Anthropic's Leaked Claude Mythos AI Sparks Cybersecurity Concerns

Anthropic is developing a new AI model called Claude Mythos-its most capable to date-after draft materials about the system were leaked from an unsecured data cache.

The company confirmed Mythos represents a “step change” in reasoning, coding, and cybersecurity performance, surpassing its current top-tier Opus models. Internal documents describe a future version, dubbed “Capybara,” positioned even higher.

Anthropic attributed the leak to human error in its content management system configuration and has since restricted access. The draft warned that Mythos’s offensive cyber capabilities could outpace defenders’ ability to respond.

As a result, Anthropic plans a cautious rollout, initially limiting access to cybersecurity defense organizations.

Financial markets reacted swiftly: Palo Alto Networks (PANW) fell 7%, CrowdStrike (CRWD) dropped 6.4%, Zscaler (ZS) declined 5.8%, and Fortinet (FTNT) slipped 4%.