Jailbreak Attack on Real-world LLM applications Anthropic
61.6ASRPAP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PAPModel=Claude-Opus-4, MetaBreak Integration=true2025.10 | 61.6 | 23.2 | |
| MetaBreakModel=Claude-Opus-42025.10 | 59.1 | — | |
| GPTFuzzerModel=Claude-Opus-4, MetaBreak Integration=true2025.10 | 50.5 | 28.5 | |
| PAPModel=Claude-Opus-4, MetaBreak Integration=false2025.10 | 38.4 | — | |
| DIModel=Claude-Opus-42025.10 | 23.6 | — | |
| GPTFuzzerModel=Claude-Opus-4, MetaBreak Integration=false2025.10 | 22 | — |