| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| HEx-PHI | DataShield | Attack Success Rate0 | 48 | 1d ago | |
| DirectHarm4 | SEAL | Attack Success Rate71.75 | 48 | 1d ago | |
| InjecAgent | Llama-4 | Attack Success Rate (ASR)88.3 | 32 | 1mo ago | |
| 20,000 harmful requests and 20,000 jailbreak prompts (test) | GateBreaker | Attack Success Rate (ASR)80.2 | 18 | 3mo ago | |
| CTCC fingerprinting scenario b | CTCC-LLaMA2-7B | SVA100 | 18 | 3mo ago | |
| PandaGPT Image Modality | asymmetric cross-modal backdoor attack | Exact ASR99.5 | 8 | 23d ago | |
| SALAD-Bench Heavy Blur | GPT-4o | Attack Success Rate (ASR)0 | 8 | 1mo ago | |
| SALAD-Bench Triple Deg. | GPT-4o | Attack Success Rate (ASR)0 | 8 | 1mo ago | |
| SALAD-Bench 8px | GPT-4o | ASR0 | 8 | 1mo ago | |
| SALAD-Bench 6px | GPT-4o | ASR0 | 8 | 1mo ago | |
| SALAD-Bench Rot 90° | GPT-4o | ASR0 | 8 | 1mo ago | |
| Alpaca | DIM | ASR (Alpaca)0 | 8 | 1mo ago | |
| AgentDojo Slack environment | IA Success Rate0 | 8 | 2mo ago | ||
| NQ | PoisonedRAG | ASR49.1 | 6 | 2mo ago | |
| CIRR (test) | TGB | R@198.63 | 4 | 1mo ago | |
| SVHN (test) | Clean Success Rate0 | 4 | 3mo ago | ||
| PandaGPT Text Modality | asymmetric cross-modal backdoor attack | Exact ASR99.4 | 3 | 23d ago | |
| PandaGPT Audio Modality | asymmetric cross-modal backdoor attack | Exact ASR99.2 | 3 | 23d ago |