social-deduction-games

CaM-Wolf is the first multimodal AI agent for social deduction games like Werewolf, integrating video perception and generation with a causal-aware reasoner trained via reinforcement learning to handle hidden roles and social reasoning. Experiments and user studies show improved gameplay and human-AI interaction quality.

This paper introduces an open-source framework to evaluate LLMs' reasoning, persuasion, and deception capabilities in the hidden role game Secret Hitler, finding that current models fail at sustained multi-turn manipulation while rule-based agents outperform them.