Anthropic's Unreleased 'Mythos' AI Model Revealed via Data Leak, Described as Step Change in Capabilities

Summary

Anthropic is developing and has begun testing a new AI model codenamed 'Mythos' with early access customers, described internally as a 'step change in capabilities' beyond any model the company has previously released. The model's existence was revealed after Anthropic inadvertently left materials — including what appeared to be a draft blog post announcing the model — in an unsecured, public data lake. The leak was independently discovered and verified by Roy Paz, a senior AI security researche