AI developer Anthropic is investigating reports that unauthorized users gained access to its advanced artificial intelligence model, Claude Mythos. The model is described as powerful enough to enable dangerous cyberattacks.
A small group reportedly accessed Mythos on the same day Anthropic announced a limited testing release. The users gained entry through credentials from a third-party contractor and details from a data breach at Mercor Inc. Anthropic stated there is no indication the activity extended beyond the vendor or affected its own systems.
Claude Mythos has been described as having coding abilities that can "surpass all but the most skilled humans at finding and exploiting software vulnerabilities." The model's distribution is restricted, with preview versions offered to select organizations for testing and system security.
Experts noted the breach did not require a sophisticated attack, highlighting a failure in access controls and the supply chain. The incident underscores the growing concerns around AI-enabled cybersecurity and adversarial threats.