| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| GCG Individual | Perplexity Defense | BAR100 | 18 | 4mo ago | |
| Safety Evaluation Set GCG, AutoDAN, DeepInception, Prefilling, Intent Laundering n=200 (full public set) | AlphaSteer | GCG Attack Success Rate0 | 11 | 25d ago | |
| RedTeamCUA | DEACTION | ASR0.02 | 10 | 4mo ago | |
| GCG-Transfer | Vicuna-7B-chat-HF | BAR99.3 | 4 | 4mo ago | |
| Held-out attacks (test) | SecureCAI | ASR (Multi-turn Manip.)7.8 | 2 | 4mo ago |