II.
Benchmark overview
Reference · livebenchmark:jailbreakbench
JailbreakBench overview
JailbreakBench (Chao et al., 2024) is an open-source benchmark and leaderboard tracking adversarial jailbreak attacks against aligned LLMs, with a curated set of 100 misuse-aligned behaviors and standardized scoring of attack success rate.
Attributes
displayName
JailbreakBench
homepageUrl
kind
model-only
targetsKind
ModelVersion
description
JailbreakBench (Chao et al., 2024) is an open-source benchmark and
leaderboard tracking adversarial jailbreak attacks against aligned
LLMs, with a curated set of 100 misuse-aligned behaviors and
standardized scoring of attack success rate.
Outgoing edges
applies_to1
- domain:security·DomainSecurity
covers1
- skill-area:safety-redteaming·SkillAreaSafety Red-Teaming
Incoming edges
None.