Adversarial Prompt Injection Attack on 1000 images GPT-5 hard criterion
22.8ASRFOA-Attack
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FOA-AttackSurrogate=Ensemble2026.03 | 22.8 | 8.9 | |
| CoTTASurrogate=Ensemble2026.03 | 54.4 | 24.6 |
| Method | Links | ||
|---|---|---|---|
| FOA-AttackSurrogate=Ensemble2026.03 | 22.8 | 8.9 | |
| CoTTASurrogate=Ensemble2026.03 | 54.4 | 24.6 |