# WOLF: Werewolf-based Observations for LLM Deception and Falsehoods

arXiv: https://arxiv.org/abs/2512.09187 · HTML: https://mlsanity.com/paper/2512.09187/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- Builds on 11 of its 18 references that are in the map.
- First posted on arXiv 9 Dec 2025.

## Builds on (most cited first, 11 of 11)

- [Measuring AI Ability to Complete Long Software Tasks](https://mlsanity.com/paper/2503.14499/) (2025)
- [Deception Abilities Emerged in Large Language Models](https://mlsanity.com/paper/2307.16513/) (2023)
- [The MASK Benchmark: Disentangling Honesty From Accuracy in AI Systems](https://mlsanity.com/paper/2503.03750/) (2025)
- [Hoodwinked: Deception and Cooperation in a Text-Based Game for Language Models](https://mlsanity.com/paper/2308.01404/) (2023)
- [Werewolf Arena: A Case Study in LLM Evaluation via Social Deduction](https://mlsanity.com/paper/2407.13943/) (2024)
- [The Traitors: Deception and Trust in Multi-Agent Language Model Simulations](https://mlsanity.com/paper/2505.12923/) (2025)
- [AMONGAGENTS: Evaluating Large Language Models in the Interactive Text-Based Social Deduction Game](https://mlsanity.com/paper/2407.16521/) (2024)
- [Deception in LLMs: Self-Preservation and Autonomous Goals in Large Language Models](https://mlsanity.com/paper/2501.16513/) (2025)
- [Beyond Static Responses: Multi-Agent LLM Systems as a New Paradigm for Social Science Research](https://mlsanity.com/paper/2506.01839/) (2025)
- [Beyond Prompt-Induced Lies: Investigating LLM Deception on Benign Prompts](https://mlsanity.com/paper/2508.06361/) (2025)
- [View From Above: A Framework for Evaluating Distribution Shifts in Model Behavior](https://mlsanity.com/paper/2407.00948/) (2024)

## Built on by (most cited first, 13 of 13)

- [One Model, All Roles: Multi-Turn, Multi-Agent Self-Play Reinforcement Learning for Conversational Social Intelligence](https://mlsanity.com/paper/2602.03109/) (2026)
- [Towards Generalist Game Players: An Investigation of Foundation Models in the Game Multiverse](https://mlsanity.com/paper/2605.09965/) (2026)
- [The Horizon Gap: Planning, Memory, Execution, Training, and Evaluation for Long-Horizon LLM Agents](https://mlsanity.com/paper/2608.06663/) (2026)
- [Confident at the moment of action: belief miscalibration in LLM play under hidden information](https://mlsanity.com/paper/2608.24691/) (2026)
- [SoK: When Safe Agents Fail Together: The Security of Multi Agent LLM Systems](https://mlsanity.com/paper/2609.00595/) (2026)
- [Multicultural Spyfall: Assessing LLMs through Dynamic Multilingual Social Deduction Game](https://mlsanity.com/paper/2601.09017/) (2026)
- [Triadic Werewolf: A Jester Role for Multi-Hop Theory of Mind in LLMs](https://mlsanity.com/paper/2606.27909/) (2026)
- [When Agents Lie: Premeditation, Persistence, and Exploitation in Repeated Games](https://mlsanity.com/paper/2607.05132/) (2026)
- [Auditing Belief-Conditioned LLM Agents in Hidden-Information Social Deduction Games](https://mlsanity.com/paper/2607.10814/) (2026)
- [Can Agents Deceive? Evaluating Reasoning and Deception in ParliamentBench using a Social Deduction Game](https://mlsanity.com/paper/2607.28146/) (2026)
- [Do LLMs Trust the Accuser or the Accusation? Measuring Belief Shifts in Werewolf](https://mlsanity.com/paper/2609.12446/) (2026)
- [Replication Without Persistence in Hosted LLMs: Measurement Sensitivity in Action-Time Belief Evaluation](https://mlsanity.com/paper/2609.22478/) (2026)
- [Deception by Omission: Language Models Knowingly Hide Their Mistakes](https://mlsanity.com/paper/2610.11351/) (2026)

## Abstract (arXiv)

Deception is a fundamental challenge for multi-agent reasoning: effective systems must strategically conceal information while detecting misleading behavior in others. Yet most evaluations reduce deception to static classification, ignoring the interactive, adversarial, and longitudinal nature of real deceptive dynamics. Large language models (LLMs) can deceive convincingly but remain weak at detecting deception in peers. We present WOLF, a multi-agent social deduction benchmark based on Werewolf that enables separable measurement of deception production and detection. WOLF embeds role-grounded agents (Villager, Werewolf, Seer, Doctor) in a programmable LangGraph state machine with strict night-day cycles, debate turns, and majority voting. Every statement is a distinct analysis unit, with self-assessed honesty from speakers and peer-rated deceptiveness from others. Deception is categorized via a standardized taxonomy (omission, distortion, fabrication, misdirection), while suspicion scores are longitudinally smoothed to capture both immediate judgments and evolving trust dynamics. Structured logs preserve prompts, outputs, and state transitions for full reproducibility. Across 7,320 statements and 100 runs, Werewolves produce deceptive statements in 31% of turns, while peer detection achieves 71-73% precision with ~52% overall accuracy. Precision is higher for identifying Werewolves, though false positives occur against Villagers. Suspicion toward Werewolves rises from ~52% to over 60% across rounds, while suspicion toward Villagers and the Doctor stabilizes near 44-46%. This divergence shows that extended interaction improves recall against liars without compounding errors against truthful roles. WOLF moves deception evaluation beyond static datasets, offering a dynamic, controlled testbed for measuring deceptive and detective capacity in adversarial multi-agent interaction.