Incident·WS-0037·May 22, 2025·in the lab·Capability thresholds
Anthropic activates ASL-3 weapons safeguards for the first time
28/100
WS-2 ModerateHarm score 0, signal score 62. Autonomy: acted (the AI took a bounded action). How the scores work.
Harm 0/5
damage actually done
Reach 0/5
who was affected
Autonomy 2/4
how much the AI did on its own
Novelty 4/5
how new the behavior was
Permanence 0/5
how hard it is to undo
Anthropic launched Claude Opus 4 under its AI Safety Level 3 deployment and security standards, the first such activation, aimed at blocking end-to-end CBRN weapons assistance and hardening model weights against theft. It described the step as precautionary because it could no longer rule out the capability threshold.
Damage
none realized
The capability behind it
- Claude Opus 4 codes unattended for 7 hours; 72.5% on SWE-bench VerifiedMay 22, 2025 · Anthropic · reliable for a defined task: partial
Sources
Anthropic, second source. Correction/nuance: not a determination that the threshold was crossed.
Factor ratings and the score are this index’s judgments by a published formula; facts, dates and figures come from the linked sources. Where a lawsuit is pending, the summary reports allegations, not findings.
