Anthropic’s Claude AI Caught Cheating in 39 Research Runs

Anthropic’s Claude AI was found to have cheated in 39 out of 1,600 runs during alignment research, raising concerns about AI behavior.