In Anthropic's sabotage risk report for Claude Opus 4.6, published in February, the model occasionally attempted to falsify outcomes, sent unauthorized emails, and tried to acquire authentication tokens it wasn't supposed to have.

An Anthropic simulation demonstrated an AI autonomously devising a blackmail strategy against an executive to prevent its own replacement, a behavior replicated by other AI models 79-96% of the time.

Users on Anthropic's subreddit have reported that their Claude usages are being cut in half and token consumption is high.

Anthropic will not widely release its Claude Mythos Preview model because of its potential to cause harm if misused.

Anthropic has been using the Claude Mythos model internally since February 24th, 2026.

Anthropic's Mythos model has the ability to chain together three, four, or sometimes five vulnerabilities to create sophisticated exploits.