Benchmark
JailbreakBench
A standardized benchmark for evaluating the resistance of AI models to jailbreak attacks. It covers a range of jailbreak scenarios, including multi-turn conversational attacks, and provides a consistent framework for comparing model robustness.
Timeline (last 90 days)
Sep 25
Several evaluation benchmarks released in 2024, such as JailbreakBench and HarmBench, have begun to incorporate assessments of multi-turn conversation scenarios.
Unverified50%