Multi-turn Deception Detection on OpenDeception
0.654Response AUROCDECOR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DECORBackbone=GPT-4o2026.05 | 0.654 | 0.772 | |
| CoT-Red-HandedBackbone=GPT-4o2026.05 | 0.597 | 0.615 | |
| Constitutional MonitorBackbone=GPT-4o2026.05 | 0.548 | 0.617 | |
| Prompt-based Zero-shotBackbone=GPT-4o, Mode=Zero-shot2026.05 | 0.399 | 0.469 | |
| DeceptionBenchBackbone=GPT-4o2026.05 | 0.304 | 0.443 |