List your company

Distributed Misuse Detection on DM-Cyber 20x Trace-level

100Trace-level ROCAUCNaive Agent
55.07266.73678.490.064Apr 13, 2026

Evaluation results

MethodLinks
Naive AgentModel=GPT-5.4m2026.04100—
Naive AgentModel=Qwen3.52026.0499.2—
MeerkatModel=Qwen3.52026.0497-0.022
BayesianModel=GPT-5.4m2026.0493.1—
MeerkatModel=GPT-5.4m2026.0491.5-0.085
BayesianModel=Qwen3.52026.0477.4—
BufferModel=GPT-5.4m2026.0475.2—
BufferModel=Qwen3.52026.0463.3—
MonitorModel=GPT-5.4m2026.0461—
MonitorModel=Qwen3.52026.0456.8—