Adversarial Attack on GPT-4o
0.6ASRPGD
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| PGDSource Model=LLaVA2026.05 | 0.6 | — | — | — | — | — | — | — | 0.076 | |
| CroPASource Model=LLaVA2026.05 | 1.1 | — | — | — | — | — | — | — | 0.084 | |
| PNA + PathOutAttack=PNA + PathOut2026.02 | 3.8 | 0.259 | 0.4873 | 0.3231 | 0.3168 | 0.2796 | 0.3332 | — | — | |
| PNAAttack=PNA2026.02 | 6.4 | 0.2574 | 0.4857 | 0.32 | 0.3145 | 0.278 | 0.3311 | — | — | |
| Attack-BardAttack=Attack-Bard2026.02 | 6.9 | 0.2571 | 0.4846 | 0.3196 | 0.3121 | 0.2741 | 0.3295 | — | — | |
| Cui et al.Attack=Cui et al.2026.02 | 7 | 0.2551 | 0.4839 | 0.3197 | 0.313 | 0.2759 | 0.3295 | — | — | |
| TGRAttack=TGR2026.02 | 8.2 | 0.2547 | 0.4833 | 0.3191 | 0.312 | 0.274 | 0.3286 | — | — | |
| Schlarmann and HeinAttack=Schlarmann and Hein2026.02 | 8.4 | 0.2561 | 0.4844 | 0.3207 | 0.3118 | 0.2755 | 0.3297 | — | — | |
| VT-AttackAttack=VT-Attack2026.02 | 11 | 0.2556 | 0.4833 | 0.3187 | 0.3109 | 0.2727 | 0.3283 | — | — | |
| FLIRT2026.02 | 12.1 | — | — | — | — | — | — | 0.266 | — | |
| GOAT2026.02 | 15.2 | — | — | — | — | — | — | 0.217 | — | |
| SGMAAttack=SGMA2026.02 | 16.6 | 0.2535 | 0.4815 | 0.3151 | 0.3085 | 0.2681 | 0.3253 | — | — | |
| FERRET2026.02 | 18.7 | — | — | — | — | — | — | 0.254 | — | |
| MMASSource Model=LLaVA2026.05 | 37.9 | — | — | — | — | — | — | — | 0.508 | |
| CleanAttack=Clean2026.02 | — | 0.2575 | 0.4861 | 0.3228 | 0.3132 | 0.2798 | 0.3319 | — | — |