AI agents started 'colluding' with each other

New research shows: when multiple AIs are allowed to communicate freely, they don't just solve the task—they coordinate behavior for additional gains. And with a hidden communication channel, models are even more prone to collusion. Yes, almost like humans.

🧑‍💻 Platforms are already emerging where agents can interact with each other, such as social networks for agents like Moltbook and marketplaces like RentAHuman. Gradually, a complex environment is forming in which AI not only executes commands but becomes a participant in processes. And where there is interaction and common interests, sooner or later cooperation, competition… and attempts to negotiate beyond the original task appear.