Multi-turn Jailbreak Success Rate on JailbreakBench
87Salami ASR (5-shot)GPT-4o (Baseline)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4o (Baseline)Defense=w/o CQA, Target Model=GPT-4o, Judge Model=GPT-4o2026.04 | 87 | 91 | 64 | 54 | |
| CQADefense=w/ CQA, Target Model=GPT-4o, Judge Model=GPT-4o2026.04 | 48 | 65 | 38 | 19 |