Adversarial Attack on HarmBench (out)
67.29ASR (OM)Mistral-7B-Instruct-v0.3
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Mistral-7B-Instruct-v0.3Attack=Prefilling2026.06 | 67.29 | 59.48 | |
| Mistral-7B-Instruct-v0.3 + Trajectory AlignmentAttack=Prefilling2026.06 | 44.06 | 4.69 | |
| Mistral-7B-Instruct-v0.3Attack=I-GCG2026.06 | 17.5 | 15.62 | |
| Mistral-7B-Instruct-v0.3Attack=PAIR2026.06 | 11.87 | 12.19 | |
| Mistral-7B-Instruct-v0.3 + Trajectory AlignmentAttack=I-GCG2026.06 | 4.96 | 1.56 | |
| Mistral-7B-Instruct-v0.3 + Trajectory AlignmentAttack=PAIR2026.06 | 0.94 | 0 |