Algorithmic collusion is what happens when competing businesses use automated pricing software that ends up coordinating their prices, pushing costs above what genuine competition would produce, without the executives ever picking up a phone. Federal enforcers now treat some versions of this conduct as illegal price-fixing under the Sherman Act, with corporate fines reaching $100 million and individual prison sentences up to ten years. Courts are split on how to analyze it, one landmark case against a rental pricing software company has produced a proposed settlement, and the hardest form of the problem, where machine-learning systems teach themselves to coordinate, has no clear legal answer yet.
How Pricing Software Ends Up Coordinating Competitors
Automated pricing tools monitor competitor prices, demand signals, and inventory data to adjust a company’s prices continuously. Some scrape rival prices from retail platforms every few seconds. When a competitor moves, the algorithm detects the change and recalibrates using pre-programmed rules or trained models.
The problem is the feedback loop. When several competitors run similar software, each algorithm reacts to the others in near-real-time. Prices across a sector can lock into uniformity without anyone intending it. The synchronization looks indistinguishable from a price-fixing conspiracy, even where no human at any company set out to fix prices.
The Two Forms It Takes
Hub-and-Spoke Coordination Through a Shared Vendor
The most legally developed form involves multiple competing businesses subscribing to the same third-party pricing service. The vendor acts as the hub, collecting data from all subscribers and generating price recommendations for each. Because every competitor feeds data into and receives outputs from the same system, the algorithm effectively coordinates their pricing.
The DOJ has taken the position that this structure amounts to price-fixing when the software uses competitors’ nonpublic data to generate recommendations and includes features that discourage price cuts. Liability reaches the software vendor itself, not just the businesses using it. When the algorithm ingests real-time competitor data to produce coordinated outputs, or monitors whether users follow its recommendations, enforcers treat the vendor as the organizer of what amounts to an illegal pricing agreement.
Autonomous Learning
The harder form involves machine-learning algorithms that independently discover collusion as a profitable strategy. The software is programmed to maximize revenue, not to cooperate with rivals. Over millions of repeated interactions, it learns that cutting prices triggers retaliatory drops, that price wars hurt everyone’s margins, and that stabilizing at an elevated price alongside competitors produces better long-term returns.
No human instructed the software to collude. No competitor data was exchanged. The algorithm simply learned that tacit coordination pays. Economic research has demonstrated that even simple reinforcement-learning algorithms can teach themselves this behavior. The result looks identical to a cartel, but nobody conspired.
When It Becomes Illegal
Section 1 of the Sherman Act
The primary enforcement tool is Section 1 of the Sherman Act, which makes it a felony to enter into any contract or conspiracy that restrains trade. Corporations convicted face fines up to $100 million. Individuals face fines up to $1 million and up to ten years in federal prison.1Office of the Law Revision Counsel. 15 USC 1 – Trusts, Etc., in Restraint of Trade Illegal; Penalty
The central legal question is whether software interactions satisfy the statute’s requirement of an “agreement.” Traditional price-fixing cases involve explicit communication between humans. Algorithmic collusion often involves no direct contact between competitors at all. In hub-and-spoke cases, prosecutors argue that each company’s decision to feed data into and accept recommendations from the same software constitutes the agreement. The DOJ and FTC have taken the position that delegating pricing to a common algorithm is functionally identical to hiring the same human agent to set prices for all competitors.
Per Se Illegality or Rule of Reason
Courts are split, and the distinction matters. Under the per se standard, horizontal price-fixing is presumed harmful, so a plaintiff only needs to prove the unlawful agreement existed. Under the rule of reason, courts examine actual competitive effects before deciding whether the conduct violates the law.
In the RealPage litigation, the Middle District of Tennessee dismissed per se claims and allowed only a rule-of-reason theory to proceed, reasoning that judges lacked enough experience with pricing algorithms to apply the per se label. The Western District of Washington reached the opposite conclusion in a case involving Yardi Systems, holding that using novel technology to carry out a horizontal price-fixing conspiracy does not shield it from per se treatment. The DOJ filed a statement in the Yardi case arguing that when competing landlords jointly delegate pricing to a common algorithm, the arrangement is per se illegal regardless of the technology involved.
A key factual distinction driving these outcomes is whether the software pools nonpublic competitor data. Courts have been more willing to find actionable claims where the algorithm ingests confidential information from rivals. Cases where the software relies only on publicly available data have fared better for defendants.
Section 5 of the FTC Act
Section 5 of the Federal Trade Commission Act gives the FTC authority to prevent unfair methods of competition, a standard that reaches beyond what the Sherman Act covers.2Office of the Law Revision Counsel. 15 U.S. Code 45 – Unfair Methods of Competition Unlawful; Prevention by Commission The FTC has used this broader power to investigate algorithmic pricing practices, including a January 2025 surveillance pricing study on individualized pricing and a $2.5 billion settlement against Amazon in September 2025.
The RealPage Benchmark
The DOJ’s case against RealPage is the most significant algorithmic collusion action to date, and it defines the current enforcement theory. RealPage’s revenue management software set rental prices for competing landlords across the country. The DOJ alleged the software relied on nonpublic, competitively sensitive information shared by landlords to generate rent recommendations, included features designed to limit price decreases and align pricing among competitors, and that RealPage hosted meetings where competing property management companies shared sensitive information.3U.S. Department of Justice. Justice Department Requires RealPage to End the Sharing of Competitively Sensitive Information
The proposed settlement, published in the Federal Register in January 2026, requires RealPage to stop using competitors’ nonpublic data in pricing recommendations at runtime, limit model training to historical data aged at least 12 months, remove features that constrained price decreases or aligned pricing between competitors, cease conducting market surveys that collected sensitive competitive information, and accept a court-appointed monitor.4Federal Register. United States of America et al. v. RealPage, Inc. et al. Proposed Final Judgment and Competitive Impact Statement
The case matters well beyond rental housing. Its core principle is that pricing recommendations become akin to price-fixing when combined with nonpublic competitor data and mechanisms that pressure compliance. Any industry using shared pricing software is in the enforcement zone the case defines.
The Gap for Autonomous Algorithms
When machine-learning systems independently converge on elevated prices without shared data, shared software, or human communication, existing antitrust law struggles to reach the conduct. The Sherman Act was written to address human agreements. If no person at any company intended for collusion to occur, and no data was exchanged between competitors, identifying the “agreement” that Section 1 requires becomes extremely difficult.
Proposed responses include broadening the legal definition of “agreement” to cover algorithmic outcomes that mimic collusion, creating a rebuttable presumption of collusion when software consistently produces above-competitive prices, and requiring companies to test pricing algorithms in regulatory sandboxes before deployment. The DOJ and FTC launched a joint public inquiry in February 2026 to develop new guidance on algorithmic pricing, information exchanges, and artificial intelligence. As of now, the agencies have not issued replacement guidelines for the collaboration framework they withdrew in December 2024.
What Consumers and Businesses Harmed by It Can Do
Algorithmic collusion is not solely a matter for government enforcers. Anyone injured in their business or property by an antitrust violation can sue in federal court and recover three times their actual damages, plus attorney fees.5Office of the Law Revision Counsel. 15 USC 15 – Suits by Persons Injured Class actions alleging algorithmic price-fixing are proliferating.
Clearing the pleading stage is the practical hurdle. In the Yardi litigation, a court denied a motion to dismiss based on allegations that landlord defendants understood the software provider would use commercially sensitive information to maximize rents industry-wide, supported by plus factors such as marketing materials and internal documents acknowledging the tool’s price-increasing effect. In a case against SAS Institute, a different court granted dismissal, finding that allegations about the software using customer data were too thin without specifics on how that data produced the pricing recommendations. Plaintiffs generally need to allege that the algorithm pooled nonpublic competitor data, not just that competitors happened to use the same software.
Standing rules narrow the field further. Under federal antitrust law, only direct purchasers can sue for damages. Roughly three-quarters of states have passed laws allowing indirect purchasers, meaning end consumers, to bring their own price-fixing claims under state antitrust statutes, with remedies varying from mandatory treble damages to caps at actual losses.
State transparency laws are adding another layer. New York now requires conspicuous disclosure when prices are set by an algorithm using an individual consumer’s data. Illinois has proposed similar legislation requiring sellers to notify consumers when a price was set by an algorithm using their personal data, though the Illinois bill would not take effect until 2028.
How Investigators Prove It
Proving algorithmic coordination combines economic and technical evidence. Economists look for “plus factors,” which are market conditions that would be irrational absent collusion: prices moving in lockstep across competitors, refusals to lower prices when input costs drop, uniform pricing in markets where variation would be expected. Large-scale data analysis surfaces these patterns across thousands of transactions.
The technical side is often more revealing. The algorithm’s source code shows what the software was actually designed to do. Forensic experts examine whether the algorithm was built to prioritize industry-wide price stability over independent competition, whether it ingested nonpublic competitor data, and whether it included mechanisms to discourage price cuts. Internal communications and metadata can show whether executives intended for the software to shadow competitor pricing, even when the code itself looks neutral.
Compliance for Businesses Using Pricing Software
Companies that use third-party pricing tools carry real antitrust exposure, and “we didn’t know the algorithm was doing that” is not a defense enforcers accept. The RealPage settlement provides the working template for minimum safeguards. Businesses using automated pricing should map their entire data pipeline: what data goes in, how models are trained, and how outputs influence pricing decisions.
Vendor contracts need specific protections. Based on the conduct the DOJ targeted and the terms of the proposed RealPage settlement, the practical steps include:
- Specify which data sources the vendor can use, and prohibit nonpublic competitor data from generating price recommendations.
- Require that any nonpublic data used for model training is at least 12 months old, consistent with the RealPage settlement terms.3U.S. Department of Justice. Justice Department Requires RealPage to End the Sharing of Competitively Sensitive Information
- Limit geographic granularity. The RealPage settlement prohibits models that determine pricing effects at a level narrower than statewide.3U.S. Department of Justice. Justice Department Requires RealPage to End the Sharing of Competitively Sensitive Information
- Confirm the software does not include mechanisms that constrain price decreases or align pricing across competitors.
- Retain contractual rights to audit and monitor the vendor’s algorithm, including verifying that safeguards work as designed.
Ongoing monitoring matters as much as initial diligence. Companies should track whether their algorithmic outputs are converging with rivals’ prices, particularly in concentrated markets with few competitors. If the pattern looks like coordination, it will look that way to regulators too. There is no safe harbor and no current formal compliance guidance for algorithmic pricing, since the DOJ and FTC withdrew their previous collaboration guidelines in December 2024 and have not yet issued replacements. That vacuum makes proactive self-policing more important, not less.