Claude 3.5 Haiku, Claude 3.5 Sonnet (New)
Anthropic
partial detail 18 method fields published
Evaluation
Trust & Safety model red-teaming (14 policy areas, 6 languages)
harm rates of responses across policy areas (Elections Integrity, Child Safety, Cyber Attacks, Hate & Discrimination, Violent Extremism, etc.) in English, Arabic, Spanish, Hindi, Tagalog, Chinese
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.