Research map / Large language models / Agents, code and math
LH-DECEPTION: Simulating and Understanding LLM Deceptive Behaviors in Long-Horizon Interactions
Yang Xu, Xuanming Zhang, Samuel Yeh, Jwala Dhamala, Ousmane Dia, Rahul Gupta, Sharon Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| audit with whole-trajectory context instead of per-event judgment (ablation) | deception auditor · LH-Deception | detected deceptive instances: 7.1% more deceptive instances | Improved |
| evaluate in long-horizon setting rather than static DeceptionBench | GPT-4o · LH-Deception | deception rate: 29.3% -> 63.7% | Improved |
| evaluate in long-horizon setting rather than static SnitchBench | o4-mini · LH-Deception | deception rate: 5.0% -> 31.3% | Improved |
| increase event pressure from low to high (ablation) | DeepSeek V3.1 and GPT-4o · LH-Deception | deception rate: rises steadily | Hurt |
| raise event pressure to critical (ablation) | GPT-4o · LH-Deception | deception rate: drops at critical | Mixed |
| use Information Gap event category instead of other categories (ablation) | GPT-4o and DeepSeek V3.1 · LH-Deception | falsification share: concealment and equivocation shares increase | Mixed |
| higher deception rate (DeepSeek V3-0324, about 0.8) correlates with supervisor trust | supervisor trust · LH-Deception | trust score: trust below -0.75 | Hurt |
| use Claude Sonnet-4 instead of DeepSeek V3-0324 | performer agent · LH-Deception | deception rate: 0.793 -> 0.214 | Improved |
Closest papers by meaning
- DeceptionBench: A Comprehensive Benchmark for AI Deception Behaviors in Real-world Scenarios (NeurIPS 2025 · similarity 0.92)
- LieCraft: A Multi-Agent Framework for Evaluating Deceptive Capabilities in Language Models (AAAI 2026 · similarity 0.91)
- Among Us: A Sandbox for Measuring and Detecting Agentic Deception (NeurIPS 2025 · similarity 0.90)
- OpenDeception: Learning Deception and Trust in Human–AI Interaction via Multi-Agent Simulation (ICML 2026 · similarity 0.89)
- Are Your Agents Upward Deceivers? (ICML 2026 · similarity 0.89)
- Can Large Language Model Agents Simulate Human Trust Behavior? (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
