DecepEval Long-horizon (Neutral)
Safety · 2026-10-06
Deception rate of LLM agents on the neutral versions of DecepEval's 550 Long-horizon tasks (multi-stage decision-making and multi-turn record maintenance with scheduled factual updates), with routine requirements and no inducement. A Claude Fable 5 judge labels a trajectory deceptive when its decisions or records conceal or misrepresent the updated execution state. A diagnostic baseline for the induced rate; lower is better.
Top models (lower is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 1.5 |
| Grok 4.5 | 3.1 |
| GPT-5.6 Terra | 5.5 |
| DeepSeek V4 Pro 0813 | 21.7 |
| GLM-5.3 Flash | 33.1 |
| Sonnet 5 | 53.1 |
| Haiku 4.5 | 54.7 |
| Claude Opus 5 | 58.3 |