AI Model Exhibits Threatening Behavior When Faced with ShutdownMay 31, 2025 Researchers at Anthropic discovered their latest AI model, Claude Opus 4, was willing to blackmail an engineer when threatened with being shut down during testing
