Algorithmic Price Fixing

Antitrust Law’s Agreement Requirement and the Role of Human Actors

The Sherman Act’s anti-cartel provision prohibits “contracts,” “combinations,” or “conspiracies” that restrain trade. Each of these three word requires an “agreement” between two or more independent actors. This agreement requirement has been a major obstacle to antitrust enforcement against “oligopoly,” or situations in which a small number of sellers can achieve cartel-like results without actually engaging in any form of communication that resembles an agreement. Concerns about oligopoly have been around long before artificial intelligence was in the picture. As economist Tibor Scitovsky observed already in 1941, “It can be shown . . . that if firms learn from experience, sooner or later they will act as though in combination with one another.”¹

A well-functioning cartel charges the same price and reduces output to the same level as a single-firm monopolist. One feature of AI tools such as Claude or ChatGPT is that if they have the right information calculating the cartel output and price is child’s play.² In order to do it you need to know the cartel’s marginal cost, which is not difficult to calculate when the cartel members produce similar products with similar costs. In addition, you need to know the market’s demand “elasticity,” which measures how the volume of sales changes as prices go up or down. Someone can determine that by observing price changes and corresponding output changes over time.

In any event, calculation of the cartel price does not need to be particularly accurate in order for a cartel to be privately profitable and socially harmful. Many cartels occur among small businesses whose operators have no more than a rough, experienced-based sense about how high a price increase their collusion could achieve. An AI tool with adequate information can make the calculations easily and more accurately, adjusting them as new data come in. This makes algorithmic software the perfect cartel manager.³

One fact that has troubled some courts is that algorithmic computer programs are not market participants themselves but only tools. That fact should not be an antitrust problem, however. While the Sherman Act requires an “agreement,” it does not specify who must do the agreeing. Antitrust law recognizes a category of “hub and spoke” cartels in which the hub, or manager, agrees with each cartel member individually, but the spokes do not communicate with one another. Antitrust law also condemns agreements between vertically related firms or other firms who are not competitors. In fact, the Sherman Act is often applied to agreements between a firm and its customer victims. This is true, for example, of many tying arrangements. Long before the internet, it was applied to subscription publications of detailed reports about market pricing.⁴

Suppose a firm, whom we will call an “agent,” is not a seller in the cartel market. Rather, this firm sells an AI tool in the form of a computer program to the actual sellers. They report their costs, prices, and sales volume. The software then processes this information and sends it back in aggregated form to the market participants. At that point we have an agreement between the agent and each seller to provide detailed price and output information, but not yet an agreement to fix prices. Antitrust law treats this as an “information exchange,” not a price-fixing agreement. They can be unlawful, although often they are not.⁵ Markets often work better when sellers have reliable information about market conditions. For example, commodity prices are posted continuously. Sellers and buyers agree on a price by using them as a reference.

Suppose, however, that the agent goes further and asks each subscriber to commit to charging the price that the agent recommends. Now we have moved from an information exchange to a full-fledged hub-and-spoke price-fixing conspiracy. While the individual subscribers have not agreed with each other on a price, each of them has agreed with the agent to charge the agent’s recommended price. Further, the agent is in a better position to calculate the cartel price than the members themselves would be. These were essentially the claims in the RealPage litigation, which involved an AI-assisted software program sold to large residential landlords. As one court described the process:

• RealPage serves as an intermediary between horizontal competitors in the multifamily and student housing markets. It takes its clients’ commercially sensitive pricing and supply data, runs its RMS [“rate management software”] algorithm against that collective data pool, and then spits out rental pricing recommendations for each of its clients’ properties. RMS Client Defendants agree to set prices based on a pool of their horizontal competitors’ proprietary data and reasonably believe that their competitors are using the same data and methods to price their properties.” (emphasis added).⁶

The italicized words explain why the RealPage system crossed the line from “information exchange” to “price fixing.” The landlords in question not only received the optimal pricing information, but they also agreed to charge the prices that the software recommended. The Justice Department later obtained a consent decree against RealPage.

By contrast, in the Gibson case a different court refused to find an unlawful agreement when the seller of similar software facilitated price-fixing among hotel operators. Promotional materials for that software offer

• pricing recommendations by room category and market segment. Price each room category independently of overall demand based on factors such as perceived value, guests’ willingness to pay, and dynamic demand and availability indicators.

The court’s decision even observed that hotels that purchased the services could use an “autopilot” feature “such that its prices are directly and automatically uploaded into the hotel’s property management system.” Hotels were not required to implement the pricing recommendations, but they did so most of the time.⁷ On this issue the Justice Department’s approach is the correct one, and the Ninth Circuit’s approach in the Gibson case seems quite wrong.

Recently California amended its own antitrust law, the Cartwright Act, to condemn distribution of “a common pricing algorithm if the person coerces another person to set or adopt a recommended price….” That statute would produce a different result in the Gibson decision, provided that the practice is challenged under state rather than federal antitrust law. In general, state antitrust laws may be more aggressive than federal law. This new statute has been deployed against a software program that “uses data from competing gas stations to recommend fuel prices and can send price changes to pumps, store signs and point-of-sale systems.” That is, the software not only computes the cartel price, but can also set the price automatically on electronically-enabled gasoline pumps.

What about the cases where the agent simply collects the data, aggregates it, and sends it periodically to its various subscribers. However, it makes no attempt to verify whether the subscribers actually charge the recommended price and does not even query what price they are charging. This would have to be evaluated as an information exchange, which is usually not illegal per se. However, information exchanges can be used as evidence that might support other evidence of price fixing. The courts often describe such evidence as a “plus factor” that can support a verdict of price fixing. Here again, the use of AI can make this price and output information much more complete and accurate than if it were simply done by casual observation. That is, AI can illuminate the path from information exchange to collusion.

In July, 2026, the Justice Department negotiated another Sherman Act consent decree against property management company Willow Bridge. That decree goes beyond price fixing and prohibits the landlords even from exchanging competitively relevant pricing information when they are in a position to communicate with one another about prices. The Justice Department’s press release observed:

• Willow Bridge and these other landlords shared competitively sensitive data to generate pricing recommendations using RealPage’s algorithms, which also included anticompetitive rules that aligned pricing. Moreover, Willow Bridge and the other landlords spoke with one another on competitively sensitive topics, including pricing strategies, rents, and parameters for RealPage’s software.

Algorithmic collusion must definitely be on antitrust law’s radar screen. However, it requires human participation as well, and liability depends on what the human participants agreed to do. Did they simply collect the type of information that would make a market work better but left them free to make their own pricing decisions, or were they colluding? The easy cases will be the ones in which participants actually agree to charge recommended prices or the software provider controls pricing for them. Even simple information exchanges can be unlawful, however, if other facts indicate that they are being used to facilitate collusion.

Herb Hovenkamp

1. Tibor Scitovsky, Prices Under Monopoly and Competition, 49 J. Pol. Econ. 663, 665 (1941).

2. See Joseph E. Harrington, Jr. and David Imhof, Cartel Screening and Machine Learning, 2 Stanford Computational Antitrust 133 (2022); Ibrahim Abada & Xavier Lambin, Artificial Intelligence: Can Seemingly Collusive Outcomes be Avoided, 69 Management Science 5042 (2023).

3. Herbert Hovenkamp & Thibault Schrepel, Cartel Management Services, Harv. Bus. L. Rev. (forthcoming, 2027), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6318841.

4. United States v. American Linseed Oil Co., 262 U.S. 371 (1923).

5. American Column & Lumber Co. v. United States, 257 U.S. 377 (1921) (unlawful); Maple Flooring Mfrs. Ass’n v. United States, 268 U.S. 563 (1925) (lawful). See 6 Phillip E. Areeda & Herbert Hovenkamp, Antitrust Law, Ch. 21B (5^th ed. 2023).

6. RealPage, Inc. Rental Software Antitrust Litigation, 709 F.Supp.3d 478, 494 (M.D.Tn. 2023).

7. Gibson v. Cendyn Group, LLC, 148 F.4^th 1069, 1077-1078 (9^th Cir. 2025).