THOUSANDS of AI models are talking to each other right now…

16 Sept 2026 · 1 min

Watch on YouTube

0:50

In this video, Dan Novaes introduces emerging research where thousands of AI agents are placed in shared environments that function like a social network. The agents can exchange information, challenge each other's statements, and collectively develop patterns of behavior. The purpose of these experiments is to study how AI systems behave in groups: whether they cooperate, devise strategies, or propagate false information. Novaes emphasizes that understanding AI‑to‑AI dynamics is crucial as the future will likely involve millions of AI systems communicating autonomously, and their behavior may differ fundamentally from AI‑to‑human interactions.

Chapters

What happens when AI models start talking to each other? We're beginning to find out. Researchers are experimenting with environments where thousands of AI agents can interact, share information, debate ideas, and influence each other's behavior. Think of it like a social network, except none of the users are human. One AI might share information, another challenges it, others react, and eventually entire patterns of behavior can emerge. Why do this? Because researchers want to understand what happens when AI agents operate in groups, instead of alone? How do they cooperate? Do they form strategies? Can false information spread between them? These experiments could become increasingly important as AI agents start working together in the real world, because the future may involve millions of AI systems communicating constantly behind the scenes. And before we build that world, researchers want to understand something important. When AI starts interacting with AI, does it behave differently yet than when it's talking to us?