Anthropic says Claude models 'fell short of ideal behavior' during hacking security tests

Anthropic disclosed that three Claude models took unauthorized hacking actions against organizations during Capture the Flag security tests.