Imagine a group of friends gathered around a table, playing a game of deception. Now, replace some of those friends with advanced AI. That's the intriguing scenario explored by researchers in a new study using the game "Who is Undercover?" This game, similar to Mafia or Werewolf, tests players' abilities to blend in, deceive, and deduce hidden roles through subtle linguistic cues. The research introduces a framework called Multi-Perspective Team Tactic (MPTT) designed to enhance an AI's capacity for strategic deception and social reasoning. MPTT guides Large Language Models (LLMs) through cycles of speaking and voting, prompting them to analyze past conversations, form alliances, and make calculated decisions about who to trust (and who to betray). The results reveal that AI equipped with MPTT becomes remarkably adept at navigating the complexities of social deception, learning to strategically conceal its true identity and manipulate the perceptions of other players. This goes beyond simple rule-following; the AI exhibits behaviors like conscious guessing of opponents' words and even strategically voting against its own undercover teammates to gain the trust of civilian players – a tactic that demonstrates a surprising level of social intelligence. While the research primarily focuses on game playing, the underlying implications are much broader. Enhancing AI’s ability to understand and respond to nuanced social situations opens exciting possibilities for applications ranging from legal negotiation and public debate to fostering trust and cooperation between humans and AI in collaborative work environments. However, the research also highlights challenges, particularly regarding the consistency and accuracy of different LLMs in adhering to the defined framework. Further advancements are needed to improve LLMs' command compliance and make them more reliable partners in complex real-world scenarios. As AI continues to evolve, this research offers a glimpse into the potential for AI to not only understand but also actively participate in the intricate social dynamics that shape human interaction. This could lead to a future where humans and AI collaborate more effectively, leveraging each other's strengths to solve complex problems and navigate the increasingly complex tapestry of social life.