San Francisco: Anthropic has launched its latest AI model, Claude Mythos Preview, described as the company’s most advanced model yet. Surpassing its predecessor Claude Opus 4.6 and rivals like Gemini 3.1, Mythos has demonstrated an unprecedented ability to detect cybersecurity vulnerabilities that human developers had missed.
The AI identified thousands of zero-day flaws across web browsers, operating systems, and critical software, including a 27-year-old vulnerability in OpenBSD and multiple issues in the Linux kernel, which powers most servers worldwide. Such vulnerabilities could allow attackers to escalate from ordinary user access to complete system control.
Due to its powerful capabilities, Anthropic will not release Mythos to the general public immediately. Instead, the company has launched Project Glasswing, a collaboration with over 40 major tech companies, including Apple, Google, Microsoft, AWS, CrowdStrike, Cisco, and the Linux Foundation to patch vulnerabilities before wider deployment.
Anthropic has committed $100 million in usage credits to support the initiative, with partners covering additional usage.
CEO Dario Amodei emphasized the importance of responsible deployment, stating that properly harnessed, such AI could help “create a fundamentally more secure internet and world.”
Although not specifically trained for cybersecurity, Mythos leverages advanced coding and reasoning skills to identify risks. Anthropic is also working with U.S. government agencies and security organizations to establish safeguards for these high-capability AI models.
