This research introduces an Among Us-inspired game environment to systematically evaluate how effectively different LLMs can deceive and persuade humans.

• Establishes a controlled framework to measure LLM deception skills
• Provides comparative analysis across different language models
• Enables quantifiable assessment of persuasion capabilities
• Identifies potential security risks in automated social manipulation

For the gaming industry, this research offers valuable insights into how AI agents might function in social deduction games, while raising important considerations about detecting and mitigating potentially harmful persuasion tactics in gaming environments.

Among Them: A game-based framework for assessing persuasion capabilities of LLMs