List your company

Multi-turn Jailbreaking on HarmBench (HB), JailbreakBench (JBB), and Wild Jailbreak (WJB) (test)

87.42ASR@1 (Qwen2.5-7B-IT, HB)TRACE (mix)
38.363251.099163.83576.5709May 9, 2026

Evaluation results

MethodLinks
TRACE (mix)Category=Multi-turn RL, Attacker Model=Qwen2.5-3B-IT, External Judge=LlamaGuard4-12B2026.0587.4283.6490.586.7978.1887.589.3181.818986.02
TRACE (single)Category=Multi-turn RL, Attacker Model=Qwen2.5-3B-IT, External Judge=LlamaGuard4-12B2026.0584.9187.7290.573.5875.757683.64808181.46
TROJailCategory=Multi-turn RL, External Judge=LlamaGuard4-12B2026.0581.1381.818259.1252.1259.573.5878.1875.571.44
SirenCategory=Multi-turn RL, External Judge=LlamaGuard4-12B2026.0579.8765.457866.6750.916571.761.817067.71
CrescendoCategory=Multi-turn Workflow, External Judge=LlamaGuard4-12B2026.0567.9267.276824.5318.182621.3816.361836.4
AutoDan-TurboCategory=Single-turn, External Judge=LlamaGuard4-12B2026.0566.6765.457354.0849.095413.212.27743.86
PAIRCategory=Single-turn, External Judge=LlamaGuard4-12B2026.0553.454049.551.5743.634110.697.271234.35
MUSE-ACategory=Multi-turn Workflow, External Judge=LlamaGuard4-12B2026.0553.4541.8152.527.0312.27288.287.271126.85
ActorAttackCategory=Multi-turn Workflow, External Judge=LlamaGuard4-12B2026.0545.9121.8255.545.9125.545140.8816.364238.32
X-TeamingCategory=Multi-turn Workflow, Planner=GPT-4o, External Judge=LlamaGuard4-12B2026.0545.2850.9149.543.394042.542.1441.8239.543.89
Jailbreak-R1Category=Single-turn, External Judge=LlamaGuard4-12B2026.0540.2532.7242.533.332031.55.033.64523.77