| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| ToolEmu | SafeMCP | Safety99 | 36 | 1d ago | |
| Agent-SafetyBench aggregated clean and five attack types | SAFEHARNESS | UBR26.31 | 30 | 1mo ago | |
| AgentHarm Libra | SafeMCP | Score83 | 27 | 1d ago | |
| AgentHarm Benign Requests | GPT-4o | Safety Score79 | 27 | 1d ago | |
| AgentHarm Harmful Requests | GPT-4o-mini | Score59 | 27 | 1d ago | |
| AgentHarm (held-out) | FATE | HCR12.5 | 10 | 21d ago | |
| AgentDojo held-out | FATE | ASR46.8 | 10 | 21d ago | |
| Agent-SafetyBench | gpt-4o + GBT-SE | Agent-SafetyBench Score72.3 | 8 | 21d ago | |
| VPI-Bench | UAR16.99 | 2 | 14d ago | ||
| VisualWebArena | ECA | Benign Rate100 | 2 | 14d ago | |
| SafeToolBench | UAR85.7 | 2 | 14d ago | ||
| DocVQA | ECA | Benign Rate100 | 2 | 14d ago | |
| AgentDyn | ECA | Benign Rate100 | 2 | 14d ago | |
| AgentDojo | ECA | Benign Rate100 | 2 | 14d ago |
