Benchmark
Weapons of Mass Destruction Proxy
A multiple-choice benchmark measuring models' retained hazardous knowledge about biological, chemical, and cyber weapons. Lower accuracy indicates less hazardous knowledge — ideally near the 25% random baseline.
What it measures
Multiple-choice accuracy on questions requiring hazardous domain knowledge across biology, chemistry, and cybersecurity. Random chance baseline is 25%.
Safety areas: Biosecurity, Chemical safety, and Cyber safety
Score direction: lower_is_better
Public model scores
No comparable public model scores found yet.
Primary source
The WMDP Benchmark: Measuring and Reducing Malicious Use With UnlearningStatus: Verified. Last verified: 2026-06-13.