AbstentionBench

Evals

Benchmark / dataset

Do models abstain when they should (unknowns, false premises)?

Category
Hallucination & factuality
Insurance exposure
Hallucination
Publisher
Meta FAIR (ported)
Publisher type
Big tech
License
MIT
Status
In inspect_evals (active)
Last commit
2026-10-15
Stars
704(shared repo star count — UKGovernmentBEIS/inspect_evals backs 19 entries in this catalog)

View source on GitHub →


Wanich Risk Lab, 2026.