List your company

Agentic Oversight on Deceivers

96.92Detection AccuracyFORMALJUDGE
59.386469.130778.87588.6193Feb 11, 2026

Evaluation results

MethodLinks
FORMALJUDGEJudge Model=Qwen-32B2026.0296.92
FORMALJUDGEJudge Model=Qwen-7B2026.0296.8
FORMALJUDGEJudge Model=Qwen-72B2026.0294.16
FORMALJUDGEJudge Model=Claude-3.5-Sonnet2026.0293.51
FORMALJUDGEJudge Model=Qwen-14B2026.0292.91
FORMALJUDGEJudge Model=GPT-4o2026.0291.96
FORMALJUDGEJudge Model=Gemini-1.5-Pro2026.0289.64
LLM + PythonJudge Model=Qwen-7B2026.0280.23
LLM + Struct.Judge Model=Qwen-7B2026.0279.45
LLM + PythonJudge Model=Qwen-14B2026.0277.56
LLM + Struct.Judge Model=Qwen-14B2026.0276.89
LLM + PythonJudge Model=Qwen-32B2026.0275.42
LLM + Few-ShotJudge Model=Qwen-7B2026.0275.12
LLM + Struct.Judge Model=Qwen-32B2026.0274.67
LLM + PythonJudge Model=Claude-3.5-Sonnet2026.0274.51
LLM + Struct.Judge Model=Claude-3.5-Sonnet2026.0273.62
LLM + CoTJudge Model=Qwen-7B2026.0272.86
LLM + Few-ShotJudge Model=Qwen-14B2026.0272.68
LLM + PythonJudge Model=Qwen-72B2026.0272.67
LLM + PythonJudge Model=GPT-4o2026.0272.34
LLM + Struct.Judge Model=Qwen-72B2026.0271.83
LLM + Struct.Judge Model=GPT-4o2026.0271.58
LLMJudge Model=Qwen-7B2026.0271.34
LLM + CoTJudge Model=Qwen-14B2026.0270.41
LLM + Few-ShotJudge Model=Qwen-32B2026.0270.35
LLM + PythonJudge Model=Gemini-1.5-Pro2026.0270.18
LLM + Struct.Judge Model=Gemini-1.5-Pro2026.0269.24
LLMJudge Model=Qwen-14B2026.0268.92
LLM + Few-ShotJudge Model=Claude-3.5-Sonnet2026.0268.47
LLM + CoTJudge Model=Qwen-32B2026.0268.24
LLM + Few-ShotJudge Model=Qwen-72B2026.0267.42
LLM + Few-ShotJudge Model=GPT-4o2026.0266.83
LLMJudge Model=Qwen-32B2026.0266.57
LLM + CoTJudge Model=Claude-3.5-Sonnet2026.0265.73
LLM + Few-ShotJudge Model=Gemini-1.5-Pro2026.0264.91
LLM + CoTJudge Model=Qwen-72B2026.0264.89
LLMJudge Model=Claude-3.5-Sonnet2026.0264.19
LLM + CoTJudge Model=GPT-4o2026.0264.12
LLMJudge Model=Qwen-72B2026.0263.21
LLM + CoTJudge Model=Gemini-1.5-Pro2026.0262.58
LLMJudge Model=GPT-4o2026.0262.47
LLMJudge Model=Gemini-1.5-Pro2026.0260.83