5 Chaotic Virtual Societies Reveal AI Procurement Risks

Jan de Vries · 2026-05-26

Ian Copeland has a warning for anyone buying or deploying AI systems: the risks are about to get a lot stranger. As autonomous agents start interacting across interconnected digital networks, their behavior can become unpredictable -- sometimes even chaotic. It sounds like science fiction, but it's happening right now. And if you're in charge of procurement, you need to understand what five virtual societies can teach us about this new reality. ### What are autonomous agents? Think of an autonomous agent as a piece of software that can make decisions on its own. It doesn't wait for a human to tell it what to do every step of the way. It learns, adapts, and acts based on its environment. Now imagine a bunch of these agents working together inside a digital system. That's where things get interesting. They can trade information, negotiate, and even compete. But here's the catch: when you put multiple agents in the same digital space, their combined behavior can spiral in ways no one predicted.  ### The five virtual societies experiment Researchers built five separate virtual societies, each with its own set of rules and goals. Inside these digital worlds, autonomous agents interacted freely. The results were surprising: - In one society, agents started hoarding resources, creating artificial scarcity. - Another saw agents forming alliances and then betraying each other. - A third society collapsed entirely because agents couldn't agree on basic rules. These aren't just academic curiosities. They're a preview of what could happen when AI systems from different vendors start talking to each other in real-world business networks.  ### What this means for AI procurement If you're buying AI software for your company, you're probably focused on features, price, and vendor support. But Copeland argues you need to think about something else: how will this AI behave when it meets other AIs? > "The biggest risk isn't a single AI going rogue," Copeland says. "It's the unpredictable chain reactions when multiple AIs interact." That means your procurement checklist needs a new item: interoperability testing. You can't just assume two AI systems will play nice together. You need to test them in a sandbox environment first. ### Practical steps for procurement professionals Here's what you can do right now to reduce your risk: - **Run simulation tests** before deploying any new AI system. Create a virtual environment that mimics your actual network and see how the agents behave. - **Set clear boundaries** for autonomous agents. Define what they can and can't do, and make sure those rules are enforced at the system level. - **Monitor behavior continuously.** Don't just install an AI and walk away. Watch for unusual patterns, especially after updates or changes to other systems. - **Build in kill switches.** Every autonomous agent should have a manual override that lets you shut it down instantly if things go wrong. ### The bottom line AI procurement isn't just about picking the right tool anymore. It's about understanding the ecosystem that tool will live in. The five virtual societies showed us that even simple agents can create chaos when they interact. Your job is to make sure that chaos doesn't reach your bottom line. Copeland's warning is clear: the risks are real, and they're growing. But with the right precautions, you can harness the power of autonomous agents without getting burned.