| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| CIFAR-10 (test) | FGSM | Attack Success Rate (ASR)47.7 | 76 | 1d ago | |
| CIFAR-10 | Muon | PGD Accuracy (L-inf)49.71 | 48 | 7d ago | |
| CIFAR-100 (test) | FAT | Natural Acc66.74 | 46 | 2mo ago | |
| ImageNet 1k (test) | RVT-B* | FGSM Robustness53 | 34 | 3mo ago | |
| CIFAR-10 | Proposed method (FGSM) | FGSM Robust Accuracy99.2 | 30 | 1mo ago | |
| CIFAR100 1000 images (val) | SA | Clean Acc70.25 | 24 | 3mo ago | |
| CIFAR10 1000 images (test) | Square Attack (SA) | Robust Accuracy72.3 | 24 | 3mo ago | |
| GCG | GCG Rate0.13 | 21 | 3mo ago | ||
| CIFAR-100 LT (test) | AT-BSL-AuA | Clean Accuracy50.66 | 20 | 3mo ago | |
| GTSRB (test) | FGSM-Neuro-Symbolic | FGSM Accuracy74 | 18 | 3mo ago | |
| AutoDAN | ASR0 | 18 | 3mo ago | ||
| PAIR | ASR26 | 18 | 3mo ago | ||
| ImageNet 1k (val) | BG_Random | Accuracy13.4 | 18 | 3mo ago | |
| CIFAR-100 | Cutout | Final Auto-Attack Accuracy98.5 | 16 | 2mo ago | |
| ImageNet sr=90% (val) | AT | Clean Accuracy83.29 | 14 | 3mo ago | |
| Skill-Composed 2k queries (test) | Explicit Refusals Count169 | 12 | 1mo ago | ||
| WildJailbreak 2k queries (test) | AUTOSKILL | Number of Explicit Refusals498 | 12 | 1mo ago | |
| HarmBench | DR56.25 | 12 | 3mo ago | ||
| MUSE-Book Harry Potter | SFT | ASR1.5 | 11 | 1mo ago | |
| Adversarial Robustness Prompts | GCG ASR92 | 11 | 3mo ago | ||
| NSFW | AdvUnlearn | ASR4.69 | 11 | 3mo ago | |
| CIFAR-10 L2, epsilon=8/255 | Stutz et al. (2020) | Robust Acc (AA)77.64 | 11 | 3mo ago | |
| HarmBench | DirReq17 | 10 | 22d ago | ||
| RoboPAIR Adversarial Behaviors | SPINE | Attack Success Rate92.3 | 10 | 3mo ago | |
| STL10 (test) | ZePAD | Baseline Acc82.07 | 10 | 3mo ago |
