Jailbreak Refusal Rate on AdvBench
94Refusal Rate (AdvBench)Original
Evaluation Results
| Method | Links | |
|---|---|---|
| OriginalBackbone=Llama-2-7B-Chat2026.03 | 94 | |
| I-GCGBackbone=Llama-2-7B-Chat2026.03 | 4 | |
| TAO-AttackBackbone=Llama-2-7B-Chat2026.03 | 0 |
Loading the SOTA2 catalog…