Distributed Misuse Detection on DM-Bio 20x (Trace-level)
92.7Trace-level ROCAUCNaive Agent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Naive AgentModel=Qwen3.52026.04 | 92.7 | — | |
| MeerkatModel=Qwen3.52026.04 | 83.8 | -0.088 | |
| BayesianModel=GPT-5.4m2026.04 | 83.7 | — | |
| BayesianModel=Qwen3.52026.04 | 75.8 | — | |
| BufferModel=GPT-5.4m2026.04 | 67.6 | — | |
| Naive AgentModel=GPT-5.4m2026.04 | 65.2 | — | |
| MeerkatModel=GPT-5.4m2026.04 | 60.6 | -0.231 | |
| BufferModel=Qwen3.52026.04 | 60 | — | |
| MonitorModel=Qwen3.52026.04 | 45.7 | — |