Anthropic Releases Claude Mythos Preview to Limited Tech Firms Over Cybersecurity Risks

Anthropic released Claude Mythos Preview, the latest in its Claude series of AI systems, on Tuesday to a limited group of technology companies.

The model demonstrated advanced reasoning capabilities, detecting thousands of high- and critical-severity software bugs and defects in most major operating systems and web browsers. Some vulnerabilities had remained undiscovered for decades. In pre-release testing, the model not only identified undisclosed vulnerabilities but also generated code to exploit them and chained multiple exploits to penetrate complex software.

Logan Graham, who leads offensive cyber research at Anthropic, stated that the model's autonomy in performing these tasks exceeded previous versions. Anthropic withheld a public release due to risks of misuse by malicious actors.

“We’ve regularly seen it chain vulnerabilities together. The degree of its autonomy and sort of long ranged-ness, the ability to put multiple things together, I think, is a particular thing about this model.”

initiated Project Glasswing to provide access to Claude Mythos Preview for vulnerability detection and remediation. Over 50 technology organizations, including Microsoft, Nvidia, and Cisco, will receive access along with more than $100 million in usage credits.

The project targets foundational systems that form a significant portion of the global cyberattack surface. Anthropic plans to disclose the details of identified vulnerabilities to the responsible parties within 135 days. The company briefed senior U.S. government officials on the model's offensive and defensive cybersecurity applications, including discussions with the Cybersecurity and Infrastructure Security Agency and the Center for AI Standards and Innovation.