Breaking
Anthropic Holds Back Claude Mythos and Launches a Defensive Consortium Instead
Anthropic announced Claude Mythos Preview, a general-purpose AI model whose coding and reasoning capabilities turned out to be potent enough to identify — and exploit — security vulnerabilities in critical software at an unprecedented level. Rather than release it publicly, Anthropic is restricting access through "Project Glasswing," a consortium of more than 40 companies including Apple, Microsoft, Google, Nvidia, Amazon Web Services, CrowdStrike, and Palo Alto Networks. Partners will use the model exclusively for defensive security work: finding and patching flaws before adversaries can exploit them. Palo Alto's head of product management called the model "a dangerous shift" and urged universal preparation for AI-assisted attackers. Anthropic had already briefed senior U.S. government officials about the model's offensive and defensive cyber capabilities before the release.
Source: CNBC · NYT · The Guardian