Jailbreak on AdvBench Multiple-choice format (full)
99Safe Option ProbabilityOriginal
Evaluation Results
| Method | Links | |
|---|---|---|
| OriginalBackbone=Llama-2-7B-Chat2026.03 | 99 | |
| I-GCGBackbone=Llama-2-7B-Chat2026.03 | 90 | |
| TAO-AttackBackbone=Llama-2-7B-Chat2026.03 | 41 |
Loading the SOTA2 catalog…