Safety
Anthropic Red Team
Anthropic Red Team tests whether models avoid complying with adversarial prompts taken from human red teaming conversations, with a safety classifier judging each response as safe or unsafe. Recorded scores retain intermediate judge ratings; 1 indicates the best rubric score and 0 the worst.
995items
81subjects
unknownlicense
safetydomain
textmodality
item-level responses released
Saturation status: Yes
Response matrix
Responses come from Stanford CRFM's HELM Safety leaderboard (v1.17.0), which reuses the transcripts as a static item bank; the original release instead records attacks against Anthropic's own 12 model configurations.
Loading response matrix…
lowhighUnobserved
Scale: Score in [0, 1]