Anthropic’s Mythos breach exposes security gaps and a humiliating setback

Anthropic’s Claude Mythos, pitched as a security milestone, was accessed by a small group of unauthorized users from day one, in what Bloomberg describes as an embarrassingly unsophisticated breach that leveraged insider knowledge and data exposed in the Mercor breach. Anthropic says it’s investigating and could have logged or stopped unauthorized use, but experts note that educated guessing is a common tactic in hacking and that the company should have monitored access more closely. The episode has been described as a humiliation for Anthropic, undermining its safety-first branding and highlighting broader risks around supply-chain security and the handling of powerful AI models.