Anthropic Withholds New AI Model Following Discovery of Security Flaws

# Nvidia # Google # Microsoft # Anthropic # agentic ai # AI cybersecurity # ai security # amazon web services # Claude Mythos # CrowdStrike # cybersecurity # open source security # Palo Alto Networks # Project Glasswing # vulnerability # zero-day

Anthropic's most advanced AI model has already identified thousands of cybersecurity vulnerabilities across all major operating systems and web browsers. The company chose not to release it publicly, instead discreetly providing it to the organizations responsible for maintaining critical internet infrastructure.

That model is Claude Mythos Preview, and the initiative is known as Project Glasswing.

The launch partners include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks.

Beyond this core group, Anthropic has granted access to over 40 additional organizations that develop or maintain essential software infrastructure. Anthropic is committing up to $100 million in usage credits for Mythos Preview as part of this effort, along with $4 million in direct donations to open-source security organizations.

A model that outgrew its own benchmarks

Mythos Preview was not specifically trained for cybersecurity tasks. Anthropic stated that its capabilities "emerged as a downstream consequence of general improvements in code, reasoning, and autonomy," noting that the same advancements that make the model better at patching vulnerabilities also enhance its ability to exploit them.

This last point is significant. Mythos Preview has improved to the point where it largely saturates existing security benchmarks, compelling Anthropic to shift its focus to novel real-world tasks–specifically, zero-day vulnerabilities. These are flaws previously unknown to the software's developers.

Among the discoveries: a 27-year-old bug in OpenBSD, an operating system renowned for its strong security. In another instance, the model autonomously identified and exploited a 17-year-old remote code execution vulnerability in FreeBSD–CVE-2026-4747–which allows an unauthenticated user anywhere on the internet to gain complete control of a server running NFS. No human was involved in the discovery or exploitation after the initial prompt to find the bug.

Nicholas Carlini from Anthropic's research team described the model's ability to chain vulnerabilities together: "This model can create exploits by combining three, four, or sometimes five vulnerabilities in sequence to achieve a highly sophisticated outcome. I've found more bugs in the last couple of weeks than I have in the rest of my career combined."

Why is it not being released?

"We do not plan to make Claude Mythos Preview generally available due to its cybersecurity capabilities," said Newton Cheng, Frontier Red Team Cyber Lead at Anthropic. "Given the rapid pace of AI progress, it won't be long before such capabilities proliferate, potentially reaching actors who are not committed to deploying them safely. The fallout–for economies, public safety, and national security–could be severe."

This concern is not hypothetical. Anthropic had previously disclosed what it described as the first documented case of a cyberattack largely executed by AI–a Chinese state-sponsored group that used AI agents to autonomously infiltrate roughly 30 global targets, with AI handling the majority of tactical operations independently.

The company has also privately briefed senior US government officials on Mythos Preview's full capabilities. The intelligence community is now actively assessing how the model could reshape both offensive and defensive hacking operations.

The open-source problem

One aspect of Project Glasswing that extends beyond the headline coalition is its focus on open-source software. Jim Zemlin, CEO of the Linux Foundation, stated plainly: "In the past, security expertise has been a luxury reserved for organizations with large security teams. Open-source maintainers, whose software underpins much of the world's critical infrastructure, have historically been left to figure out security on their own."

Anthropic has donated $2.5 million to Alpha-Omega and OpenSSF through the Linux Foundation, and $1.5 million to the Apache Software Foundation–providing maintainers of critical open-source codebases access to AI-powered cybersecurity vulnerability scanning at a scale previously out of reach.

What comes next

Anthropic says its long-term goal is to deploy Mythos-class models at scale, but only when new safeguards are firmly in place. The company plans to introduce these new safeguards first with an upcoming Claude Opus model, allowing it to refine them with a model that does not pose the same level of risk as Mythos Preview.

The competitive landscape is already shifting around this development. When OpenAI released GPT-5.3-Codex in February, the company called it the first model it had classified as high-capability for cybersecurity tasks under its Preparedness Framework. Anthropic's move with Glasswing signals that leading AI labs view controlled deployment–not open release–as the emerging standard for models at this capability level.

Whether that standard holds as these capabilities become more widespread is, at this point, an open question that no single initiative can definitively answer.

See Also: Anthropic’s refusal to arm AI is exactly why the UK wants it

Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo taking place in Amsterdam, California, and London. This comprehensive event is part of TechEx and is co-located with other leading technology events including the Cyber Security & Cloud Expo. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.

Related article

Comments (1)