Definition
Algorithmic collusion refers to competing enterprises achieving coordinated, anti-competitive outcomes using algorithm-based software, resulting in price, output, or other competitive parameters being raised or maintained beyond levels that would have otherwise prevailed under effective competition.
Such coordination may be explicit, i.e., the use of algorithms to support conscious and voluntary coordination. The algorithm here acts as a facilitator or enabler of human collusion. Or such coordination may be implicit, i.e., where independently deployed algorithms autonomously converge on anti-competitive strategies without any express agreement or direct communication. The collusion can occur without any human involvement or reciprocal interaction and is based on conscious parallelism of behaviors by competing undertakings.
The increasing use of algorithms enhances market transparency, accelerates price adjustments, and optimizes responses to competitors’ conduct. These features, while efficiency-enhancing in certain contexts, may also create market conditions conducive to collusion.
Commentary
The use of algorithms to determine prices or outputs is not a recent phenomenon, and such tools have long been deployed across industries, including aviation, hospitality, and retail. The Competition Commission of India (“CCI”) has previously examined allegations of cartelization under Sections 3(1) and 3(3) of the Competition Act, 2002 (“Act”) in markets characterized by algorithmic pricing, notably in the domestic airline industry and in the cab aggregator platform market.
Approach to algorithmic pricing under the Competition Act, 2002
In Re: Alleged Cartelization in the Airlines Industry (“Airline Cartelization Case”),1 The CCI has noted the Director General's observations that technological advancements have intensified competition, as airlines can now monitor competitors’ fares in real time through publicly available sources. In such a dynamic and demand-responsive market, price parallelism was found to be a natural and foreseeable outcome, particularly during peak and lean seasons, and was not construed as evidence of agreement.
A central factor influencing the CCI’s analysis in airline pricing has been the role of human intervention in the pricing process. In Airline Cartelization Case, the CCI observed that revenue management personnel play a pivotal role in the determination of airfares, with software tools merely facilitating decision-making. The revenue management teams provide parameters to software developers, and the algorithms embedded in the revenue management software are designed and programmed by third-party vendors on the basis of airline-specific inputs. The output generated by such software is further analyzed by route analysts, who take the final decision regarding inventory allocation and pricing before uploading fares. Accordingly, the role of the software is limited to aiding the revenue management team. This finding was reiterated in Shikha Roy v. Jet Airways (India) Limited2, where the CCI emphasised that the ultimate call on inventory allocation is taken by the respective route analysts of each airline, thereby preserving decisional independence. Even where similar revenue management software was used by multiple airlines, manual intervention was found to remain decisive in determining final prices.3
A similar approach was adopted by the CCI in Samir Agrawal v. ANI Technologies Pvt. Ltd.4, which involved allegations of algorithmic collusion in the cab aggregator market. The informant argued that platform-determined pricing facilitated collusion among drivers and amounted to a hub-and-spoke cartel. The CCI rejected this argument, holding that a hub-and-spoke arrangement requires a prior agreement or coordination among the drivers themselves, facilitated by the platform through the exchange of price-sensitive information. In the absence of such coordination, the unilateral setting of prices by a platform’s algorithm, and their acceptance by drivers, does not constitute an agreement or concerted practice under Section 3 of the Act.
Typologies of algorithmic collusion
The CCI, in its market study report on “Artificial Intelligence and Competition” (“AI Market Study”) has classified algorithmic collusion concerns arising from algorithms as monitoring algorithms, parallel algorithms, signalling algorithms, and self-learning algorithms.
Monitoring Algorithms implement or monitor human-colluded agreements.5 It facilitates collusion by collecting information concerning competitors’ business decisions. Such algorithms require explicit collusion between human actors for them to result in market distortion.
A clear illustration of this category was United States v. Topkins6 the defendant, where David Topkins, sold posters on Amazon Marketplace, entered into an arrangement with a competing seller to fix and maintain prices. The algorithm used by the defendants gathered pricing data to determine the lowest price available in the market. Pursuant to this arrangement, the colluding sellers set their prices marginally below the prevailing market price, resulting in the suppression of genuine competition.
Parallel Algorithms or Hub and Spoke Collusion, where the competitors may use the same algorithm or algorithm provider to facilitate the indirect exchange of competitively sensitive information. The use of a hub may extend to the generation of pricing recommendations based on co-mingled data. In such cases, indirect information exchange through a third party, including an algorithm provider, may constitute anti-competitive conduct. While the use of such algorithms does not, by itself, amount to collusion, concerns arise where the hub collects data from all spokes and influences overall pricing in the market.
These concerns were reflected in the litigation concerning RealPage’s revenue management software7, which recommended rental prices to multi-family apartment owners based on non-public data supplied by competing users. Now a settlement has been reached between the US Department of Justice and RealPage Inc., where RealPage has agreed to ensure that its algorithms will not use competitors’ non-public, competitively sensitive information to determine rental prices in runtime operation.8
In a factually similar matter, the U.S. District Court for the Western District of Washington refused to dismiss antitrust claims against a group of multifamily housing operators alleged to have violated the Sherman Act.9 The court held that the pleadings sufficiently suggested that the defendants knowingly shared commercially sensitive information with a pricing software provider and adopted the pricing output of the software. According to the court, the defendants entered into a collective restraint by enrolling in RENTmaximizer with the awareness that the scheme would yield benefits only if rival landlords also participated.
The courts have, however, also emphasized limits on liability in algorithmic hub-and-spoke cases. In Gibson v. Cendyn Group, LLC10, claims alleging algorithmic collusion among Las Vegas hotel operators were dismissed where pricing recommendations were non-binding, and hotel operators retained full pricing discretion. The absence of allegations regarding the pooling or commingling of confidential competitor data was decisive. A similar conclusion was reached in Cornish-Adebiyi v. Caesars Entertainment, Inc.,11 in the absence of any pooled dataset against which the algorithm operated. These decisions suggest that the exchange or pooling of competitively sensitive information and the degree of pricing discretion retained by users are central to the analysis.
Signalling algorithms react predictably to market events and may increase the risk of tacit coordination by reducing strategic uncertainty. While price transparency alone is not anti-competitive, algorithms that rapidly detect and respond to rivals’ pricing changes may stabilize prices and discourage aggressive competition. Where such algorithms operate independently and without coordination, they may not raise competition concerns.
Self-learning algorithms are capable of maximizing profits through reinforcement learning and rapid adaptation to market conditions. Although such algorithms may not be designed to achieve collusion, they may nonetheless reach collusive outcomes depending on market structure and learning objectives.
Conclusion
Across these cases, a clear theme emerges that the use of algorithms does not shield firms from antitrust liability. Competition authorities have consistently held undertakings accountable for the conduct of their automated systems, rendering the technological mechanism used to implement price-fixing legally immaterial. The decisional practice of the Competition Commission of India also reflects a technology-neutral application of competition law, a position reinforced by the Competition (Amendment) Act, 2023, which expressly recognized hub-and-spoke cartels, an issue of particular relevance in AI-driven markets. The current framework, therefore, appears capable of addressing algorithmic collusion in certain cases.
However, difficulty arises in cases of purely automated conduct, where there is a risk of penalizing lawful parallel behaviour. The Supreme Court has recognized conscious parallelism as a rational response to the market conditions and economic factors12 Where algorithmic decision-making is opaque even to its creators, identifying the plus factors necessary to establish collusion becomes more difficult. How competition law will adapt to this challenge remains an open and evolving question.
Suo Motu Case No. 03 of 2015↩︎
Case No. 32 of 2016↩︎
Supra n.1, Para 51.↩︎
2018 SCC OnLine CCI 86↩︎
Renato Nazzini & James Henderson, Overcoming the Current Knowledge Gap of Algorithmic “Collusion” and the Role of Computational Antitrust, Stanford Computational Antitrust Project (Feb. 2024) https://law.stanford.edu/wp-content/uploads/2024/02/Algorithmic-Collusion.pdf↩︎
Case 3:15-cr000201-WHO (N.D. Cal. 2015)↩︎
In re Realpage, Inc., Rental Software Anti. Litig., 709 F.Supp.3d 478 (M.D. Tenn. 2023)↩︎
United States v. RealPage, Inc., No. 1:24-cv-00710-WLO-JLW (M.D.N.C. Nov. 24, 2025).↩︎
Duffy v. Yardi Sys., Inc., No. 2:23-cv-01391-RSL (W.D. Wash. Dec. 4, 2024),↩︎
2024 WL 2060260 (D. Nev. May 8, 2024).↩︎
2024 WL 435618 (D.N.J. Sept. 30, 2024).↩︎
Rajasthan Cylinders & Containers Ltd. v. Union of India, (2020) 16 SCC 615↩︎
