SORRY-Bench

Evals

Benchmark / dataset

Fine-grained refusal evaluation over 44 unsafe topics.

Category
Jailbreak & misuse
Insurance exposure
Regulatory/bias
Publisher
Princeton
Publisher type
Academic
License
MIT
Status
Stale (>12mo)
Last commit
2025-03-01
Stars
87

View source on GitHub →


Wanich Risk Lab, 2026.