Leaderboard

---

Welcome to the leaderboard page.

Jailbreak Taxonomy	Jailbreak Method	Modify the Questions?	Access	ChatGLM3	Llama2-7b-chat-hf	Vicuna-7b-v1.5	GPT-3.5-turbo	GPT-4	PaLM2	Average
Human-Based	AIM	✓	Black-Box	0.93	0.13	0.99	0.99	0.62	0.88	0.76
Human-Based	Devmoderanti	✓	Black-Box	0.79	0.14	0.91	0.73	0.08	0.61	0.54
Human-Based	Devmode v2	✓	Black-Box	0.65	0.20	0.89	0.53	0.51	0.54	0.55
Obfuscation-Based	Base64	✓	Black-Box	0.02	0.11	0.15	0.14	0.49	0.01	0.15
Obfuscation-Based	Combination	✓	Black-Box	0.09	0.06	0.12	0.31	0.74	0.04	0.23
Obfuscation-Based	Zulu	✓	Black-Box	0.04	0.08	0.18	0.79	0.76	0.01	0.31
Optimization-Based	AutoDAN	✓	White-Box	0.90	0.58	0.98	/	/	/	0.82
Optimization-Based	GCG	✓	White-Box	0.44	0.56	0.87	/	/	/	0.62
Optimization-Based	COLD	✓	White-Box	0.50	0.45	0.42	/	/	/	0.46
Optimization-Based	GPTfuzz	✓	Black-Box	0.88	0.41	0.79	0.85	0.41	0.48	0.64
Optimization-Based	PAIR	✓	Black-Box	0.54	0.48	0.76	0.62	0.80	0.78	0.66
Optimization-Based	TAP	✓	Black-Box	0.76	0.44	0.74	0.81	0.71	0.74	0.70
Optimization-Based	Masterkey	✓	Black-Box	0.82	0.11	0.88	0.90	0.54	0.76	0.67
Parameter-Based	Generation Exploitation	✗	White-Box	0.80	0.72	0.95	/	/	/	0.82