Silicon Cartels and Digital Dominance: Competition Law in the Age of AI Pricing

- by Adveer Singh Narang & Aayushka Pandey, students at Hidayatullah National Law University, Raipur. This is the 4th best entry in the National Article Writing Competition organized by CBLT.

Introduction

Artificial Intelligence (AI) and algorithmic technologies are increasingly transforming how firms operate, especially in the realm of pricing. AI pricing refers to the use of algorithms, including machine learning and data-driven models, to set and adjust prices dynamically in response to real-time variables such as consumer behaviour, market demand, competitor prices, and historical trends. As digitalisation deepens across sectors,[1] AI-driven pricing is no longer confined to e-commerce; it is now prevalent in industries such as insurance and credit,[2] finance, travel, and hospitality.[3] 

Unlike traditional pricing strategies, AI algorithms are capable of operating autonomously, processing vast volumes of datasets in real-time and adjusting continuously from market inputs. This has led to the emergence of several competition-related concerns. One of the most debated is algorithmic collusion, where autonomous pricing algorithms, even without direct coordination or human intervention, may end up converging on supra-competitive pricing strategies.[4] 

However, the risks are not limited to collusion alone. AI pricing tools can also be used to engage in unilateral conduct such as self-preferencing, dynamic price discrimination, predatory pricing and hub-and-spoke arrangements.[5] These scenarios blur the traditional boundaries of antitrust enforcement, where intent and explicit agreements have historically been central to establishing liability.

This article examines AI-driven pricing through a legal lens, assessing the dichotomy between its potential to enhance efficiency and its ability to facilitate anti-competitive conduct. Further, this article delves into the existing anti-trust framework while also proposing possible solutions to mitigate the distortion of competition at the hands of artificial intelligence.

AI pricing: Catalysing Competition

An ethical AI levels the field; not tips the scale.

This shift towards algorithmic pricing has brought about significant pro-competitive benefits by increasing efficiency both on the demand and supply side.[6]AI tools have significantly improved and advanced the insurance industry, AI tools have made great strides in industries such as  insurance , with AI-driven platforms such as Lemonade based on a transparent fee model that streamlines insurance claims.[7] Furthermore, AI pricing also enables market entry and innovation by enabling new entrants to use available data to provide data-based prices of products and services, thus increasing consumer choice and creating a virtuous cycle of innovation.

For the consumer, the introduction of AI has increased efficiencies in the demand side by providing real-time price information and reducing search and transaction costs. Additionally, AI tools have also created greater bargaining power by making available price comparison tools such as Skyscanner, which enhance transparency and empower consumers by aggregating data from multiple sources in real time, allowing users to compare prices, routes, and services with ease.[8] By reducing information asymmetry, they enable consumers to make more informed decisions. Furthermore, the introduction of AI goes beyond just increasing efficiencies at the supply and demand sides; it also helps balance supply and demand through dynamic pricing models, maintaining market equilibrium. Algorithms also possess the ability to expand the competitive parameters by assessing factors such as product quality, customer service ratings, delivery timelines, and sustainability which were hitherto not accounted for.[9]

However, it is pertinent to note that AI pricing is a double-edged sword with increased efficiency on one side and possible anti-competitive effects on the other. Alongside the advantages that AI inherently possesses, the dilemma that AI leads to the sophistication of anti-competitive agreements and abuse of dominance poses complex challenges for market regulation and competition law. Therefore, AI pricing must be benchmarked against transparency and ethical pricing mandates.

The Price of Progress: AI and Market Manipulation

a. Anti-Competitive Agreements

Collusion is not always born in smoke-filled rooms. Sometimes, it takes the shape of parallel algorithms speaking in silence.

Algorithmic pricing tools challenge traditional conceptions of anti-competitive agreements under Section 3 of the Competition Act. 2002 (“Act”).[10]  The provision has traditionally been applied to explicit collusion where there is communication or a “meeting of minds, often evidenced through exchange of information or coordinated pricing. In contrast, tacit collusion refers to parallel conduct without any formal agreements or human communication, where independent algorithms align pricing to mirror unlawful collusion. This is especially evident in digital markets, where transparency and data sharing enable mutual price adjustments.[11] Empirical studies have demonstrated that such algorithms can avoid competitive pricing and stabilize supra-competitive results, mirroring cartel-like behaviour.[12]

A particularly concerning type of algorithmic collusion arises through hub-and-spoke arrangements. In this, several competing firms (the spokes) shared the same third-party pricing software (the hub) which applies uniform pricing logic. Even without any direct coordination, this can result in algorithmic collusion creating outcomes similar to cartel behaviour.[13] In the Shikha Roy case, the CCI acknowledged the possibility of such collusion without human involvement but did not find enough evidence to conclude contravention.[14] Nevertheless, such arrangements can give rise to “constructive agreements” where the algorithm acts as a tool. The economic profit is mutual where firms comply with the platform’s pricing to avoid retaliatory behaviour or exclusion from the service.[15]

Moreover, algorithms can allow companies to fluctuate prices by quickly detecting any price cuts and responding with retaliation. This creates a feedback loop that keeps prices changing. In the Topkins case,[16] the U.S. District Court considered algorithmic collusion as a conspiracy to manipulate the prices of commodities.[17] This ruling clearly established a precedent against any form of collusion, by traditional means, explicit or otherwise, involving price-fixing was anti-competitive.

These advancements put pressure on conventional understandings of "agreement" under the Act, where the appreciable adverse effect to competition (AAEC) presumption relies on evidence of coordination. Under Section 3 of the Act, an agreement, whether formal or informal, that causes or is likely to cause an AAEC is considered anti-competitive and void. The AAEC standard serves as the core analytical framework for determining whether practices such as price-fixing, bid-rigging, or market allocation harm consumer welfare or distort market dynamics. Traditionally, this requires proof of intent or communication among the competitors. However, in algorithmic collusion where direct evidence is hard to find, researchers and regulators increasingly promote a functionalist perspective, attuned to structural market realities, algorithmic structure, and effects, rather than intent or communication alone.

b. Abuse of Dominant Position

A giant may walk among men, but if it begins to trample, the law must cut it down to size.

The general discussions on potential harms by algorithm were focused on explicit or tacit collusion. However, recent discussions have raised concerns regarding the algorithms enabling dominant firms to abuse their market power. The dominant firm can indulge in self-preferencing, predatory pricing, price discrimination and other activities contravening Section 4 of the Act.[18] These practices may result in both monetary and non-monetary exploitation including degrading the quality of the services, reducing privacy protections, extracting more personal data or manipulating recommendation algorithms to favour profit-maximizing outcomes at the cost of user satisfaction. The monetary exploitation can be such as personalized pricing where the companies customize prices for various customers according to data about their traits or behaviours. Personalized pricing is more technically complex than algorithmic targeting, which enables the company to set prices for marginal and inframarginal clients differently (i.e., prices for two sets of consumers at-risk and safe).[19] From an economic perspective, when a company has market dominance, it can set prices higher than they would be in a competitive market. This leads to allocative inefficiency, where resources are not optimally allocated to meet consumer demand. Consequently, consumer surplus, the extra benefit consumers gain when they pay less than what they are willing to pay is reduced, harming the overall consumer welfare. Algorithmic tools such as personalized pricing and price discrimination enable companies to extract the maximum willingness to pay from the consumer, reducing consumer surplus. Further, it can also result in unfair outcomes for the consumers[20] as the prices shown are as per the personal data collected but not the cost. For example, algorithms may display higher prices to frequent users based on behavioural data, thereby boosting profits with reduced consumer welfare driven by data profiling rather than cost-based pricing. Though price discrimination purported by AI pricing is not violative of competition law in all cases, it largely risks going foul of the price-cost / economic cost test and giving rise to excessive pricing. The price-cost test is a key analytical tool which determines whether a price charged by a dominant firm is excessive by comparing the price charged to the cost of production, usually measured by an appropriate notion of economic cost such as average total cost or marginal cost. If the price significantly exceeds the economic cost without justification,  it may be deemed excessive and thus abusive under the law. If AI-enabled pricing leads to sustained high prices without corresponding improvements in quality or innovation, it may amount to excessive pricing, a form of exploitative abuse under competition law. Hence, the use of AI in pricing, while not unlawful per se, demands careful scrutiny when it affects market fairness and consumer welfare.

The Road to Regulation

Law is not the enemy of progress, but its measure—ensuring speed does not outrun sense.

Though algorithmic pricing has ushered new dilemmas for competition law, where anti-competitive agreements and abuse of dominant position by enterprises have become much more clandestine, scholars such as Professor Thibault Schrepel believe the paradigm shift brought by algorithm-based pricing is merely an “old wine in a new bottle”[21] and that the existing framework of antitrust regulation would largely fill the legal lacunas as they come. However, it is essential to pre-empt the changing legal landscape as algorithmic pricing wields the power to enable concerted action or collaboration under the guise of supra-competitive pricing.[22] Additionally, with respect to the abuse of the dominant position, algorithmic pricing can act as a facilitator by ensuring greater price discrimination, giving rise to predatory and excessive pricing. Therefore, this section of the paper delves into the existing legal landscape of competition law across jurisdictions and potential solutions to further streamline the vices of algorithmic pricing.

The potential danger of algorithmic pricing in India came before the CCI in the cases of Samir Agrawal v ANI Technologies[23] (Uber case), Re Alleged Cartelization[24] & Shikha Roy v Jet Airways[25]. Through the abovementioned cases the CCI cautioned that even though algorithmic pricing itself is not enough to establish cartelisation but the same can be used as a tool for tacit collusion by using shared parameters, communication and coordinated use of algorithms.

The flow of judicial ink by the CCI indicates that the Indian antitrust landscape has traversed from a position where the possibility of a hub-and-spoke arrangement was missed owing to the rationale of a third-party platform setting prices based on technical and historical data to a two-step test for determining algorithmic collusion. The two-step algorithmic test provides for assessing collusion in the absence of algorithms, followed by an inquiry into algorithmic collusion based on common software being used and the extent of human intervention. Though the test provides a potential safeguard against algorithmic collusion, the legal lacunas with respect to algorithmic pricing cannot entirely be filled by it.

For a much more complete picture, the test must be seen alongside the legislature, i.e., the Draft Digital Competition Bill, 2024[26], which is yet to take effect and possible learnings from the European law, i.e., Digital Markets Act[27], AI Act[28] & Digital Services Act[29]. In terms of what is yet to come with the advent of the Digital Competition Bill, it is clear that increasing data interoperability and portability, prohibition on self-preferencing, tying and bundling, unauthorised data use, prohibition of anti-steering provisions and increased reporting and compliance mandates are what is largely there for the Indian competition landscape. Further, to transpose the remaining inconsistencies by learning from the European legislature, though, seems a rather ideal and straightforward path but might not cater to the Indian legal landscape. It must be noted that a statute is like a tree, drawing life from the soil it springs from; to uproot it and plant it in foreign ground is to risk a law that stands but does not live. Therefore, filling the gaps in the Indian competition law by transposing the provision of the AI Act; which supplements the Digital Markets Act might not be ideal.

Therefore, this article further delves into possible solutions to novel problems posed by algorithmic pricing by categorising them into solutions based on an over-regulatory spirit and possible softer law propositions based on minimal interference. Before, delving into the possible solutions it is pertinent to note that the evolution of competition law in India can be categorised as a move from the Harvard School (roughly Pre-2015), emphasising a structure-based approach to the Chicago School focusing on an effect-based analysis.[30] The Harvard School has been rooted in the ideology that presumes concentrated market structures inherently harm competition. On the other hand the Chicago school stresses on an effect-based analysis, focusing on actual consumer welfare rather than indulging in a theoretical outlook.

However, it is pertinent to note that the Indian regulator has always measured anti-competitive practices against the anvil of public interest and development. Hence, it can be concluded that the Indian landscape is not one to shy away from a strict regulatory approach but at the same time is not supportive of onerous regulations that stifle innovation.

a. Hard law-based approach

The primary issue with algorithmic pricing leading to collusion is that the same disincentives undercutting prices and ordinary competitive practices as any price-cut would be met by immediate reduction by rivals as well.[31] Therefore, the recognition that supra-competitive markets arising out of algorithmic pricing can take the form of tacit collusion must be recognised as a reality. The same has been categorised by Ezrachi and Maurice to take the shape of hub-and-spoke markets, algorithms becoming a ‘digital eye’ that colludes on its own, the possibility of persons and enterprises recognising the pervasive use of the same/similar software which turns them into predictable agents that collude and finally software that dons the role of messengers ensuring that agreed prices are maintained.[32]

Therefore, a hard law approach can be the competition commission donning the role of a price regulator, not one in the absolute sense but fixing consumer surplus and producer surplus as variable ranges linked to the already established price-cost test. Hence, the producers will still have the incentive to reduce costs to increase profits, but at the same time, the price for the commodities cannot exceed the predetermined higher-end range of the consumer surplus and producer surplus variable.

The consumer surplus and producer surplus variable would be the range that would reward firms that are able to reduce the costs below the average industry median by affording them the right to charge a higher proportion of the cost than those firms operating at a cost equivalent to industry median and above the industry median production costs. To further streamline the same and recognise that dominant firms and firms with deeper pockets stand to benefit from this; categorisation based on available resources to firms would allow new entrants to compete with existing behemoths due to them being placed in different categories that factor in the size of firms and economies of scale. Hence, this would enable firms to innovate and reduce prices, but at the same time firms operating at lower cost levels can also indulge in ordinary competitive processes, and the increase of prices which was earlier incentivised by algorithmic pricing would be pre-emptively curbed by price regulation.

The same can be simply explained through a six-step process

1. The Competition Commission acts as a price regulator;

2. It then links consumer surplus and producer surplus to price-cost test to prohibit predatory and excessive pricing;

3. Then it sets variable ranges linked to the price-cost test;

4. Under this paradigm producers who reduce costs can charge at the higher end of the range while producers who cannot reduce costs can only charge lower end of the range;

5. Further, prices cannot exceed the predetermined higher end of the range;

6. Additionally, categorisations place firms with similar resources, size and economies of scale equally.

However, with the proposed hard-law approach it is admitted that the same would give rise to additional regulatory burdens and defining the range of consumer surplus variables would be difficult. Additionally, it is proposed that the consumer surplus range must not be static and should be adaptive to markets. To simply put the above model works on the principle of a rate-of-return regulation which is essential given the paradigm shift towards ex-ante competition law regulations.

b. Softer-law based approach

Understanding the rigour of the hard-law approach and potential regulatory burdens, the article also delves into possible soft-law approaches to remedy the burgeoning vices of algorithmic pricing. One of the more common approaches that can be seen across other legal landscapes in curbing the pace at which algorithmic pricing functions is by imposing a time lag.[33] A time lag enforces competition by inserting a delay between market signals and corresponding price adjustments. The same prevents instantaneous and self-reinforcing changes, reducing the risk of tacit collusion.

The same has been seen as a potential solution to algorithmic software undertaking human action in capital markets, i.e., High-frequency Trading emanating from algorithmic trading. [34]Hence, it is attractive to transpose the same to the competition law and the same can incentivise the competitive process of price-cutting and also deter the supra-competitive paradigm.

Additionally, in the absence of a sectoral regulator, the growing possibility of the Competition Commission of India undertaking behavioural remedy measures must be considered. Firstly, most Artificial Intelligence and algorithmic software are black-box in nature. The black-box characteristic refers to the nature of AI that cannot be explained. Simply put, the AIs don’t function in a way where the input and output can be explained, the way the deep neural networks function largely remains opaque.[35] Hence, models that can be explained, i.e., White-box models and the ones that can be reverse-engineered, can be the only permissible algorithmic pricing software.

Additionally, behavioural measures that the CCI can impose must include mandatory AI audits to mitigate anti-competitive practices emanating from the growing presence of Artificial Intelligence. The CCI can impose mandatory AI audits on Systematic Significant Digital Enterprises, to begin with as the magnitude of harm remains the largest when these enterprises carry out anti-competitive practices.

Lastly, the CCI could also consider the introduction of regulatory sandboxes i.e., that are introducing algorithmic pricing in a controlled manner by enterprises before they are allowed to implement the same at a larger scale. Given that the Digital Competition Bill is an ex-ante regulation, sandboxes can be instrumental in enabling firms to test whether their algorithms give rise to any potential harm and can act as a safe regulatory space to address the same.[36]

However, the possible concerns with the soft-law approaches remain the requirement of a proactive regulator as the regulatory burden in implementing the same would be greater than that in the hard-law approach and the current status quo. The CCI would be required to take up the tasks of introducing regulatory sandboxes, provide standards and regulations surrounding AI audits and also undertake the exercise of classifying AIs as black and whit box AIs. Further, the soft-law approach would not have the same rigour as the hard-law approach.

Conclusion

The promise of AI is dazzling, but brilliance without boundaries can blind. When innovation outruns regulation, it’s not progress—its peril dressed in code.

The growth of artificial intelligence has brought about a paradigm change in all domains, and competition law is no exception. However, the growth of artificial intelligence requires us to juxtapose its pro-competitive benefits of increasing efficiency and innovation against the distortion of competition by bringing a different dimension to collusion and abuse of dominance. What must be recognised is that artificial intelligence has ushered in an era where the invisible handshake wields more power than Adam Smith’s invisible hand. Therefore, the legal landscape must recognise that measures against anti-competitive practices such self-preferencing, tying and bundling are well-founded but not enough. In particular the measures fall short with the burgeoning field of data where unauthorised use of data and increased data portability and interoperability exemplify existing anti-competitive practices.

The ability of AI pricing to streamline collusion into ways that slip under the radar calls for recognising that the current regime of establishing a ‘plus factor’ in addition to conscious price parallelism might not be enough. Recognition of the fact the ex-ante competition law regime can enable the regulators to take a proactive price regulatory approach is essential. Therefore, possible remedies for the preservation of competition and consumer protection include providing for a consumer and producer surplus range linked to the existing price-cost test to incentivise ordinary competitive behaviour in the market. Additionally, behavioural remedies include calling for increasing AI audits, allowing only white-box AIs to fix prices and

[1] Renato Nazzini and James Henderson, ‘Overcoming the Current Knowledge Gap of Algorithmic “Collusion” and the Role of Computational Antitrust’ (2024) 4 Stanford Computational Antitrust <https://law.stanford.edu/wp-content/uploads/2024/02/Algorithmic-Collusion.pdf> accessed 11 April 2025.

[2] ‘When Algorithms Set Prices: Winners and Losers’ (Oxera Consulting LLP, 28 June 2017) <https://www.oxera.com/insights/agenda/articles/when-algorithms-set-prices-winners-and-losers/> accessed 9 April 2025.

[3] OECD (2017), ‘Algorithms and Collusion: Competition Policy in the Digital Age’, OECD Roundtables on Competition Policy Papers No. 206, OECD Publishing, Paris <https://doi.org/10.1787/258dcb14-en> accessed 30 March 2025.

[4] Ariel Ezrachi and Maurice E. Stucke, Virtual Competition: The Promise and Perils of the Algorithm-Driven Economy (Harvard University Press 2016).

[5]Digvijay R. Singh, ‘Protect against Self-Learning Algorithms?’ (IndiaCorpLaw, 6 January 2022) <https://indiacorplaw.in/2022/01/algorithmic-collusion-can-the-competition-act-protect-against-self-learning-algorithms.html> accessed 25 March 2025.

[6] OECD, ‘Algorithms and Collusion’ (n 3).

[7] Oliver Ralph, ‘Insurance and the Big Data Technology Revolution’, Financial Times (24 February 2017)

<www.ft.com/content/bb9f1ce8-f84b-11e6-bd4e-68d53499ed71> accessed 28 March 2025.

[8] Gonenc Gurkaynak, ‘Algorithms and Artificial Intelligence: An Optimist Approach to Efficiencies’ (2019) 5 CLPD 29.

[9] ibid.

[10] The Competition Act, 2002, s.3.

[11] Calvano, Emilio, Giacomo Calzolari, Vincenzo Denicolò, and Sergio Pastorello, ‘Artificial Intelligence, Algorithmic Pricing, and Collusion’ (2020) 110(10) American Economic Review 3267 <https://www.aeaweb.org/articles?id=10.1257/aer.20190623> accessed 20 March 2025.

[12] ibid.

[13] Shambhavi Jha & Simran Nagra, ‘An Analysis of Algorithmic Collusion under Indian Competition Law: Comparative Study with EU and US’ (2024) 5 Jus Corpus LJ 22 <https://www.juscorpus.com/wp-content/uploads/2025/01/3.-Shambhavi-Jha.pdf> accessed 30 March 2025.

[14] Shikha Roy v. Jet Airways (India) Limited, 2021 SCC OnLine CCI 31 (Shikha Roy case).

[15] Abhishek Rudra, ‘Hubs and Spokes Arrangement through Algorithmic Collusion - Economic

and Legal Implications’ (2022-2023) 4 Indian Journal of Law and Legal Research 1 <https://www.ijllr.com/post/hubs-and-spokes-arrangement-through-algorithmic-collusion-economic-and-legal-implications> accessed on 30 March 2025.

[16] United States v Tompkins (ND Cal 2015) Crim Case No 15-cr-00201.

[17] Raghuvansh Seth, ‘Interplay of Algorithmic & Tacit Collusion with Competition Law’

(2022) 16 NUALS LJ 131 <https://nualslawjournal.com/wp-content/uploads/2023/01/nuals-law-journal-issue-16-1.pdf> accessed on 5 April 2025.

[18] The Competition Act, 2002, s.4.

[19] OECD (2023), ‘Algorithmic Competition, OECD Competition Policy Roundtable Background Note’

<www.oecd.org/daf/competition/algorithmic-competition-2023.pdf> accessed on 1 April 2025.

[20] Qian Li, Niels Philipsen and Caroline Cauffman, ‘AI‑enabled price discrimination as an abuse of dominance: a law and economics analysis’ (2023) 9 China-EU Law Journal 51 <https://link.springer.com/article/10.1007/s12689-023-00099-z> accessed on 30 March 2025.

[21] 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 13 April 2025.

[22] Ariel Ezrachi and Maurice E Stucke, ‘Sustainable and Unchallenged Algorithmic Tacit Collusion’ (2020) 17(2) Northwestern Journal of Technology and Intellectual Property 217 <https://ora.ox.ac.uk/objects/uuid:fde32e64-30b0-41b8-9477 de4cc716d902/files/m4ccfcc7d271be9b6a42f5ca60b02e26f> accessed on 10 April 2025.

[23] Samir Agrawal v. CCI (Cab Aggregators Case), (2021) 3 SCC 136.

[24] Alleged Cartelization in the Airlines Industry, In re, 2021 SCC OnLine CCI 3.

[25] Shikha Roy case (n 12).

[26] ‘MCA invites public comments on Report of Committee on Digital Competition Law and Draft Bill on Digital Competition Law’ (2024) PIB Delhi < https://pib.gov.in/PressReleasePage.aspx?PRID=2013947> accessed on 5 April 2025.

[27] Regulation (EU) 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 [2022] OJ L265/1.

[28] Regulation (EU) 2024/1689 of 12 July 2024 on laying down harmonised rules on artificial intelligence [2024] OJ L219/1.

[29] ​Regulation (EU) 2022/2065 of 19 October 2022 on a Single Market For Digital Services and amending Directive 2000/31/EC [2022] OJ L 277/1.

[30] Saakshi Agarwal and Chintan Bhardwaj, ‘Goals of Competition Law in India’ (20 February 2021) SSRN Electronic Journal.

[31] Edward M Iacobucci, ‘AI Pricing and the Case for Permissive Competition Law Enforcement’ (2024) Oxford Business Law Blog <https://blogs.law.ox.ac.uk/oblb/blog-post/2024/07/ai-pricing-and-case-permissive-competition-law-enforcement> accessed 13 April 2025;  Zach Y Brown and Alexander MacKay, ‘Competition in Pricing Algorithms’ (2023) 15(2) American Economic Journal: Microeconomics 109 <https://doi.org/10.1257/mic.20210158> accessed on 10 April 2025.

[32] Ezrachi & Stucke, ‘Sustainable and Unchallenged Algorithmic Tacit Collusion’ (n 20).

[33] Michal S. Gal, ‘Limiting Algorithmic Coordination’ (2023) 38(1) Berkeley Technology Law Journal <https://btlj.org/wp-content/uploads/2023/10/0004-38-1-Gal.pdf> accessed on 5 April 2025.

[34] Yesha Yadav, ‘How Algorithmic Trading Undermines Efficiency in Capital Markets’ (2015) 68 Vanderbilt Law Review 1607.

[35] G Montavon, W Samek and K Müller, ‘Methods for Interpreting and Understanding Deep Neural Networks’ (2018) 73 Digital Signal Processing 1 <https://www.sciencedirect.com/science/article/pii/S1051200417302385> accessed 13 April 2025.

[36] Competition & Markets Authority, ‘Algorithms: How They Can Reduce Competition and Harm Consumers – Summary of Responses to the Consultation’ (2021) <https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/991676/Summary_of_responses_to_algorithms_paper_publish.pdf>  accessed 13 April 2025.