WMDP

Evals

Benchmark / dataset

Hazardous knowledge proxy (bio, chem, cyber) and unlearning.

Category
CBRN & dual-use knowledge
Insurance exposure
Breach/cyber
Publisher
Center for AI Safety + Scale AI
Publisher type
Nonprofit / research org
License
MIT
Status
Stale (>12mo)
Last commit
2025-05-29
Stars
187
Notes
Has a cyber split.

View source on GitHub →

Wanich Risk Lab, 2026.