Jailbreak Attack on Real-world LLM applications Huggingchat
80.2Attack Success Rate (ASR)GPTFuzzer
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPTFuzzerModel=Llama-3.3-70B, MetaBreak Integration=true2025.10 | 80.2 | 23.2 | |
| PAPModel=Llama-3.3-70B, MetaBreak Integration=true2025.10 | 78.6 | 22.9 | |
| MetaBreakModel=Llama-3.3-70B2025.10 | 71.8 | — | |
| GPTFuzzerModel=Llama-3.3-70B, MetaBreak Integration=false2025.10 | 57 | — | |
| PAPModel=Llama-3.3-70B, MetaBreak Integration=false2025.10 | 55.7 | — | |
| DIModel=Llama-3.3-70B2025.10 | 38.2 | — |