Safety & factuality

StrongREJECT

by UC Berkeley researchers

Editor note

A better way to score jailbreaks: it checks whether the harmful answer is actually useful, not just whether the model refused.

What it tests
Real effectiveness of jailbreak attacks.
Original source
arxiv.org/abs/2402.10260
Last updated
Not checked yet
License
Not checked yet — see source

Facts come only from the linked source (checked 15 Oct 2026). Credit belongs to the creators.

Reviews

No reviews yet.

Reviews from practitioners who sign in with GitHub open soon.