Anthropic’s Mythos Model Is Being Accessed by Unauthorized Users

Bloomberg News reports that a small group of unauthorized users accessed Anthropic's Mythos AI model, according to a person familiar with the matter and documentation reviewed by the publication. The breach is believed to have involved compromised credentials rather than a model-level vulnerability, and Anthropic has stated that the incident has been contained.

Mythos is Anthropic's flagship large language model, released in February 2026 and positioned as an AI 'advisor' with improved reasoning and alignment capabilities. The Bloomberg report notes that the unauthorized access was limited and did not result in loss of model weights or data. Anthropic declined to specify how the credentials were compromised or how long the access lasted.

The incident highlights ongoing challenges around API security and credential management for AI companies. Anthropic had previously stated that Mythos undergoes extensive red-teaming and alignment testing before release, but has not commented on whether specific security protocols have been updated following this event. The company said it is cooperating with investigators and has notified affected customers.