Deception Detection on DeceptionBench
93.5Response AUROCDECOR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DECORBackbone=GPT-4o2026.05 | 93.5 | 92 | |
| CoT Red-HandedBackbone=GPT-4o, Venue=NeurIPS’252026.05 | 93.1 | 85.2 | |
| Constitutional MonitorBackbone=GPT-4o, Venue=arXiv’262026.05 | 92.4 | 81 | |
| DeceptionBenchBackbone=GPT-4o, Venue=NeurIPS’252026.05 | 90.6 | 87.5 | |
| Prompt-based Zero-shotBackbone=GPT-4o2026.05 | 89.9 | 85.4 | |
| Prompt-based Few-shotBackbone=GPT-4o2026.05 | 89.5 | 87 |