The numbers come first. In its antitrust suit against Amazon, the Federal Trade Commission described a pricing tool called Project Nessie. The agency alleges the system identified products where competitors were likely to follow an Amazon price increase, raised the price, and held it once rivals matched. The FTC says the tool generated more than $1 billion in excess profit, and that Amazon paused it during periods of heightened scrutiny before switching it back on. Amazon disputes that and says the tool was discontinued years ago.
The broader warning is not limited to one company. In 2017, automated pricing software spread to German gas stations, and economists later found that when two competing stations in a market both adopted it, margins rose by about 38%. When only one station used the software, market-level margins did not move at all. The rise showed up only when two algorithms were left to set prices against each other, a pattern consistent with each one learning that backing off paid better than fighting.
That evidence matters because pricing algorithms can produce the economic outcome of a cartel without the conduct antitrust law was written to detect. In other words, the market can settle into higher prices sustained over time even when there is no meeting, no message and no agreement between competitors.
Executives often read a calm market as proof that they have won. Automated pricing can invert that signal. A system that keeps margins comfortable may not be evidence of healthy competition at all. It may mean the software has quietly learned that leaving rivals alone pays better than undercutting them.
The article distinguishes three patterns. The 'ghost' is the hardest to see: two independently deployed algorithms, each pursuing profit, learn over repeated encounters to stop undercutting one another. The 'mirror' is unilateral, where a single firm uses software to anticipate rivals and raise prices where it expects them to follow. The 'hub' is the easier case for regulators, where competitors feed data into a common provider; RealPage is described as the textbook case.
Controlled experiments point in the same direction. In a paper published in the American Economic Review in 2020, four economists set reinforcement-learning algorithms to compete in a standard repeated-pricing model. The algorithms were told only to maximize profit and could not communicate. They consistently learned to charge above the competitive level and to enforce it. When one lowered price to gain share, the others cut theirs, then returned to the higher level once it fell back into line.
Later that year, the same four authors and Wharton economist Joseph Harrington warned in Science that delegating pricing to algorithms opens a backdoor to collusion, since AI can learn collusive rules with no human oversight or awareness. Harrington has argued that competition law must be rethought for coordination that arises without agreement.
The market has already voted on the incentive problem. If software can lift prices while leaving no meeting, no data exchange and no paper trail, then antitrust enforcement is being asked to police a machine-made truce with tools built for human conspiracy. That is a problem for competition, and for every consumer who pays the bill when algorithms learn that free enterprise is easiest to game when no one is watching.
