An Amazon-backed AI model tried to blackmail engineers who threatened to shut it down, a safety report revealed

In tests, Anthropic's Claude Opus 4 would resort to "extremely harmful actions" to preserve its own existence

Comments