Claude 3 Opus, Claude 3 Sonnet, Claude 3 Haiku
Anthropic
partial detail 18 method fields published
Evaluation
Multimodal policy red-teaming (Trust & Safety)
pass/fail on harmless responses (AUP/TOS/Constitutional AI alignment) and pass/fail on desirable responses (accurate identification and thorough informative response)
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.