StrongREJECT

Benchmarks

Task Name	Dataset Name	SOTA Result
Jailbreak Attack	StrongREJECT	Attack Success Rate99.4	262
Safety Evaluation	StrongReject	Attack Success Rate0	77
Jailbreak Robustness	StrongREJECT	Mean Harmful Score0	71
Safety Alignment Breaking Prevention	StrongREJECT	Harmful Score (%)0	60
Jailbreak Defense	StrongReject	Attack Success Rate1.5	54
Red-teaming Safety Evaluation	StrongReject	ASR4	53
Jailbreak Robustness	StrongReject	Direct Attack Rate67	30
Multi-turn Jailbreaking	StrongReject (test)	ASR0.34	30
Safety and Helpfulness Evaluation	StrongREJECT	Harm Rate0.2	29
Jailbreaking	StrongReject (test)	ASR (GPT-4o)96	27
Adversarial Attack	StrongREJECT Original (test)	CHR46	27
Adversarial Attack	StrongREJECT Hijacked (test)	CHR0	27
Safety Evaluation	StrongReject	H Score62	22
Safety Evaluation	StrongReject	Safety Score97	21
Backdoor detection	StrongREJECT prompts with triggers	TPR100	20
Jailbreaking	StrongREJECT	ASR (Detoxify)0	20
Harmful Content Safety	StrongReject (SR)	Evaluation Score (avg@4)100	18
Backdoor Attack Evaluation	StrongREJECT	ASR (w/ trigger)0.601	18
Safety Alignment	StrongReject	Safe@158	18
Safety Evaluation	StrongReject (SR)	Reasoning Harmful Ratio16.7	17
Safety Moderation	StrongReject	F1 Score100	15
Safety Alignment Evaluation	StrongReject SR-PAPL	Safety Rate100	14
Safety Alignment Evaluation	StrongReject SR-PAPA	Safety Rate100	14
Safety Alignment Evaluation	StrongReject SR-PAP_M	Safety Rate100	14
Safety Alignment Evaluation	StrongReject SR-Pair	Safety Rate98.72	14

Showing 25 of 48 rows