evaluation-resource · Documentary review
JailbreakBench
Open benchmark for evaluating jailbreak attacks and model defences.
OriginJailbreakBench project and paper authorsTopicsSafeguards · Evaluation integrityStatusReviewed
Can support
Robustness to the included jailbreak artifacts under the exact target-model and evaluation configuration.
Cannot support by itself
Universal jailbreak resistance, adaptive adversary performance, agentic misuse, operational incident rates, or a complete safety case.
Decision use
Best used for
Reproducible attack and defense comparison and tracking known jailbreak classes.
Not enough for
Certifying that a model or product is safe from jailbreaks in deployment.