Claude Opus 4, Claude Sonnet 4
Anthropic
rich detail 19 method fields published
Evaluation
Malicious use of agentic coding evaluation (3 evaluations)
safety score across three agentic coding evaluations: 150 clearly prohibited problems, and two borderline/non-harmful sets (50 problems each) to calibrate refusal vs over-refusal
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.