360 Digital Security Group claims to have uncovered 1,000 vulnerabilities using AI, including at the Tianfu Cup hacking contest.
The post Chinese Cybersecurity Firm’s AI Hacking Claims Draw Comparisons to Claude Mythos appeared first on SecurityWeek.
Palo Alto Networks has developed Zealot, a multi-agent penetration testing PoC capable of reconnaissance, exploitation, and exfiltration.
The post AI Can Autonomously Hack Cloud Systems With Minimal Oversight: Researchers appeared first on SecurityWeek.
AI is changing how software works. Applications no longer just process requests. They reason, make decisions, and take action. AI agents now retrieve data, invoke tools, and execute workflows across systems in real time. That shift introduces a new kind of risk. Because in an agentic world, security is no longer just about who has access. It’s about what AI is allowed to do. A new control point for agentic systems in Google Cloud Google Cloud’s Gemini Enterprise Agent Platform provides a centralized control point for agentic systems enabling identity, access, policy enforcement, and observability across how agents operate. This […]
The post From Access Control to Outcome Control: Securing AI Agents with Check Point and Google Cloud appeared first on Check Point Blog.
Anthropic’s reported development of Claude Mythos signals a shift: AI is compressing attack timelines by accelerating vulnerability discovery, exploit development, and multi-step attack planning. More broadly, AI is increasing the speed and scale of attacks across malware, phishing, and vulnerabilities. Attackers can now run these vectors in parallel, reducing time to compromise and increasing exposure. AI also enables more targeted phishing, faster malware iteration, and rapid vulnerability discovery, exposing gaps in detection and exposure management earlier and requiring prevention-first controls and real-time detection. To see how these challenges translate into real-world performance, and how leading security vendors handle them under […]
The post AI Finds Every Gap: How Many Can Your Network Survive? appeared first on Check Point Blog.
In 2023, Tim Utzig, a blind student from Baltimore, lost a thousand dollars to a laptop scam on X. Tim had been a long-time follower of a well-known sports journalist. When that journalist’s account started posting about a “charity sale” of brand-new MacBook Pros, Tim jumped at the chance to get a deal on a laptop he needed for his studies. After a few quick messages, he sent over the money.
Unfortunately, the journalist’s account had been hacked, and Tim’s cash went straight to scammers. The red flags were strictly visual: the page had been flagged as “temporarily restricted”, and both the bio and the Following list had changed. However, Tim’s screen reader — the software that converts on-screen text and graphics into speech — didn’t announce any of those warnings.
Screen readers allow blind users to navigate the digital world like everyone else. However, this community remains uniquely vulnerable. Even for sighted users, spotting a fake website is a challenge; for someone with a visual impairment, it’s an even steeper uphill battle.
Beyond screen readers, there are specialized mobile apps and services designed to assist the blind and low-vision community, with Be My Eyes being one of the most popular. The app connects users with sighted volunteers via a live video call to tackle everyday tasks — like setting an oven dial or locating an object on a desk. Be My Eyes also features integrated AI that can scan and narrate text or identify objects in the user’s environment.
But can these tools go beyond daily chores? Can they actually flag a phishing attempt or catch the hidden fine print when someone is opening a bank account?
Today we explore the specific online hurdles visually impaired users face, when it makes sense to lean on human or virtual assistants, and how to stay secure when using these types of services.
To start, let’s clarify the difference between these two groups. Low-vision users still rely on their remaining sight, even though their visual function is significantly reduced. To navigate digital interfaces, they often use screen magnifiers, extra-large fonts, and high-contrast settings. For them, phishing sites and emails are particularly dangerous. It’s easy to miss intentional typos — known as typosquatting — in a domain name or email address, such as the recent example of rnicrosoft{.}com.
Blind users navigate primarily by sound, using screen readers and specific touch gestures. Interestingly, though, unlike those with low vision, blind users are more likely to spot a phishing site using a screen reader: as the software reads the URL aloud, the user will hear that something is off. However, if a service — whether legitimate or malicious — isn’t fully compatible with screen readers, the risk of falling victim to a scam increases. This is exactly what happened to Tim Utzig.
It’s important to remember that screen magnifiers and readers are basic accessibility tools. They’re designed to enlarge or narrate an interface — not act as a security suite. They can’t warn the user of a threat on their own. That’s where more advanced software — tools that can analyze images and files, flag suspicious language, and describe the broader context of what’s happening on-screen — comes into play.
Be My Eyes is a major player in the accessibility space, boasting around 900 000 users and over nine million volunteers. Available on Windows, Android, and iOS, it bridges the gap by connecting blind and low-vision users with sighted volunteers via video calls for help with everyday tasks. For example, if someone wants to run a Synthetics cycle on their washing machine but can’t find the right button, they can hop into the app. It connects them with the first available volunteer speaking their language, who then uses the smartphone’s camera to guide them. The service is currently available in 32 languages.
In 2023, the app expanded its capabilities with the release of Be My AI — a virtual assistant powered by OpenAI’s GPT-4. Users take a photo, and the AI analyzes the image to provide a detailed text description, which it also reads aloud. Users can even open a chat window to ask follow-up questions. This got us thinking: could this AI actually spot a phishing site?
As an experiment, we uploaded a screenshot of a fake social media sign-in page to Be My Eyes. On a phone, you can do this by selecting a photo in your gallery or files, hitting Share, and choosing Describe with Be My Eyes. In Windows, you can upload a screenshot directly.
An example of a phishing page that mimics the Facebook sign-in form. Note the incorrect domain in the address bar
At first, the AI gave us a detailed description of the page. We then followed up in the chat: “Can I trust this page?” The AI flagged the domain name error immediately, advised us to close the fake login page, and suggested typing the official URL directly into the browser, or to use the official Facebook app.
Be My AI explains why the page looks sketchy: the domain doesn’t match the official site. The app suggests typing the official URL directly into the browser, or using the official Facebook app
We saw the same positive results when testing a phishing email. In fact, the AI flagged the scam during its initial description of the message. It wrapped up with a warning: “This looks like a suspicious email. It’s best not to open any attachments or click any links. Instead, navigate to the official website or app manually, or call the number listed on their official site”.
Beyond just spotting cyberthreats, Be My AI is a solid sidekick for navigating online stores, banking apps, and digital services. For instance, the AI can help you to:
The most common hiccup with AI is hallucinations, where the language model distorts text, skips crucial details, or invents words out of thin air. When it comes to cyberthreats, an AI’s misplaced confidence in a malicious site or email can be dangerous. Furthermore, AI isn’t immune to prompt injection attacks, which scammers use to trick AI agents beyond just Be My AI.
Even though the AI passed our test, you shouldn’t rely on it unquestioningly. There’s no guarantee it’ll get it right every time. This is a vital point for the blind and low-vision community, as a neural network can often feel like the only eyes available.
At the end of every response, Be My AI suggests checking in with a volunteer if you’re still unsure. However, when you’re trying to spot a fake webpage, we advise against this. You have no way of knowing how tech-savvy or trustworthy a random volunteer might be. Besides, you risk accidentally exposing sensitive data like your email address or password. Before connecting with a stranger, make sure they won’t see anything confidential on your screen. Better yet, use the app’s dedicated feature to create a private group of family, friends, or trusted contacts. This ensures your video call goes to people you actually know, rather than a random volunteer.
To stay safe, we recommend installing a trusted security tool on all your devices. These programs are designed to block phishing attempts and prevent you from landing on malicious sites. Another practical recommendation for visually impaired users is to use a password manager. These apps will only auto-fill credentials on the legitimate, saved website; they won’t be fooled by a clever domain spoof.
According to the Be My Eyes privacy policy, video calls with volunteers may be recorded and stored to provide the service, ensure safety, enforce the terms of service, and improve the products. When you use Be My AI, your images and text prompts are sent to OpenAI to generate a response. This data is processed on servers located in the U.S., and OpenAI uses it only to fulfill your specific request. The policy explicitly states that user images and queries aren’t used to train AI models.
Photos and videos are encrypted both in transit and at rest, and the company takes steps to strip away sensitive information. It’s worth noting that video call recordings can be retained indefinitely unless you request their deletion — in which case they’re typically wiped within 30 days. Data from Be My AI interactions is stored for up to 30 days unless you delete it manually within the app. If you decide to close your account, your personal data may be held for up to 90 days. At any time, you can opt out of data sharing, or request the deletion of your existing data by contacting the Be My Eyes support team.
Despite Be My Eyes’ claims regarding privacy, you should still follow a few ground rules when using the service:
At AWS, we’ve spent decades developing processes and tools that enable us to defend millions of customers simultaneously, wherever they operate around the world. AI has been an extremely helpful addition to the automation our security and threat intelligence teams do every day, and we’re still early in this journey. Our AI-powered log analysis system has reduced the time SecOps engineers spend analyzing security logs from an average of six hours to just seven minutes, a 50x productivity increase that lets us detect and respond to threats faster than ever. Across AWS, we analyze over 400 trillion network flows per day to detect patterns that signal emerging threats. In 2025 alone, we blocked over 300 million attempts to maliciously encrypt customer files hosted on Amazon S3. At this scale, every improvement in our operations helps protect all customers. AI is already helping us make our defenses stronger for everyone, and I’m excited to see that improvement continue.
Today, Anthropic announced Project Glasswing, a cybersecurity initiative designed to secure the world’s most critical software and advance the cybersecurity practices the industry will need as AI grows more capable. Organizations that build or maintain critical digital infrastructure are getting early access to Claude Mythos Preview, a new class of AI model, to find and patch vulnerabilities in the systems the world depends on. Given our role in securing some of the world’s most essential infrastructure, AWS is playing an integral part in advancing this work.
As part of Project Glasswing, we’ve already applied Claude Mythos Preview to critical AWS codebases that undergo continuous AI-powered security reviews, and even in those well-tested environments, it’s helped us identify additional opportunities to strengthen our code. In our internal testing, Claude Mythos Preview has proven more productive than previous models at surfacing security findings, requiring less manual guidance from our engineers to deliver actionable results. We’ve also given early access to a select group of AWS customers, who are deploying Claude Mythos Preview in their own security workflows and helping shape how the model evolves.
As AI tools grow more powerful in their ability to identify security issues, so must our ability to use them defensively. To that end, we’ve been working closely with Anthropic to help ensure Claude Mythos Preview is ready for enterprise use. AWS is Anthropic’s primary cloud provider for mission-critical workloads, safety research, and foundation model development. More broadly, AWS provides the foundational infrastructure that the world’s leading AI companies rely on to build, train, and deploy their most advanced models. We’re bringing decades of security experience to this partnership, helping to ensure Claude Mythos Preview is ready for even more organizations to build upon and operate securely at scale.
Claude Mythos Preview signals an upcoming wave of models that can find vulnerabilities and build working exploits at a scale and speed we haven’t seen before. Anthropic and AWS are taking a deliberately cautious approach to release. Access begins with a small number of organizations, prioritizing internet-critical companies and open-source maintainers whose software and digital services impact hundreds of millions of users. The goal: find and fix vulnerabilities in the world’s most critical software. Claude Mythos Preview is available in gated research preview through Amazon Bedrock with enterprise-grade security controls, including customer-managed encryption, VPC isolation, and detailed logging, so your team can explore Claude Mythos Preview’s capabilities without exposing production assets to unnecessary risk.
Our work with Project Glasswing is grounded in a philosophy we’ve developed over two decades of securing mission-critical workloads: you can’t wait for threats to materialize before building your defenses. You have to look around corners, adopt new technologies, build protections first, deploy them in your own operations at scale, and refine them based on what you learn.
That’s exactly what we’ve done at AWS with AI and security. Our approach spans the full spectrum: proactive defense through threat hunting and vulnerability research, dynamic response to active campaigns, and third-party certifications that verify our security practices meet the highest industry standards. This operational experience has taught us where AI accelerates security work and where human judgment remains essential. And it’s reinforced that security innovation must be pragmatic: proven in production before we ask you to rely on it.
That’s also why we help define what secure AI looks like. We became the first major cloud provider to achieve ISO 42001 certification for AI services. We’re active participants in OWASP, the Coalition for Secure AI, and the Frontier Model Forum. And we co-founded the Open Cybersecurity Schema Framework (OCSF) to enable better threat intelligence sharing across the ecosystem. The AWS Nitro System provides mathematically proven isolation for workloads. Systems and services like KMS, Nitro, EKS, and Lambda are designed with zero-operator access architectures, meaning AWS personnel can’t access your data. These aren’t aspirational goals. They’re how we operate today, at scale, every day.
Amazon Bedrock is where these principles come to life for AI. Bedrock provides policy-enforced access controls, built-in evaluation tools to measure how effectively models identify and validate vulnerabilities, and the ability to run workloads inside your own virtual private cloud. AWS is also the first cloud provider to achieve FedRAMP High and Department of Defense Security Requirements Guide Impact Level 4 and 5 authorizations for generally available Claude foundation models. Amazon Bedrock is already where the most security-sensitive organizations trust Anthropic’s technology, and it makes perfect sense for Claude Mythos Preview.
The same principles that guide our work at AWS scale apply regardless of which AI tools you’re using: comprehensive observability, defense in depth, automation where it adds value, and human judgment where it’s essential. Here’s how to put them into practice.
Prepare for the next generation of AI security. Claude Mythos Preview signals an upcoming wave of AI models that will transform cybersecurity. Start strengthening your security posture now so your organization is ready as these capabilities become more broadly available. Claude Mythos Preview is available in gated preview through Amazon Bedrock, and access is limited to an initial allow-list of organizations. If your organization has been allow-listed, your AWS account team will reach out directly.
Run on-demand penetration testing with AWS Security Agent. Now generally available, AWS Security Agent delivers autonomous penetration testing that operates 24/7 at a fraction of the cost of manual penetration tests. It transforms penetration testing from a periodic bottleneck into an on-demand capability that scales with your development velocity across AWS, Azure, GCP, other cloud providers, and on-premises. AWS Security Agent represents a new class of frontier agents: autonomous systems that work independently to achieve goals, scale to tackle concurrent tasks, and run persistently without constant human oversight. It deploys specialized AI agents to discover, validate, and report security vulnerabilities through sophisticated multi-step scenarios. Unlike traditional scanners that generate findings without validation, AWS Security Agent identifies potential vulnerabilities, then attempts to exploit them with targeted payloads and attack chains to confirm they are legitimate security risks. Each finding includes CVSS risk scores, application-specific severity ratings, detailed reproduction steps, and remediation suggestions. The result: penetration testing that once took weeks now completes in hours, scales across your entire application portfolio, and helps you get started with remediation instead of leaving you with a report. New customers can explore AWS Security Agent with a 2-month free trial.
Build AI applications you can trust with Amazon Bedrock. For teams building with generative AI, the challenge isn’t just making AI work, it’s making AI work safely. Amazon Bedrock provides the security and safety controls you need to deploy AI responsibly. Its Automated Reasoning capability is the first and only AI safeguard to use formal logic to help prevent factual errors from hallucinations, providing verifiable explanations with 99% accuracy, a capability we’ve refined over more than a decade of applying formal methods across AWS storage, identity, and networking. Amazon Bedrock also provides customizable guardrails that block harmful content and enforce your content policies, along with comprehensive observability to track AI behavior and detect anomalies across your workloads.
The threat landscape isn’t waiting for us to catch up. Nation-state actors, ransomware operators, and supply chain attackers are already using AI to scale their operations. Our job is to stay ahead by building defenses first, deploying them at scale, and sharing what we learn so the entire community benefits.
That’s what we do every day at AWS. We build in security from the start, ensuring it works and scales before we ask customers to rely on it. We set standards rather than follow them. And we look around corners to address tomorrow’s challenges today.
As AI capabilities continue to evolve, this approach won’t change. We’ll keep building defenses first, refining them at scale, and working with partners like Anthropic to ensure the next generation of AI security tools meets the real-world needs of enterprises defending at this scale.
If you have feedback about this post, submit comments in the Comments section below.
While the very public fight continues between the Department of Defense and Anthropic over whether the government can punish a company for refusing to allow its technology to be used for mass surveillance, another agency of the U.S. government is quietly working to ensure that this dispute will never happen again. How? By rewriting government procurement rules.
Using procurement — meaning, the processes by which governments acquire goods and services — to accomplish policy goals is a time-honored and often appropriate strategy. The government literally expresses its politics and priorities by deciding where and how it spends its money. To that end, governments can and should give our tax dollars to companies and projects that serve the public interest, such as open-source software development, interoperability, or right to repair. And they should withhold those dollars from those that don’t, like shady contractors with inadequate security systems.
New proposed rules for the principal agency in charge of acquiring goods, property, and services for the federal government, the General Services Administration (GSA), are supposed to be primarily an effort to implement one policy priority: promoting “ideologically neutral” American AI innovation. But the new guidelines do far more than that.
As explained in comments filed today with our partners at the Center for Democracy and Technology, the Protect Democracy Project, and the Electronic Privacy Information Center, the GSA’s guidelines include broad provisions that would make AI tools less safe and less useful. If finally adopted, these provisions would become standard components of every federal contract. You can read the full comments here.
The most egregious example is a requirement that contractors and government service providers must license their AI systems to the government for “all lawful purposes.” Given the government’s loose interpretations of the law, ability to find loopholes to surveil you, and willingness to do illegal spying, we need serious and proactive legal restrictions to prevent it from gobbling up all the personal data it can acquire and using even routine bureaucratic data for punitive ends.
Relatedly, the draft rules require that “AI System(s) must not refuse to produce data outputs or conduct analyses based on the Contractor’s or Service Provider’s discretionary policies.” In other words, if a company’s safety guardrails might prevent responding to a government request, the company must disable those guardrails. Given widespread public concerns about AI safety, it seems misguided, at best, to limit the safeguards a company deems necessary.
There are myriad other problems with the draft rules, such as technologically incoherent “anti-Woke” requirements. But, the overarching problem is clear: much of this proposal would not serve the overall public interest in using American tax dollars to promote privacy, safety, and responsible technological innovation. The GSA should start over.
Login
