When the Algorithm Becomes the Cartellist

April 2026 | Başaran Law Firm

Introduction

For decades, competition law's mental image of a cartel was a group of executives in a hotel room, agreeing on prices over dinner. That image is obsolete. The cartellist of the 2020s may be a line of code that reaches supracompetitive pricing not through human conspiracy, but through machine learning, shared software, and algorithmic interdependence. Regulators on both sides of the Atlantic, and increasingly in Ankara, are confronting this reality. The legal framework is catching up but unevenly, and with a doctrinal gap at its centre that no jurisdiction has yet closed.

1. Three Scenarios of Algorithmic Collusion and Where the Law Stands

Competition law's prohibition of anticompetitive agreements — Article 101 TFEU in the EU, Article 4 of Law No. 4054 (4054 sayılı Kanun) in Turkey, Section 1 of the Sherman Act in the US — requires, at its core, some form of understanding between competing firms. Algorithmic pricing arrangements strain that requirement in three increasingly complex ways.

The first scenario is legally settled: firms agree on prices and use software to implement the cartel. The UK Competition and Markets Authority's Posters case and its US equivalent established that the use of repricing software to execute a preexisting price-fixing arrangement does not change its legal character. The agreement came first; the algorithm served it.

The second scenario is more complex: competing firms use the same third-party pricing algorithm, producing coordinated prices without any direct communication. This is the hub-and-spoke structure (merkez-kol yapısı) — the algorithm provider is the hub, the competing firms are the spokes. The EU's Eturas judgment (C-74/14) established that passive participation in a common pricing platform can constitute a concerted practice (uyumlu eylem) under Article 101 TFEU, provided firms were aware of the system and continued to use it.

The US Department of Justice's civil action against RealPage, whose YieldShare software aggregated non-public rental pricing data from competing landlords to generate pricing recommendations, tested the same logic under Sherman Act Section 1. The November 2025 settlement imposes a three-year oversight regime and restricts the use of non-public competitor data. It stops short, however, of a judicial finding that algorithmic rent coordination itself violates US antitrust law. For practitioners, the settlement's data-governance conditions are the operative compliance benchmark: ingesting competitors' non-public pricing data through a shared platform is precisely the conduct that triggered enforcement.

The third scenario is the most legally challenging and the most dangerous: independently deployed machine learning algorithms, each instructed to maximise profits, converge on supracompetitive prices through reinforcement learning without any human communication or shared platform. Experimental research confirms that Q-learning algorithms can learn collusive strategies autonomously, including punishment of deviations, in simulated markets. This autonomous algorithmic collusion is a digital evolution of the classic oligopoly problem. It may fall entirely outside the reach of Article 101 TFEU and its equivalents: no agreement, no concerted practice, just parallel machines reaching the same harmful destination. Every major jurisdiction has acknowledged the gap; none has yet closed it.

2. Vertical Restraints: Resale Price Maintenance in the Age of Price-Monitoring Software

Resale price maintenance (RPM), the practice of a supplier dictating minimum retail prices, has long been treated as a "by object" restriction (amaç itibarıyla kısıtlama) under both EU and Turkish competition law. What has fundamentally changed is the enforcement technology available to suppliers, and consequently the scale and precision of the harm. Price-monitoring software now enables suppliers to detect retailer deviations from recommended prices within hours. In a series of decisions against Asus, Denon & Marantz, Philips, and Pioneer, the European Commission imposed over €110 million in combined fines for online RPM schemes in which nominally "recommended" prices were enforced through algorithmic surveillance, followed by commercial pressure, threats of supply withdrawal or removal of promotional support. The online context amplified the harm: because competing retailers used their own pricing algorithms to match one another, a restriction imposed on a single low-pricing retailer propagated automatically through the market.

The Turkish Competition Board's 2025 enforcement record reflects the same dynamic. Decisions in food, fast-moving consumer goods, and cosmetics markets increasingly involve fact patterns where recommended retail prices function as effective price floors not through explicit contractual obligation, but through algorithmic detection of deviations and subsequent commercial intervention by suppliers. The legal analysis is classical; the evidentiary record is digital.

3. Dominance, Self-Preferencing, and the AI Infrastructure Problem

Article 102 TFEU and Article 6 of Law No. 4054 prohibit dominant firms from abusing their market position. Algorithmic conduct generates two structurally distinct forms of abuse.

The first is exclusionary: dominant platforms manipulate ranking algorithms to favour their own products over those of rivals. The European Commission's Google Search (Shopping) decision established the legal framework; the Digital Markets Act (DMA) now addresses this ex ante. Designated "gatekeepers" firms controlling core platform services are required to apply transparent, fair, and non-discriminatory conditions to ranking, removing the competitive advantage of self-preferencing before harm materialises, rather than remedying it years later.

The second form of exploitative conduct through personalised pricing remains less adequately addressed. Article 102 TFEU's tools fit imperfectly: the excessive pricing provision requires establishing dominance and a supra-competitive margin; the price discrimination provision targets firm-to-firm rather than firm-to-consumer discrimination. The GDPR, Consumer Rights Directive, and the recently effective EU AI Act GPAI regime each apply partially, but a coherent enforcement framework for algorithmic consumer exploitation has yet to crystallise across any jurisdiction.

The deepest structural challenge is upstream. The AI supply chain, specialised in chips, hyperscale cloud compute, curated training data, and foundation models, is concentrated in a small number of firms. The FTC's January 2025 Section 6(b) staff report identified the Microsoft–OpenAI, Amazon–Anthropic, and Google–Anthropic partnerships as arrangements involving over USD 20 billion in combined investment and commercial terms, including equity rights, revenue-share provisions, and minimum-spend commitments, that warrant sustained competition scrutiny. European merger control has struggled to keep pace: following the Illumina/Grail judgment narrowing Article 22 referral powers, the effective acquisition of substantially all Inflexion AI staff by Microsoft was found unreviewable under the EU Merger Regulation. The residual route under Article 102 TFEU ex post, via the Towercast doctrine, is structurally ill-suited to partnership arrangements that fall short of a formal transfer of control.

4. Regulatory Architecture: EU, US, and Turkey Compared

The EU and US approaches diverge sharply on instrument design while converging on diagnostic categories. Europe has constructed an ex ante architecture. The AI Act's general-purpose AI (GPAI) obligations, which entered into force on 2 August 2025, impose documentation, transparency, and systemic-risk assessment requirements on providers of large-scale foundation models. The DMA adds structural gatekeeper obligations that operate independently of any competition investigation. Together, these create a regulatory layer that ex post competition enforcement cannot replicate on its own.

The United States has no federal equivalent to the DMA or AI Act. Enforcement is case-by-case under the Sherman Act and FTC Act, shaped by the Ninth Circuit's 2025 ruling narrowing the doctrine under which common vendor use constitutes a Section 1 agreement. The DOJ has nonetheless signalled continued vigilance against algorithmic and AI-driven coordination, and ongoing monopolisation proceedings against Google in search and ad-tech indirectly shape the competitive environment in which AI assistants are distributed.

Turkey's framework mirrors the EU model structurally: Article 4 of Law No. 4054 is the functional equivalent of Article 101 TFEU; Article 6 mirrors Article 102. What sets Turkey apart is its operational approach: as confirmed in its 2025 Annual Report, the Turkish Competition Authority has deployed AI-assisted market-anomaly detection and algorithmic-coordination identification tools, which are now integrated into its enforcement decision-making. The TCA is among the first competition authorities globally to operationalise such tools at scale. Firms using algorithmic pricing in Turkish digital marketplaces, food supply chains, or fast-moving consumer goods markets now face a surveillance infrastructure that combines the EU's doctrinal framework with machine-learning-assisted pattern detection.

On 23 July 2024, the European Commission, the UK CMA, the DOJ, and the FTC issued a joint statement identifying shared concerns: concentrated AI input markets, entrenchment through cloud-model partnerships, and the risk that such arrangements escape classical merger review. The analytical convergence across jurisdictions, even where instruments diverge, is itself a compliance signal: the same structures are being scrutinised by regulators who are now explicitly coordinating their frameworks.

Forward Look: The three algorithmic collusion scenarios range from settled law to a genuine regulatory gap. The first scenario applies classical competition analysis to new technology; the second necessitates thorough documentation of pricing systems and client independence; and the third, involving autonomous machine learning convergence, currently falls outside existing legal frameworks. However, this gap won't last, as legislative reforms are underway, academic consensus on economic harm exists, and regulators are gaining both the mandate and technical ability to address these issues. Firms using machine learning in pricing must ensure compliance architecture is established before the law catches up.

*This publication is for general information only and does not constitute legal advice.