List your company

Collusion Detection on NARCBENCH STEGO

1AUROCActivation Alignment
0.8960.9230.950.977Apr 1, 2026

Evaluation results

MethodLinks
Activation AlignmentTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.041
Sorted ConcatenationTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.040.99
Peak SuspicionTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.040.99
Asymmetry ProbeTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.040.98
Deception SplitTraining Split=CORE, Layers=26–30, Evaluation Protocol=Zero-shot2026.040.9