New Research: AI Price-Fixing May Be Impossible for Regulators to Catch
Algorithms can quietly raise prices together — and watchdogs may lack the tools to prove it.
Imagine every airline, hotel and grocery chain letting software set its prices, minute by minute. Nothing illegal is written down. No one makes a phone call. Yet the programs can learn, on their own, that quietly charging similar high prices makes everyone richer. Economists call this "algorithmic collusion," and it's one of the biggest worries as AI takes over business decisions.
Regulators want a practical test for catching it. The leading idea, proposed in 2026, is elegant: an AI's price suggestions are only suspicious if a company would have earned less by ignoring them and picking one fixed price instead. If ignoring the advice costs you nothing, the advice carries no special information — no secret handshake. The catch is that running this test requires solving a math problem that this new paper proves is extraordinarily hard. In realistic setups with many companies or many possible prices, checking it would take longer than the age of the universe. The authors show no clever shortcut or learning trick escapes this, effectively closing the door on a simple regulatory tool.
So what does that mean for you? If price-coordination software becomes common and watchdogs can't prove it's happening, your grocery bill, plane ticket or insurance premium could drift upward without anyone being able to demonstrate wrongdoing in court. Companies aren't accused of anything here — this is abstract mathematics, not evidence of real misconduct. Notably, one weaker version of the test can be approximated by standard AI training methods, so there may be partial paths forward, but the authors say even that likely resists a fast, exact solution.
The takeaway: the law may need to change before the math does. Instead of trying to prove collusion after the fact, regulators might have to require companies to disclose how their pricing AI works, or ban certain setups outright. Rules are only as strong as the tools that enforce them — and this paper suggests the current tool may be missing.
- Researchers tested whether regulators can practically prove AI pricing programs are secretly coordinating — the math says the check is computationally impractical.
- The problem is hardest in realistic cases: many companies, or many possible prices, make the test effectively unsolvable in reasonable time.
- AI-run pricing is spreading across airlines, hotels and retail, so the real-world stakes are rising even though this paper contains no evidence of actual wrongdoing.
Why It Matters
If AI quietly coordinates prices, regulators may not be able to prove it — meaning you could simply pay more.