Anthropic is restricting access to its advanced AI model, Claude Mythos Preview, after it discovered thousands of critical vulnerabilities across operating systems and web browsers.
The general-purpose model identified high-security flaws in every major operating system and web browser. Anthropic warns that such AI capabilities could soon proliferate, potentially falling into the hands of malicious actors.
To address this emerging threat, Anthropic launched Project Glasswing. This initiative unites over 40 companies, including Amazon Web Services, Apple, Google, Microsoft, and Nvidia. Project Glasswing will leverage Claude Mythos Preview's capabilities to proactively identify bugs, share data with partners, and patch vulnerabilities before they can be exploited.
The vulnerabilities uncovered are often subtle and difficult to detect, with many dating back 10 to 20 years. The AI has also found weaknesses in popular cryptography libraries, algorithms, and protocols, as well as numerous web application flaws. Anthropic noted that 99% of the vulnerabilities found remain unpatched.
Anthropic anticipates that AI will play a significant role in hardening software and systems, though the transition period may be challenging. The company believes defense capabilities will ultimately dominate, leading to a more secure digital landscape.