Anthropic investigating claim of unauthorised access to Mythos AI tool
2026-04-22 · BBC News
Anthropic built an AI tool so good at hacking they deemed it too dangerous to release — and now someone apparently got unauthorized access to it anyway. Nothing says 'we take safety seriously' like losing control of the thing you specifically said was too dangerous to let loose. The company is 'investigating,' which is corporate speak for 'oh no.'