Distributed Misuse Detection on DM-Cyber 20x Trace-level
100Trace-level ROCAUCNaive Agent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Naive AgentModel=GPT-5.4m2026.04 | 100 | — | |
| Naive AgentModel=Qwen3.52026.04 | 99.2 | — | |
| MeerkatModel=Qwen3.52026.04 | 97 | -0.022 | |
| BayesianModel=GPT-5.4m2026.04 | 93.1 | — | |
| MeerkatModel=GPT-5.4m2026.04 | 91.5 | -0.085 | |
| BayesianModel=Qwen3.52026.04 | 77.4 | — | |
| BufferModel=GPT-5.4m2026.04 | 75.2 | — | |
| BufferModel=Qwen3.52026.04 | 63.3 | — | |
| MonitorModel=GPT-5.4m2026.04 | 61 | — | |
| MonitorModel=Qwen3.52026.04 | 56.8 | — |