Author: Anirudh Subhaian Padmanabhan, Tamil Nadu National Law University

This paper examines algorithmic collusion in mergers, the research addresses the legal and economic problem of how competition law frameworks adapt to detect anticompetitive conduct that emerges without human interference, but via parallel algorithmic strategies. The paper combines legal research with economic theory on tacit collusion and game theoretic modelling of algorithmic interactions in both national and international including the EU, US, and India, the paper identifies key regulatory gaps. It particularly examines how mergers involving firms with significant digital infrastructure may inadvertently facilitate anticompetitive algorithmic behaviour after transaction.This article highlights the traditional antitrust tools are ill equipped to address the opaque and self learning nature of pricing algorithms. Moreover, many mergers miscalculate the capabilities of the AI, and conclude with frameworks to address tacit collusion.

KEYWORDS:

Algorithmic Collusion, Antitrust Law, Mergers and Acquisitions, Predictive Pricing, Competition Policy, Digital Markets

In recent times, the price fixing of products from many major firms are decided by AI rather than human interactions in real time; by absorbing competitors without any human input, this creates a new form of price coordination. In a traditional collusion, companies agree on a fixed price but at the right time both companies change their price therefore maintaining a stable market rate, this can be legally challenged, but as for algorithm collusion, the market rates are changed by AI rather than humans which raises concerns current law can't address. The real risk comes when companies merge and combine their technological infrastructures. Mergers lead to combined data and fewer pricing models, increasing risks, so these increase the likelihood of coordinated outcomes in the market done by AI. This paper argues that algorithmic collusions present a growing challenge to competition and modern M&A activity, especially in digital markets, which cannot be dealt with by current antitrust law.

While classic collisions require continuous coordination between competitors, algorithmic collusion can emerge successfully when the firms provide self learning pricing algorithms that can align behaviour to maximize the profits of the company. There are two types of collusion. Explicit collusion and tacit collusion, explicit collusion involves secret deals, while tacit collusion occurs when firms independently align behaviour without an agreement for example, company’s price varies with respect to others, which is done by AI without human knowledge through repeated interactions.

Algorithms use machine learning to watch how the market reacts to price changes, they lack ethics or legal awareness, optimizing solely for profit. For the firm. The algorithm observes what happens when the rate changes and it observes how the competitors react when the market changes, over time, the algorithm tests different prices and learns which strategies maximize profit the algorithm stores feedback for future use, overtime the algorithm realizes that avoiding price cuts leads to more stable and higher profits so it starts to keep the prices high and also obverse the competitors so they don't undercut.

REAL WORLD EXAMPLES

• Uber and Ola:

Apps like Uber and Ola use dynamic pricing algorithms. This algorithm changes the prices of a ride based on supply and demand, but the real problem is these companies use similar types of algorithms therefore even without communication, similar algorithms can raise prices in tandem, leading to uniform high fares.

• Airlines and Online travel:

Airlines have used revenue management algorithms to change ticket prices based on bookings, timing, and demand. At the same time, online travel agencies use their algorithms to adjust prices and show “best deals.” Using similar data, these systems often align prices over time, reducing competition. In some countries, regulators have raised competition concerns because travellers see nearly identical prices across all the platforms.

The legal framework of the majority of international antitrust laws are built around one key idea which is collusion requires an agreement or some form of coordination or shared intent. As for United States Sherman Act, Section 1, the law bans any agreement which interferes or restrain trade, in order to prove a collusion, the courts need proof of concerted action which is either direct or through emails or even unusual pricing patterns. The most important evidence is the intent which is a proof of a plan to work together.

As for European Union Article 101(1)TFEU outlaws any agreements or concerted practices that limit competition, this includes both formal contracts and informal coordination, the law also focuses on the object or effect that are anti-competitive but still assumes some common understandings between parties.

As of India Competition Act, 2002 Section 3 prohibits anti competitive agreements which includes cartels (Section 3(3)). It presumes certain conduct is harmful, but still requires proof of an agreement even if informal.

All three systems US, EU, and India depend on the idea of a meeting of minds.  Whether it’s a written deal or a silent nod, some form of human coordination is expected and that is exactly why algorithmic collusion slips through the cracks.

Algorithms don’t shake hands, make calls, or send shady emails. But they can still learn to behave like cartels by working quietly, efficiently, and legally undetected. Antitrust law needs to show that a company was meant to collude, but if an algorithm raises prices on its own, based on market patterns, the company can say “We didn’t plan this. This is what the code has learned.”  No intent means no liability under current rules for the company.  Normally there are trails left by companies like emails, WhatsApp chats or secret meetings but algorithms just watch, learn and respond to the pricings of the competitors so there is nothing to trace back to an illegal deal. Current laws focus only on the conduct of what the company did, rather than the result produced by the company so when algorithms act through automated behaviour it is tough to say who did anything wrong it is just Algorithm responding to incentives. The current scenario is a self-learning pricing algorithm that notices what happens whenever it lowers prices, competitors follow, or profits drop so over time, it stops lowering prices entirely not because anyone told it to, but because it learned that stable pricing makes more money.

• Ola/Uber Surge Pricing Case (2017–2022)

This took place in Bangalore, and was filed before the Competition Commission of India (CCI). A law student filed a complaint against Uber and Ola, alleging that their algorithm based surge pricing was effectively a cartel system in disguise.  Though the drivers are technically independent contractors, the apps set prices centrally via algorithms so instead of each driver competing freely, everyone ends up charging similarly high rates during peak times, Initially, the CCI dismissed the case (2017) on the basis that there was no evidence of an agreement between the drivers and Ola and Uber were just providing technological platforms and not directly colluding with each other. Later, the Delhi High Court reopened the issue, citing potential future antitrust risks.

• Eturas Case (2016) – European Union

A Lithuanian online travel platform sent a system wide message to travel agencies suggesting that they cap their discounts, but this is not done by the company, rather it is The platform that used automated software to enforce it, but the judgement given by The EU Court is that this system wide communication could count as an illegal agreement, The EU ruled against it only because a human sent the system-wide message that someone wrote and sent that system wide message, If not for that human action there wouldn't have been any legal actions.

• U.S. v. David Topkins (2015) – United States

Topkins and other sellers of online posters used pricing algorithms to fix prices on Amazon, but they agreed in advance to use the software that way so that this comes under explicit collusion being carried out through tech, therefore there was a legal action. 

These cases show that Modern competition law was never designed to regulate autonomous, adaptive pricing systems that align behaviour without coordination. As a result, enforcement agencies are often constrained by the very frameworks that once gave them clarity.

Why it’s Economically Dangerous

• Algorithms Scale Instantly

One powerful pricing algorithm can impact millions of transactions in real time, worse, if multiple firms use similar models or even same vendors the entire markets start behaving alike without any planning or collusion. This results in the reshaping of competition at a national or even global scale just with a few lines of codes.

• Black Box Problem

Most pricing algorithms are opaque that even companies don’t fully understand how they evolve so that the regulators can’t inspect what they can’t see therefore there will be no proof that a collusion has taken place.

• Illusion of Competition

On the surface there are multiple firms that have flashy branding or app choices, but behind the scenes, algorithms harmonize prices to avoid undercutting, and punish deviation, so that the market looks competitive, but behaves like a monopoly.

Mergers today aren’t just about buying factories or delivery trucks rather they are about owning data, algorithms, and digital decision making power. When two companies merge, their data and algorithms merge resulting in more data and smarter algorithms. Combining customer and transaction data results in a better training set and also algorithms get stronger at predicting behaviour and pricing for profit when it merges. When the companies fall under the same roofs, firms often standardize the merged algorithms that reduces price variation and boosts synchronized pricing. At the end if everyone uses the same or similar systems post-merger competition shifts from dynamic and unpredictable to stable and uniform.  Digital M&As make algorithmic collusion the default.

Real-World M&A Examples with Algorithmic Implications:

• Google–Fitbit Merger (2021)

The EU flagged that Google could merge Fitbit’s health data with its powerful ad and search algorithms because it raised fears of dominance in health-related ads, where Google could predict, target, and price more effectively than rivals. The major concern was the presence of massive data and machine learning which when coupled gives an unfair advantage in algorithmic decision making.

• Zomato–Uber Eats India (2020)

In 2020, Zomato bought Uber Eat’s India operations which reduced the number of independent food delivery players. Both companies had smart pricing engines and are now under one roof, currently they have no evidence of price fixing, but common optimization logic can quietly emerge. The result of this merger is that with Swiggy as the only competitor, the market became more susceptible to parallel pricing.

Mergers in the digital age can supercharge anticompetitive behaviour, even without traditional monopolies or intent to collude. Some markets like airlines, food delivery, or online travel already show natural price alignment. Now when they merge the algorithms train on larger datasets, respond faster, and quietly settle on shared pricing behaviour. As a result, the prices will go up and the competition goes quiet, but no one agrees to anything, it is just AI learning new things. These algorithms learn by testing actions and rewarding results. In a stable market with fewer players, the AI quickly finds that not undercutting prices earns the most, as the result M&A creates the perfect environment for price stabilizing behaviour by accident or by design.

The Antitrust regulators are still using old tools to solve new algorithmic problems and that is the gap. The Merger Review Still Focuses on Market concentration, Price effects, short-term consumer harm and Traditional competitive overlap which is product or service based. The major things that the mergers still miss are Whether the merging firms use compatible pricing algorithms, Whether AI or code integration could trigger tacit collusion and Any review of data flows or tech stack merging, which are crucial for algorithmic collusion, also the firms rarely disclose algorithmic pricing strategies, Reinforcement learning models and AI-based competitive tools So regulators don’t even know what to look for.

REGULATORY CATCH UP (STILL IN PROGRESS):

• EU Digital Markets Act (DMA): Targets Big Tech behaviour after mergers, but not algorithmic risks during the deal.

• CCI: Has begun probing digital M&A more closely, but tech and algorithm scrutiny is still weak or superficial.

• US FTC: 2023 draft merger guidelines mention platform effects and data power, but AI is not a core focus yet.

RETHINKING ANTITRUST: LEGAL AND ECONOMIC REFORMS FOR THE ALGORITHIMIC ERA:

The law is looking for conversations, while the collusion is happening in code. Elements that traditional enforcement misses are No emails, No calls, No deals Just algorithms quietly aligning outcomes then the loss of human intent therefore regulators cannot “blame” anyone then the absence of a formal agreement So courts don’t see a cartel and at the last No direct causation rather Just profit-maximizing behaviour from machine learning. Algorithms are the Invisible Market Players, because they adjust prices, observe rivals, and learn to avoid undercutting but the main point is they don’t conspire — they just optimize. The M&A Reviews are Still Too Human Focused and the Regulatory scrutiny remains obsessed with ownership and market share, It also ignores shared tech stacks, reused codebases, or AI alignment risks.

PROPOSED LEGAL REFORMS:

If enforcement is to catch up with code, laws must evolve beyond human intent and start addressing algorithmic behaviour. The current antitrust law hinges on explicit or tacit coordination between people, but algorithms can independently learn to align without any meeting or message. The Reform Idea is to update legal definitions to treat algorithmic parallelism as a possible concerted practice, even in the absence of intent or communication.

Merger filings should not only consider the owner of the company rather They should know about the uses of pricing algorithms, about the shared data sources or AI providers and about the consequence of the merger and its effect on competition behaviour this would perfectly help EU to Include in Form CO for merger control, India to Embed in CCI’s combination review and US to Add to FTC/DOJ HSR filings.

Antitrust bodies need in-house expertise that speaks both legalese and Python. These units should audit algorithms and simulate market effects to flag risks early. A prime example is The UK’s CMA has launched a Data, Technology & Analytics (DaTA) Unit, a model worth emulating in India, EU, and the U.S.

Before pricing algorithms are unleashed on the market, let them be tested in regulator monitored environments so Firms can trial AI based pricing tools, Regulators can observe unintended convergence or anticompetitive patterns and Adjustments can be made before real consumers feel the pinch. Financial regulatory sandboxes already used by SEBI (India) and MAS (Singapore). The current frameworks focus on market overlap. But now, code overlap is just as dangerous so the new merger reviews must consider Data interoperability risks, Shared AI vendors or platforms and potential for algorithmic imitation or convergence, right now the 2023 U.S. draft guidelines mention “platform effects” so the next step is recognizing “algorithmic effects.”

The International Coordination is Critical because Algorithms don’t respect borders but antitrust laws do so the core challenge is Pricing AI can operate in real time across markets, while enforcement is slow, fragmented, and national. Global Coordination Matters because Cross-border platforms like Amazon, Booking.com etc. use unified algorithms across jurisdictions, one country’s blind spot becomes another’s loophole so that is a regulatory arbitrage and Data sharing and common standards are essential to tackle algorithmic collusion at scale. The essential improvements are Intelligence sharing protocols among competition regulators, Common frameworks for detecting and auditing algorithmic pricing and Joint investigations for cross border M&A or collusive tech deployments.

Not all algorithms are villains if used right, they lower costs, match demand efficiently, and expand access, banning them would kill innovation, instead Focus on design incentives, not just end results, Promote algorithmic transparency so regulators and auditors understand what the code is optimizing for, and Encourage competition in AI itself so More diverse pricing logics emerges reducing chances of convergence, the main goal is not to fight algorithms but to shape the market rules that they learn from.

A Future-Proof Antitrust Framework:

This paper has shown that algorithmic collusion represents not a violation of existing antitrust rules, rather a pattern of how harm emerges silently, efficiently, and without human interference. As machine learning tools observe, adapt, and optimize for profit, they also coordinate prices, suppress competition, and punish deviation, all without a single phone call or email. The risk is increased in digital mergers, where firms not only consolidate market share, but also data ecosystems, codebases, and pricing logic. In such environments, tacit collusion becomes not a strategy but an outcome of design rather than design itself. Traditional legal frameworks, still anchored in proving “agreement” or “intent,” are not equipped to tackle this.

The solution is not to fear or ban algorithms. It is to update the legal and institutional logic that governs them. This means Expanding the definition of “agreement” to include emergent algorithmic alignment, Mandating algorithmic disclosures during M&A assessments, building technical enforcement capacity within competition agencies and coordinating internationally to prevent cross-border regulatory arbitrage, The danger lies not in artificial intelligence, but in legal complacency and if algorithms can learn to collude, then antitrust law must learn to evolve.

• United States v. Topkins, No. 3:15-cr-00201, (N.D. Cal. 2015).

• Case C-74/14, Eturas UAB v. Lietuvos Respublikos konkurencijos taryba, ECLI:EU:C:2016:42 (Jan. 21, 2016).

• Sherman Antitrust Act, 15 U.S.C. § 1 (1890).

• Consolidated Version of the Treaty on the Functioning of the European Union art. 101(1), Oct. 26, 2012, 2012 O.J. (C 326) 47.

• The Competition Act, No. 12 of 2003, § 3, Acts of Parliament, 2003 (India).

• Commission Decision, Case COMP/M.9660, Google/Fitbit, Dec. 17, 2020.

• Zomato/Uber Eats India, Combination Reg. No. C-2020/02/734, Competition Commission of India (Jan. 21, 2020).

• Commission Decision, Case M.10615, Booking/eTraveli Group, Sept. 25, 2023.