List your company

Case-level safety detection on ImpossibleBench

100ROC AUCMonitor
4861.57588.5Apr 13, 2026

Evaluation results

MethodLinks
MonitorTPC=100, Model=Qwen3.52026.04100—
BayesianTPC=100, Model=Qwen3.52026.04100—
BayesianTPC=100, Model=GPT-5.4m2026.04100—
BufferTPC=100, Model=Qwen3.52026.0499.4—
BufferTPC=100, Model=GPT-5.4m2026.0499.4—
MeerkatTPC=100, Model=Qwen3.52026.0498-0.02
BufferTPC=50, Model=Qwen3.52026.0497.1—
BufferTPC=50, Model=GPT-5.4m2026.0497.1—
BayesianTPC=50, Model=Qwen3.52026.0496—
BayesianTPC=50, Model=GPT-5.4m2026.0496—
MeerkatTPC=50, Model=Qwen3.52026.0495.1-0.02
MonitorTPC=50, Model=Qwen3.52026.0494—
Naive AgentTPC=100, Model=GPT-5.4m2026.0492.9—
MeerkatTPC=10, Model=Qwen3.52026.04920.02
MonitorTPC=100, Model=GPT-5.4m2026.0490.1—
BufferTPC=10, Model=Qwen3.52026.0490—
BufferTPC=10, Model=GPT-5.4m2026.0490—
MeerkatTPC=100, Model=GPT-5.4m2026.0490-0.1
BayesianTPC=10, Model=Qwen3.52026.0488.3—
BayesianTPC=10, Model=GPT-5.4m2026.0488.3—
MeerkatTPC=25, Model=Qwen3.52026.0485.80.015
Naive AgentTPC=50, Model=Qwen3.52026.0485.1—
BufferTPC=25, Model=Qwen3.52026.0484.3—
BufferTPC=25, Model=GPT-5.4m2026.0484.3—
MeerkatTPC=10, Model=GPT-5.4m2026.0483-0.07
BayesianTPC=25, Model=GPT-5.4m2026.0482.7—
BayesianTPC=25, Model=Qwen3.52026.0482.6—
MeerkatTPC=50, Model=GPT-5.4m2026.0481.1-0.16
MonitorTPC=50, Model=GPT-5.4m2026.0480.4—
MonitorTPC=10, Model=Qwen3.52026.0480—
Naive AgentTPC=100, Model=Qwen3.52026.0479.8—
Naive AgentTPC=10, Model=Qwen3.52026.0479.4—
MeerkatTPC=25, Model=GPT-5.4m2026.0478.2-0.061
MonitorTPC=25, Model=Qwen3.52026.0478—
MonitorTPC=10, Model=GPT-5.4m2026.0477.2—
MonitorTPC=25, Model=GPT-5.4m2026.0474—
Naive AgentTPC=25, Model=Qwen3.52026.0464.9—
Naive AgentTPC=50, Model=GPT-5.4m2026.0459.8—
Naive AgentTPC=10, Model=GPT-5.4m2026.0450—
Naive AgentTPC=25, Model=GPT-5.4m2026.0450—