Q&A: Algorithmic collusion: antitrust enforcement
August 2026 | SPECIAL REPORT: COMPETITION & ANTITRUST
Financier Worldwide Magazine
August 2026 Issue
FW discusses algorithmic collusion in antitrust enforcement with Nicole Kar, Lauren O’Brien, Katherine B. Forrest, Eyitayo St. Matthew-Daniel and Henrik Morch at Paul, Weiss, Rifkind, Wharton & Garrison LLP.
FW: How are competition authorities redefining what constitutes ‘coordination’ in markets where algorithmic systems influence pricing decisions without direct human agreement?
Morch: In the EU, the existing concept of a ‘concerted practice’ already catches a ‘meeting of minds’ that removes strategic uncertainty between competitors, with no formal agreement and no handshake required – a deliberately low threshold. Algorithmic interaction can meet it even where no human ever spoke to a rival. Brussels asks not, “did you agree?” but “did you knowingly plug into a mechanism that aligned your conduct with rivals?” So, in most cases no redefining is needed – the EU is already equipped to catch reach code-mediated alignment where there is a human somewhere in the loop.
Forrest: The US legal anchor is still an ‘agreement’. The Department of Justice (DOJ) and private plaintiffs argue the element is met when a common algorithm is offered to competitors that each sign up to knowing their rivals are doing the same. Defendants argue that even in such a ‘hub-and-spoke’ arrangement, an ‘agreement’ among the competitors around the rim is required. The genuinely open question is what facts, beyond competitors’ parallel adoption of a common algorithm, are enough for a court to infer a tacit agreement. For example, courts generally have declined to find viable federal antitrust claims where the algorithm relies only on public data, rather than sensitive, confidential competitor data, although the DOJ has not endorsed that categorical approach.
Kar: UK law in this area is derived from the European Union (EU) approach, so the law can catch ‘knowing but silent’ alignment with competitors. A recent Competition and Markets Authority (CMA) blog warned that advanced artificial intelligence (AI) systems might reach coordinated outcomes even without any human intent to collude. This creates difficulty in establishing the ‘meeting of minds’ needed for an illegal anticompetitive agreement. The CMA has drawn an analogy with an employer’s responsibility for the actions of an employee, but vicarious employer liability is not absolute. Where an AI agent’s actions in fixing prices were not foreseeable, it is not yet clear if the business would be liable – even if, for understandable reasons, the CMA’s line is that it would be. From a UK perspective, this uncertainty does not mean less compliance risk: the CMA can fall back on its market investigation powers, which enable it to review and remedy market distortions without the need to show any agreement.
“Enforcers and courts are increasingly scrutinising how algorithms work, including the code behind them.”
FW: To what extent are existing liability frameworks sufficient to address situations where algorithms align outcomes without explicit collusion?
St. Matthew-Daniel: In the US, the key requirement is to prove an ‘agreement’. This can be done through direct evidence or through indirect evidence, such as without evidence of explicit collusion. The central question is whether businesses remain engaged in independent decision making or instead use a shared algorithmic tool as a means of coordination. Recent cases illustrate that distinction. The first case before an appellate court concerning algorithmic pricing was rejected where there was no claim that there was any agreement between competing Las Vegas hotels that had independently licensed the same revenue-management software, and no pooling of each hotel’s confidential data. But other courts have allowed price-fixing claims to proceed where competitors allegedly knowingly shared non-public data through a common provider, despite no evidence of direct communication between them. As algorithmic systems evolve, including through agentic AI, the same basic inquiry should remain: whether the facts support an inference of an actual agreement, even if not an explicit one.
Morch: In the EU, the law can already catch ‘knowing but silent’ alignment. In addition, a standalone exchange of data that reduces uncertainty about a rival’s future pricing can be illegal, without needing to prove effects. The more difficult issue would be if collusion was the result of agentic AI outputs which could not be traced back to any human direction. This may be difficult to catch under existing law and precedent.
O’Brien: The UK CMA is well placed here. The law on illegal agreements mirrors the EU approach and the CMA takes a robust view of its reach. But in addition, the UK has broad market powers which enable it to review any feature of a market or the economy which may be harming competition or consumers. It could potentially use these powers if algorithmic pricing systems were thought to be skewing markets or reducing competition. The regime includes powers to impose structural or behavioural remedies to remove the distortion whether the underlying conduct is illegal or not. The potential impact of the UK market regime should not be underestimated – it has been used to restructure whole markets in the past. This is not a liability framework, and there is no scope for private enforcement using these rules, but the markets regime could enable the CMA to slice through the Gordian knot of liability for unforeseeable collusion by AI agents.
“Brussels asks not, “did you agree?” but “did you knowingly plug into a mechanism that aligned your conduct with rivals?””
FW: How significant is the role of third party data providers or pricing platforms in shaping current enforcement priorities?
Forrest: A significant number of cases in the US involving algorithmic pricing revolve around third party data providers or pricing platforms. US agencies and state enforcers and the private bar have honed in on industries in which competitors subscribe to common providers and platforms, and we expect these to be a significant source of litigation going forward.
Kar: Third parties are essentially the hub in the ‘hub-and-spoke’ analysis of price fixing, and it is well established that liability can attach to the hub and spokes alike – this is understood to be the basis of the CMA’s current investigation into hotel pricing. Using a vendor, a trade association or an algorithm provider does not insulate you – if anything, delegating pricing to a third party fed by rivals’ data raises the risk.
“All agencies are focused on the fact that technology cannot create a safe space for collusion and harmful price-fixing.”
FW: How are regulators using data analytics and economic modelling to identify and evidence coordinated effects in algorithm-driven markets?
O’Brien: The UK government has recently proposed to extend powers to review algorithms, created for vertical digital markets regulation, across the UK competition and consumer regimes – legislation is awaited. Although the CMA already can and does request information about algorithms, the proposed new powers will enable it to look under the hood of an algorithm, such as by requiring a demo of it in operation. The expanded powers will address perceived insufficiencies in written information to enable the CMA to understand the role of an algorithm properly. Businesses need to make sure they can explain their model before they are asked to. In addition, we know the CMA has built AI-driven capability to scan markets for algorithmic collusion, with a particular focus on bid-rigging, so it is deploying its investment in technical capability to fight fire with fire.
Morch: The European Commission (EC) has confirmed that it launched the tyres cartel investigation off the back of a large-scale AI review of earnings calls transcripts which surfaced what it is investigating as alleged price signalling through public statements. In addition, the EC is planning a study on modern pricing and the risks of collusion and has recently appointed a chief technology officer for its competition enforcement directorate, with a role both to enhance technical expertise in complex cases and to boost the EC’s use of digital forensics and intelligence. Businesses can expect granular information requests and deeper engagement with expert economic and technical evidence.
Forrest: US agencies are scrutinising algorithmic pricing through existing antitrust enforcement rather than a new standalone algorithm-inspection regime. The DOJ and the Federal Trade Commission have used investigations, litigation and statements of interest to develop theories based on common pricing tools, pooled non-public competitor data, high adherence rates, constrained overrides and delegation of pricing decisions to a shared platform. The RealPage settlement shows the technical direction of travel: agencies are focusing on runtime data use, model-training inputs, auto-accept defaults, asymmetric guardrails, and access to code, runtime logic and model materials. Businesses should therefore expect US agencies and state enforcers to ask not only what the algorithm is designed to do, but what data it uses, how recommendations are implemented or overridden, and whether prices, margins, output or adoption patterns evidence coordinated effects.
“Companies should think about several data-hygiene principles.”
FW: What are the practical challenges around transparency and explainability when assessing whether an algorithmic system has produced anti-competitive outcomes?
St. Matthew-Daniel: Enforcers and courts are increasingly scrutinising how algorithms work, including the code behind them. In the last several years, the DOJ has been hiring and recruiting technologists, data scientists and other technical experts to strengthen its ability to assess AI-related conduct. Courts are moving in the same direction. In Mach v. Yardi Systems, for example, a California state court permitted early discovery into how Yardi’s rental-pricing software used competitor data, and the subsequent decision granting summary judgment to Yardi treated Yardi’s early production of source code and related evidence as ‘critical’ to resolving whether the software used confidential or non-public, rental-price information to generate recommendations. This creates an opportunity for companies to rebut an algorithmic-collusion theory through proactive disclosure of source code and other technical evidence. But, without auditable code, data-flow records, recommendation logs, override evidence and plain-English explanations, it may be difficult to show that an algorithm does not raise antitrust concerns.
FW: Where are we seeing the greatest divergence across jurisdictions in approaches to algorithmic collusion, and what is driving those differences?
Kar: All agencies are focused on the fact that technology cannot create a safe space for collusion and harmful price-fixing. Any ‘divergence’ is better understood as differences in law and precedent and in the extent of private and public enforcement in the different jurisdictions. One sharp difference is criminal exposure. The DOJ has described algorithmic coordination as “old crime, new code” and warned that using a common algorithm, software as a service platform or AI tool to replace independent pricing can lead to criminal prosecution. There is no criminal enforcement at EU level. The UK has its criminal cartel offence, which was used in relation to an early algorithmic pricing case, but criminal enforcement by the CMA is now rare.
St. Matthew-Daniel: The US is increasingly fragmented. Federal case outcomes have been fact-specific, and courts are divided over whether algorithmic collusion is per se unlawful or requires a broader assessment of market effects. At the same time, states and localities have begun adopting targeted rules where they view the federal framework as insufficient. California, for example, has amended its antitrust law to prohibit common algorithms that ‘coerce’ users to adopt recommendations, as well as other uses of common algorithms to collude, while also lowering the pleading standard. New York, Connecticut and various cities, towns and counties have adopted varying levels of restrictions on common algorithms in residential housing, and Maryland has banned personalised grocery pricing outright. For US-facing businesses, this creates a patchwork of theories, thresholds and disclosure duties. The practical compliance baseline is likely to be the most aggressive applicable regime, not the most permissive.
“The clearest message is that businesses are fully accountable for what their algorithms do.”
FW: What are the key implications of these trends for how organisations design, govern and oversee automated decision-making systems?
O’Brien: The clearest message is that businesses are fully accountable for what their algorithms do and cannot offload that liability onto a tool or a vendor. For a board, that reframes pricing algorithms from an operational IT matter into a governance one – someone at the table needs to own the risk, understand at a high level what the firm’s pricing systems take in and put out, and ensure legal is consulted when onboarding a new tool. The reassuring part is that this is familiar territory, as it is essentially the same oversight discipline that boards already apply to any high-stakes legal exposure. The unfamiliar part is the speed, and the technical distance between the boardroom and the model. Closing that gap, through clear lines of accountability and a standing route for engineers and data scientists to escalate concerns, is now squarely a governance responsibility, not a technical afterthought.
Forrest: Companies should think about several data-hygiene principles. First, as to inputs, if the tool is shared with competitors, using publicly available, as opposed to non-public, data carries less antitrust risk. Second, ring-fence vendor data. Companies can further reduce risk by using a third-party tool that keeps each customer’s data genuinely separate. Third, there is the cross-portfolio trap – for a group or an investor, multiple portfolio companies in the same sector quietly using the same tool or data provider can create enforcement risk with no direct contact between them at all. Companies should bear these principles in mind early and throughout the process of licensing or designing a pricing algorithm. Additionally, safeguards such as auditing rights and commitments to only using public data and anonymisation and data-ageing should be written into any licensing agreements.
Kar: Every tool needs to be carefully analysed and evaluated for legal risk before deployment. Management must understand who owns the deployment of new tools and ensure that there is full alignment between the business and legal. Companies must understand the details of any new tool – what are the inputs and outputs, how is data protected, and what guardrails are in place to prevent collusion and preserve independent pricing decision making? Operational compliance programmes focused on these issues will be best positioned to defend the legality and pro-competitive uses of AI and algorithmic tools.
Nicole Kar is global co-head of the antitrust department, advising on global merger investigations and foreign investment screening, as well as cartel, abuse of dominance and consumer law matters. She has led over 40 merger reviews before UK and European authorities and is recognised as a leading foreign investment screening adviser, ranked Band 1 by Chambers UK and named a “Hall of Fame” lawyer by Legal 500 UK. She can be contacted on +44 (0)20 7601 8657 or by email: nkar@paulweiss.com.
A partner in the antitrust department, Lauren O’Brien is a sought-after adviser on complex antitrust, merger control, foreign direct investment and consumer law matters, counselling clients in technology, asset management, life sciences, consumer goods and industrial sectors on large-scale transactions. She completed CMA secondments in 2018 and 2021, is admitted in Brussels and England & Wales, and was recently recognised by Legal 500 UK. She can be contacted on +44 (0)20 7601 8648 or by email: lobrien@paulweiss.com.
A partner in the litigation department, Katherine Forrest serves as vice chair of the firm and co-chair of its global artificial intelligence group. A former US district judge for the Southern District of New York and deputy assistant attorney general at the Department of Justice’s antitrust division, she now handles sensitive antitrust, artificial intelligence and intellectual property matters, and is recognised by Chambers USA as an “Eminent Practitioner” in antitrust. She can be contacted on +1 (212) 373 3195 or by email: kforrest@paulweiss.com.
Deputy chair of the antitrust department, Eyitayo ‘Tee’ St. Matthew-Daniel advises clients on antitrust issues worldwide, drawing on her experience within the senior ranks of the DOJ’s antitrust division and in private practice. Dual-qualified in New York and England & Wales, she has secured clearances from the DOJ, Federal Trade Commission, European Commission and other authorities, defending clients across North America, Europe, Africa and Asia Pacific. She can be contacted on +1 (212) 373 3229 or by email: tstmatthewdaniel@paulweiss.com.
A partner in the antitrust department, Henrik Morch advises clients on EU competition law, including merger control, cartel investigations and foreign direct investment reviews. He joined the firm after more than three decades at the European Commission, most recently serving over eight years as director of transport, post and other services at DG Comp, overseeing antitrust enforcement and state aid matters in those sectors. He can be contacted on +32 2 884 0802 or by email: hmorch@paulweiss.com.
© Financier Worldwide
THE PANELLISTS
Paul, Weiss, Rifkind, Wharton & Garrison LLP
Q&A: Algorithmic collusion: antitrust enforcement
Cartel hunt: how enforcers use whistleblowers and AI to detect conspiracies
Competition regulation and enforcement trends in the EU’s digital economy
Canada tests limits with increased use of production orders
What national security review can learn from merger control