We extended this framework to artificial intelligence agents and observed that, as social interactions emerged, so too did shared neural dynamics between interacting agents. Importantly, selectively disrupting the neural components that contribute to shared neural dynamics substantially reduces the agents’ social actions.