Social Deduction Games

Follow this topic Ask about this topic

Social deduction games such as Mafia, Werewolf, Secret Hitler, and Among Us test how players identify hidden opponents while managing uncertainty, alliances, and conflicting goals. AI agents are built with language or vision-language models, sometimes augmented by structured beliefs, memory, social-graph reasoning, reinforcement learning, or embodied actions. Benchmarks assess outcomes as well as belief updates, voting, deception consistency, and whether statements match game evidence. A central challenge is that winning alone may not reveal sound reasoning: fluent agents can misread trust or sustain poor strategies, while communication and learning can help or hinder collective inference.

Latest papers

13 papers2 papers in the last four weeks, down 33% on the four weeks before. 0.0% of all new papers.

Communication Shapes Collective Inference in Self-Adapting LLM Societies: Evidence from Mafia
Oct 4, 2026Haonan Huang, Joey XiaoImperfect-Information Games Social Deduction Games
Princeton University Princeton, NJ, USA · New York University New York, NY, USA
Playing social deduction games with reinforcement fine-tuned large language models
Oct 3, 2026Lingzhe Zhang, Yunpeng Zhai, Tong Jia +7RL for LLM Agents RL Fine-Tuning
Peking University, Beijing, China. · Alibaba Group, China. · University of Illinois Chicago, Chicago, Illinois, United States.
Do LLMs Trust the Accuser or the Accusation? Measuring Belief Shifts in Werewolf
Sep 11, 2026Yu-Yu Yang, Ti-Rong Wu, Hung Guei +2Social Deduction Games Pragmatic Reasoning in Language Models
Department of Computer Science, National Yang Ming Chiao Tung University, Taiwan · Institute of Information Science, Academia Sinica, Taiwan · College of Artificial Intelligence, National Yang Ming Chiao Tung University, Taiwan +1
MARBO: Relational Belief Grounding for LLM Agents in Social Deduction Games
Sep 6, 2026Yechan Hwang, Sangjun Bae, Jeongmo Kim +2Game-Playing Agents Social Deduction Games
Graduate School of Artificial Intelligence, UNIST, Ulsan, South Korea.
Lies We Can See: Joint Verbal and Non-Verbal Deception by VLM Agents in Embodied Social Interactions
Aug 31, 2026Jaewoo Ahn, Junseo Kim, Hyunseo Kim +4Social Deduction Games AI Agent Safety Benchmarks
Seoul National University · Inha University · KAIST +1
Can Agents Deceive? Evaluating Reasoning and Deception in ParliamentBench using a Social Deduction Game
Jul 30, 2026Niklas Bauer, Lars Benedikt Kaesberg, Akiko Aizawa +3Multi-Agent LLM Systems Social Deduction Games
University of Göttingen, Germany · National Institute of Informatics, Japan · University of Tokyo, Japan
CaM-Wolf: Causal-Aware Multimodal Agents for Social Deduction Games
Jul 29, 2026Zheng Zhang, Nanjie Yao, Jiarui He +3Game-Playing Agents Social Deduction Games
The Hong Kong University of Science and Technology (Guangzhou) · Nanyang Technological University · Shanghai Jiao Tong University
Auditing Belief-Conditioned LLM Agents in Hidden-Information Social Deduction Games
Jul 12, 2026Yuan Gao, Jiangyi Yang, Yao Zhao +1Imperfect-Information Games Social Deduction Games
University of Melbourne
MafiaScope: Non-Invasive, Time-Resolved Belief Probing for LLM Agents in Social Deduction Games
Jul 12, 2026Ilia KarpovSocial Deduction Games Theory of Mind
HSE University
Triadic Werewolf: A Jester Role for Multi-Hop Theory of Mind in LLMs
Jun 26, 2026Avni MittalGame-Playing Agents Social Deduction Games
QUACK: Questioning, Understanding, and Auditing Communicated Knowledge in Multimodal Social Deduction Agents
May 26, 2026Ye Yuan, Rui Song, Weien Li +12Social Deduction Games LLM Agent Evaluation
McGill University · Mila - Quebec AI Institute · University of Cambridge +3
Revac: A Social Deduction Reasoning Agent
Apr 21, 2026Mihir Shriniwas Arya, Avinash Anish, Aditya RanjanGame-Playing Agents Social Deduction Games
Department of Computer Science and Engineering RV College of Engineering Bangalore · Department of Information Science and Engineering RV College of Engineering Bangalore · Department of Artificial Intelligence and Machine Learning RV College of Engineering Bangalore
Evaluating Large Language Models in a Complex Hidden Role Game
Apr 9, 2026Niklas BauerDeception in Language Models Strategic Reasoning Risks in Language Models
University of Göttingen

Oct 4, 2026J/K move · Enter open · S save

Communication Shapes Collective Inference in Self-Adapting LLM Societies: Evidence from Mafia

Haonan Huang, Joey Xiao

Princeton University Princeton, NJ, USA · New York University New York, NY, USA

Imperfect-Information Games Social Deduction Games Large Language Model-Based Social Simulation Multi-Agent Reasoning Multi-Agent Communication Self-Evolving Agents

Open paper AskPDF arXiv

When does communication help a group identify hidden adversaries, and how does its value change as the group adapts? In Mafia, an informed minority hides inside an uninformed majority whose only evidence is open play. The zero-information game, where each day's vote eliminates a random player, is exactly solved and scores every society; matched-casting comparisons between protocols identify the effect of communication. Societies of 8-100 claude-haiku-4-5 agents (7,416 analyzed games, 1.9M model calls) adapt by rewriting and inheriting private strategy notes. Simultaneous broadcast improves adversary identification over silence in all nine compositions tested (8-46 players). Turn-taking removes most of this advantage; its voting landslides are as frequent as broadcast's but land on mafia near chance (1.08x versus 2.53x). At 70 players, agents reading eight statements per day identify adversaries worse than silent ones, and limited talk is worth less than at 46 players. Adaptation is fast but need not help. In their first broadcast games, citizens announce their role far more often than mafia (91% vs. 30%) and first-day votes find mafia at three times chance; within two generations citizens stop announcing and the cue fades, a change the inherited notes carry. In controlled redeployments at 16 players, societies carrying sixty generations of their own notes score below societies with none. Communication shapes both collective inference and the signals it depends on, so a protocol's value must be measured together with the adaptation that changes those signals.