Introduction

The increasing availability of highly sophisticated data regarding market conditions, consumer preferences, etc. coupled with advances in machine-learning has transformed the strategies adopted by businesses for optimising profits by adjusting prices. Pricing Algorithms can constantly track rival prices, and accordingly adjust own prices and adapt to market dynamics with frequency beyond human capabilities. The increased reliance by market players on such algorithms is evident from the findings of the European Commission’s (EC) 2017 E-Commerce Sector Enquiry, when approximately one-third of the respondent retailers reported their usage in decision-making. 

While algorithmic pricing increases the efficiency of dynamic pricing, it helps create fertile conditions for tacitly coordinated outcomes and supracompetitive prices, which has been traditionally beyond the scope of antitrust enforcement. Traditional enforcement laws do not consider mere parallelism or tacit collusion as illegal in the absence of plus-factors (circumstantial economic evidence) signifying human involvement in reaching a horizontal agreement for collusion. However, the capabilities of algorithms to continuously observe market conditions in a highly transparent environment, instantly retaliate against decreases in prices charged by rivals and reinforcement learning facilitate collusive outcomes without human involvement. Recent developments confirms that concerns regarding the adequacy of traditional enforcement measures in situations where collusive outcomes emerge without human communication proving the existence of an agreement are not speculative or “an old wine in a new bottle.” The EC has initiated sectoral enquiries into algorithmic pricing practices sustaining collusion, while California and New York amended their state antitrust statutes to bar the usage of common pricing algorithms by competing businesses to coordinate parallelism.

This paper examines how algorithmic pricing alters the landscape of collusion, and offers potential liability standards and enforcement tools when competition harm occurs without human intervention. The paper is structured into four sections. The first section explores the potential challenges posed by algorithms facilitating tacit collusion. The second section examines the global jurisprudence on algorithmic collusion. The third section analyses India’s approach in cases where collusion is the alleged outcome of pricing algorithms. The last section provides a framework for attributing liability in cases involving autonomous algorithms and proposes enforcement strategies. The paper concludes by arguing for a foreseeability-based responsibility standard to impose liability on human users where collusive effects are achieved by autonomous agents. 

The Differential Degrees of Anti-Competitive Effects of Pricing Algorithms 

In concentrated oligopolistic markets with homogenous products, substitution of competition for cooperation is the dominant strategy for profit maximisation. Such cooperation can be sustained by (a) reaching an agreement regarding trade conditions such as price, quantity or quality; (b) detecting deviations from the agreed upon levels; and (c) threat of credible and effective retaliation to disincentivise deviations. The use of algorithms to determine individual prices by firms facilitates the enforcement of these conditions and stabilisation of prices at supracompetitive levels by decreasing the incentives to deviate. Rival prices in transparent digital markets can be compared in real-time using algorithms. Any reduction in price to incentivise consumers and increase sales can be met with retaliation by competitors in milliseconds, not giving consumers an opportunity to track the original price decrease and switch to the competitor.  

Section 3 of the Competition Act, 2002 bars any agreement that limits or restrains competition in the market. Section 2(h) defines any agreement as an arrangement, understanding or action in concert and therefore includes both implicit and explicit agreements. Article 101 TFEU bars any explicit agreement or concerted action that restrains competition by object or effect. However, tacit collusion or mere price parallelism can be attributed to rational reaction by competitors to market structure and dynamics and is not illegal under these provisions due to absence of proof regarding human involvement in reaching an agreement to restrict competition. While imposition of liability is relatively straightforward where algorithms merely implement a pre-existing collusive arrangement, the deployment of autonomous agents capable to make pricing decisions on their own through reinforcement learning may escape legal liability even when it results in competitive harm. Ezrachi and Stucke have developed a four-category conceptual framework for designing enforcement strategies with varying levels of human involvement in algorithmic pricing decisions. These are (a) messenger, (b) hub-and-scope, (c) predictable agent and (d) digital eye. The latter two involve deployment of advanced algorithms which facilitate price coordination with minimal human involvement and communication. These scenarios suggest the need to shift to a foreseeability or effect-based standard for imposing liability on actors who had awareness regarding the potential anti-competitive effects or could reasonably foresee them.

Where We Are: A Global Perspective on Algorithmic Price Collusion

An example of the messenger category is US v. David Topkins, where poster sellers on Amazon marketplace were held liable for horizontal price-fixing by developing an algorithm to give effect to their prior agreement not to undercut each other. Similarly, in Trod/GB Eye, poster and frame sellers were found guilty of online price fixing through algorithms to implement their agreement not to compete on prices. 

In E-Turas, the Court of Justice of the European Union (CJEU) held that the technical restrictions imposed by the online booking platform on travel agents. limiting their ability to offer discounts at the maximum rate of 3% was a hub-and-spoke cartel. The travel agents were held liable based on their continued participation on the platform despite knowledge of the anti-competitive restriction on discount offers. It was clarified that unless members publicly distanced themselves from such arrangement by refusing to participate and disclose their presence to the public and regulator agencies, their participation would be presumed for determining liability. In Proptech, Spanish real estate brokers were held liable for implementing a cartel arrangement by using an automated algorithmic software to restrict the listing of properties on the database without complying with the minimum commission rate of 4%. Further, algorithmic pricing has facilitated the enforcement of impermissible resale price maintenance agreements which restrict competition by effect. In 2018, the European Commission (EC) held consumer electronic firms Asus, Denon & Marantaz, Philips and Pioneer liable for monitoring their retailers’ compliance with the pre-agreed minimum resale price and imposed penalties on them.

In its December 2024 ruling in Duffy v. Yardi, the Washington District Court held that the reliance of landlords on a common software company to recommend rent-prices based on non-public competitive data signifies a horizontal anti-competitive agreement. This is a significant ruling which clarifies that price suggestions based on aggregation of non-competitive and commercially sensitive data will be treated as horizontal agreement where the anti-competitive effects will be presumed by law and need not be demonstrated by the plaintiff. Hence, liability can be determined as per the traditional regime because human involvement is clearly established in such cases. 

India’s Approach to Algorithmic Price Collusion

India’s approach to algorithmic collusion has witnessed an evolution from a stage of missed opportunity to acknowledge the potential anti-competitive effects to a stage of active engagement with emerging challenges and attempt at their resolution. In Samir Aggarwal, the Competition Commission of India (CCI), the National Company Law (Appellate) Tribunal (NCLAT) and the Supreme Court (SC) rejected the allegations that cab aggregator platforms such as Ola and Uber had served as ‘hubs’ in the traditional sense of a hub-and-spoke cartel. There was no evidence submitted by the informant to demonstrate that a horizontal agreement existed between the drivers to set prices by using the platforms. Neither did the arrangement fulfil the criteria of a vertical agreement because cab aggregators were not part of the same supply chain as the drivers and did not sell any input to them as upstream service providers. The possibility of expanding the scope of enquiry into the business model adopted by cab aggregators wherein they profited through higher commission from collectively higher prices charged by the drivers was not explored. 

The 2019 Report of the Competition Law Review Committee (CLRC) recommended amending Section 3(3) of the Competition Act to explicitly include such hub-and-spoke arrangements as a category which would give rise to the rebuttable presumption of their anti-competitive nature. Further, the Committee recommended amending Section 3(4) of the Act to include anti-competitive agreements which are not strictly vertical in nature to prohibit other agreements restraining competition by object. The Act was amended accordingly in 2023. However, the Committee did not suggest any amendment to address algorithmic collusion as a separate category. According to the Committee, the agreement-based framework under Section 3 is sufficient to deal with cases related to tacit collusion facilitated by algorithms. 

In Alleged Cartelisation in the Airlines Industry, the CCI held that the use of similar algorithms by the airlines did not result in collusion. The algorithms were fed with historic data sets regarding prices charged, consumer demand, cost, etc. which were unique for each airline, hence the recommended prices were not based on identical data sets. Further, the prices were modified by route analysts who considered factors such as festivals, cultural and political events, weather variations. The revenue management team determined the final prices showed to consumers while the algorithms merely facilitated the manual decision-making process. Additionally, no exchange of communication between the airlines could be established to prove a collusive arrangement and price parallelism was attributed to intelligent adaptation to the prevailing market conditions, which is beyond the purview of antitrust law. This stance was further reiterated in Shikha Roy, where the CCI held that an anti-competitive agreement between the airlines to increase airfare due to the Jat Agitation could not be established due to the absence of any communication or circumstantial evidence in this regard. 

The CCI’s insistence on providing evidence of communication for proving anti-competitive agreements puts a high burden on informants in Predictable Agent and Digital Eye scenarios where such evidence is not usually available. The CCI has recognised the potential fallbacks of this approach in its 2025 CCI Market Study on Artificial Intelligence and Competition. It has recommended the development of suitable technical tools for auditing algorithms, improving transparency, and building regulatory capacity to detect algorithmic collusion and determine liability.

Potential Roads To Be Taken: Enforcement Strategies Against Algorithmic Collusion

The decisional practice demonstrates that the current enforcement regime is sufficient to tackle cases where algorithms belonging to the first and second category are deployed to facilitate collusion among rivals. Challenges arise in the cases of Predictable Agent and Digital Eye as the firms develop their own algorithm independently, without evidence for any communication or agreement for cartelisation. This section proposes an ex-post liability based on the reasonable foreseeability of anti-competitive effects of using autonomous agents for pricing decisions. Further, enforcement agencies can use their own algorithms to audit pricing algorithms and detect the presence of any collusive properties before they are deployed in the market, enhance transparency obligations for firms and impose restrictions of the frequency of price adjustments are potential mechanisms to facilitate ex-ante regulation.

As evident from the global jurisprudence, ex-post regulation is the most common method to examine the collusive impact of algorithmic pricing and attribute liability. In the absence of traditional plus factors to prove anti-competitive agreement where price parallelism is effectuated by independently designed algorithms, informational requests regarding the objective and functioning of the algorithm can assist in determining liability. Section 41(8) of the Competition Act, introduced by the 2023 amendment empowers the Director-General (DG) to conduct duly authorised search-and-seizure operations against defendants for investigating the allegations of anti-competitive conduct. This empowers the collection of digital evidence, which can include the source code for any pricing algorithm. The source code contains information regarding the various considerations and their proportionate weightage that it has been programmed to account for while taking pricing decisions. In Predictable Agent scenarios, such codes and their behavioural responses can be assessed through regulatory sandboxing and simulation-based testing for the presence of any collusive property. A regulatory sandbox refers to a controlled experimental environment where pricing algorithms are deployed alongside simulated market participants for observing their behavioural responses to different market conditions. Regulators manipulate parameters such as cost structures, demand, pricing strategies of rival firms to examine whether the subject-algorithm exhibits behaviour consistent with tacit collusion. For example, the decision not to decrease price even when the player has achieved cost reduction to prevent an anticipated reduction in price by rivals and may signify the presence of properties facilitating tacit collusion despite the absence  of an agreement. 

Further information such as the instructions provided by the user-defendant(s) to the algorithm developers regarding their desired functional requirements can be sought. Liability can be easily determined for a single player when clear anti-competitive intent to design and use the algorithm for charging supracompetitive prices and stabilising markets is present. Internal documents, communication between the designers and deploying teams, etc. serve as sufficient evidence for proving unfair practices with potential to distort markets, as held in the context of security laws. A similar standard can be applied in the context of antitrust enforcement. 

In cases involving autonomous pricing agents which independently determine collusion as the dominant strategy from their deep learning capacities and experiences in game theory, liability can be attributed on those who had knowledge and awareness about the potential collusive effect or could reasonably foresee the same. They can be held liable even in the absence of any communication or agreement in furtherance of collusion with rivals when they had actual or constructive knowledge of the anti-competitive consequences and did not take any steps to prevent or mitigate them. The EC has recommended a risk-management approach where liability is assigned on the actor best suited to avoid the materialisation of risks posed by autonomous agent.

The dynamic nature of digital markets has led to the increased adoption of ex-ante regulatory measures for promoting deterrence of conduct that can potentially disrupt competition. The EU Digital Markets Act (DMA) and the Indian draft Digital Competition Bill provide for such ex-ante regulation in cases where dominant players may abuse their market power and distort competition, harming consumer welfare. Hence, ex-ante measures as provided below can be used even before a case of algorithmic collusion is reported to enhance deterrence. Transparency obligations can be imposed on businesses forbidding the usage of commercially sensitive data to train algorithms to reach collusive outcomes. Further, guidelines lowering the permissible speed and frequency with which price adjustments among competitors can be effectuated increase the incentives for price competition among algorithms, even if it might result in certain inefficiencies. Any autonomous pricing algorithm maybe tested in a sand box before its adoption in the market for testing its response to fluctuations in relevant factors such as demand, cost, market price, etc. Even though sand boxing does not reflect real market dynamics or the operation of algorithm based on deep learning from its operation in the market, it serves as a starting point in the enquiry for properties which should be prohibited. Further, the competition authorities can use their own algorithms to identify algorithmic collusion based on data analytics and destablise the supracompetitive equilibrium by mixed signalling for business algorithms. 

Conclusion

According to Maureen Ohlhausen, then Acting Chair of the U.S. Federal Trade Commission, an algorithm should be treated comparably to “a guy named Bob” when it attempts to engage in anti-competitive conduct. Autonomous pricing agents may give rise to a complex scenario when the algorithm learns collusion as the best strategy to increase profitability based on its own experience rather than human instructions. The current principles should be reinterpreted to include scenarios where collusion is the outcome of machine learning without evidence or intention of human agreement. As EU Commissioner Margaret Vestager has commented, businesses using algorithmic pricing systems for giving effect to collusion cannot be given ‘immunity’ because of the autonomous nature of reinforcement learning. Hence, a liability regime imposing liability on actors based on their knowledge regarding potential anti-competitive nature of such agreements or their continuance with such arrangements despite the reasonable foreseeability of the negative consequences should be developed to penalise current distortion of competition in the market and enhance future deterrence.

Notes & References

1. Commission, ‘Final Report on the E-Commerce Sector Inquiry’ COM (2017) 229 final.

2. Thibault Schrepel, ‘The Fundamental Unimportance of Algorithmic Collusion for Antitrust Law’ (2020) Harvard Journal of Law & Technology <https://jolt.law.harvard.edu/digest/the-fundamental-unimportance-ofalgorithmic-collusion-for-antitrust-law> accessed 3 December 2025.

3. Rashid Baxter, “EU reveals existence of algorithmic pricing cases’ (Global Competition Law, 9 July 2025) <https://globalcompetitionreview.com/article/eu-reveals-existence-of-algorithmic-pricing-cases> accessed 6 December 2025.

4. A.B. 325, 2025-2026 Gen. Assemb., 2025 Cal. Stat., ch. 338 (Cal. 2025) <https://legiscan.com/CA/text/AB325/id/3272268> accessed 2 December 2025.

5. Rega Justin, ‘NY Algorithmic Rent Pricing Ban Goes Into Effect Under New Law’  Bloomberg Government News <https://news.bgov.com/bloomberg-government-news/ny-algorithmic-rent-pricing-ban-goes-into-effect-under-new-law>.

6. Paul L. Joskow, ‘Firm Decision-making Processes and Oligopoly Theory’ (1975) 65(2) The Americal Economic Review <https://www.jstor.org/stable/1818864> accessed 2 December 2025.

7. George J. Stigler, ‘A Theory of Oligopoly’ (1964) 72 Journal of Political Economy 44.

8. Michal S. Gal, ‘Algorithms as Illegal Agreements’ (2019) 34(1) Berkeley Technology Law Journal <https://www.jstor.org/stable/10.2307/26755226> accessed 2 December 2025.

9. The Competition Act, 2002, §3(1).

10. The Competition Act, 2002, §2(h).

11. Consolidated Version of the Treaty on European Union [2008] OJ C115/13, Art 101.

12. Anik Bhaduri, ‘Tackling Collusion In The Digital Marketplace: Is The Competition Act Enough?’ (2020) 41(2) European Competition Law Review 99 <https://ssrn.com/abstract=3749201> accessed 3 December 2025.

13. Ariel Ezrachi & Maurice E. Stucke, ‘Algorithmic Collusion: Problems and Counter’ (OECD 2017) <https://one.oecd.org/document/DAF/COMP/WD(2017)25/en/pdf>. 

14. Ariel Ezrachi & Maurice E. Stucke, ‘Artificial Intelligence & Collusion: When Computers Inhibit Competition’ (2017) University of Illinois Law Review <https://www.illinoislawreview.org/wp-content/uploads/2017/10/Ezrachi-Stucke.pdf >.

15. United States v Topkins (ND Cal 2015) Crim Case No 15-cr-00201.

16. Online sales of posters and frames (2016) Case 50223, Decision of the CMA.

17. Case C-74/14 Eturas v Lietuvos Respublikos konkurencijos taryba,(2016) ECLI:EU:C:2016:42.

18. Case S/0003/20 Proptech (2021) <https://www.cnmc.es/novedad/sancionador-proptech-cnmc-intermediacion-inmobiliaria-cnmc-20211209> accessed 4 December 2025.

19. ‘Antitrust: Commission fines four consumer electronics manufacturers for fixing online resale prices’ European Commission (Brussels, 24 July 2018) <https://ec.europa.eu/commission/presscorner/detail/it/ip_18_4601> accessed 6 December 2025.

20. Duffy v. Yardi Systems Inc et al, No. 2:2023cv01391 - Document 187 (W.D. Wash. 2024).

21. Samir Agrawal v CCI (Cab Aggregators Case), (2021) 3 SCC 136.

22. Nikita Koradia, et al., ‘Algorithmic Collusion and Indian Competition Act: Suggestions to Tackle Inadequacies and Naivety’ in Steven Van Uytsel   (ed), The Digital Economy and Competition Law in Asia (Springer 2021).

23. Report of the Competition Law Review Committee (2019) <https://www.ies.gov.in/pdfs/Report-Competition-CLRC.pdf> accessed 5 December 2025.

24. The Competition Amendment Act, 2023.

25. Supra note 24.

26. Alleged Cartelization in the Airlines Industry, In re, 2021 SCC OnLine CCI 3.

27. Shikha Roy v Jet Airways (India) Limited and Others, CCI, Case No. 32 of 2016 (3 June 2021).

28. Competition Commission of India, Market Study on Artificial Intelligence and Competition (2025) <https://www.cci.gov.in/images/marketstudie/en/market-study-on-artificial-intelligence-and-competition1759752172.pdf> accessed 5 December 2025.

29. Francisco Beneke & Mark-Oliver Mackenrodt, ‘Remedies for Algorithmic Tacit Collusion’ (2020) 9(1) Journal of Antitrust Enforcement <https://doi.org/10.1093/jaenfo/jnaa040>  accessed 4 December 2025.

30. The Competition Act, 2002, §3(1).

31. Jospeh Harrington, ‘Developing Competition Law For Collusion By Autonomous Artificial Agents’ (2019) 14(3) Journal of Competition Law & Economics <https://academic.oup.com/jcle/article-abstract/14/3/331/5292366> accessed 3 December 2025.

32. Supra note 29.

33. In re Athena Capital Res., LLC, 3950 S.E.C. No. 3-16199 (2014).

34. Supra note 13.

35. Commission, ‘Commission Staff Working Document on the free flow of data and emerging issues of the European data economy Brussels’ COM (2017) 9 final.

36. Council Regulation 2022/1925 of 14 September 2022 on contestable and fair markets in the digital sector and amending Directives (EU) 2019/1937 and (EU) 2020/1828 (Digital Markets Act) [2022] OJ L 265.

37. The Draft Digital Competition Bill, 2024 <https://prsindia.org/policy/report-summaries/digital-competition-law> accessed 6 December 2025.

38. Supra note 29.

39.  Supra note 12.

40. Ibid.

41. Maureen K. Ohlhausen, ‘Should We Fear The Things That Go Beep In the Night? Some Initial Thoughts on the Intersection of Antitrust Law and Algorithmic Pricing’ United States Federal Trade Commission (23 May 2017) <https://www.ftc.gov/system/files/documents/public_statements/1220893/ohlhausen_-_concurrences_5-23-17.pdf> accessed 5 December 2025.

42. Foo Yun Chee, ‘EU's Vestager warns companies against abusing algorithms’ Reuters (16 March 2017) <https://www.reuters.com/article/business/eu-s-vestager-warns-companies-against-abusing-algorithms-idUSKBN16N18W/> accessed 5 December 2025.

About the Author

Andaleeb Haider is a Third Year Student at National Academy of Legal Studies and Research (NALSAR), Hyderabad