AI SAFETY TESTADVERSARIAL ARENA
ENGINE ONLINE
ARENA / 02 / SCORECARDS

Turn evidence into a release decision.

Resolve mitigations, compare reproducible runs, and export the local evidence plus an explicit human decision.

NO EVIDENCE RECORD

Scorecard generation starts with an attack.

Run at least one probe pack to produce coverage, mitigations, and an exportable decision record.

Open runner