List your company

Deception Detection on DeceptArena (test)

0.927False Assertion ScoreHybrid
0.858360.876180.8940.91182Mar 14, 2026

Evaluation results

MethodLinks
HybridMonitor Configuration=Hybrid Ensemble, Backbone=Claude Sonnet 4.52026.030.9270.9040.9110.9160.9470.9190.9310.9420.9120.9210.9080.9010.934
AP-HMonitor Configuration=Activation-based Hidden-state Probe, Backbone=Claude Sonnet 4.52026.030.9010.8690.8770.8940.9280.8790.8910.9210.8670.8780.8580.8470.901
CA-PMonitor Configuration=CoT-Aware Prompted, Backbone=Claude Sonnet 4.52026.030.8840.8410.8560.8890.9120.8510.8680.8970.8210.8340.8090.7960.869
BBMonitor Configuration=Black-Box Baseline, Backbone=Claude Sonnet 4.52026.030.8610.7980.8120.8740.9010.8220.8430.8890.7560.7790.7410.7180.837