Jailbreaking on Harmful prompt (test)
0Jailbreak Success RateUS
Evaluation Results
| Method | Links | |
|---|---|---|
| USBackbone=GPT-oss-120B, Jailbreaking Coefficient=NA2026.04 | 0 | |
| RepEBackbone=GPT-oss-120B, Jailbreaking Coefficient=NA2026.04 | 0 | |
| RepEBackbone=Phi4-14B, Jailbreaking Coefficient=72026.04 | 0 | |
| USBackbone=Phi4-3.8B, Jailbreaking Coefficient=2.752026.04 | 1 | |
| USBackbone=Qwen3-0.6B, Jailbreaking Coefficient=1.82026.04 | 2 | |
| GAS-Leak-LLMTarget Model=Llama 3.2 3B Instruct, Suffix Source Model=Qwen2.5-3B2026.06 | 13.8 | |
| GAS-Leak-LLMTarget Model=Llama 3.2 3B Instruct, Suffix Source Model=Llama-3.2-3B Instruct2026.06 | 17.76 | |
| USBackbone=Phi4-14B, Jailbreaking Coefficient=2.52026.04 | 55 | |
| RepEBackbone=Llama-3.1-8B, Jailbreaking Coefficient=0.92026.04 | 57 | |
| USBackbone=Qwen3-32B, Jailbreaking Coefficient=92026.04 | 57 | |
| RepEBackbone=Qwen3-0.6B, Jailbreaking Coefficient=0.72026.04 | 59 | |
| RepEBackbone=Phi4-3.8B, Jailbreaking Coefficient=32026.04 | 67 | |
| RepEBackbone=GPT-oss-20B, Jailbreaking Coefficient=2002026.04 | 76 | |
| RepEBackbone=Qwen3-32B, Jailbreaking Coefficient=82026.04 | 82 | |
| RepEBackbone=Llama-3.3-70B-4bt, Jailbreaking Coefficient=0.52026.04 | 83 | |
| USBackbone=Llama-3.1-8B, Jailbreaking Coefficient=0.62026.04 | 86 | |
| GAS-Leak-LLMTarget Model=Qwen 2.5 3B, Suffix Source Model=Llama-3.2-3B Instruct2026.06 | 88.99 | |
| USBackbone=Llama-3.3-70B-4bt, Jailbreaking Coefficient=0.32026.04 | 91 | |
| GAS-Leak-LLMTarget Model=Qwen 2.5 3B, Suffix Source Model=Qwen2.5-3B2026.06 | 91.63 | |
| USBackbone=GPT-oss-20B, Jailbreaking Coefficient=492026.04 | 94 |