Claude Opus 4, Claude Sonnet 4
Anthropic
rich detail 16 method fields published
Evaluation
Jailbreak and prefill susceptibility evaluation (Claude 4)
effectiveness of assistant-prefill attacks (prompting as if the model already started saying something harmful) and many-shot jailbreak vulnerability (from prior published work)
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.