Competition Law and AI-Assisted Antitrust Investigations

1. Introduction

Artificial intelligence is increasingly becoming a tool for competition authorities to identify, investigate, and analyse possible antitrust violations. AI can process very large datasets—such as transaction records, prices, emails, bidding data, platform activity, and communications—much faster than conventional manual investigation.

The OECD has specifically noted that AI can improve cartel-detection tools used to initiate ex officio investigations, while also emphasizing that authorities should use multiple complementary detection methods rather than rely on a single technological system.

At the same time, AI creates a second problem: companies themselves may use algorithms or AI systems to coordinate prices, discriminate between customers, exclude competitors, or facilitate information exchange. Therefore, an antitrust authority may need to investigate both AI-assisted conduct and AI-assisted evidence.

2. Meaning of AI-Assisted Antitrust Investigation

An AI-assisted antitrust investigation is an investigation in which artificial intelligence or advanced computational techniques assist a competition authority in:

1. detecting suspicious conduct;
2. identifying possible cartels;
3. analysing pricing patterns;
4. examining large volumes of documents;
5. identifying relationships between companies;
6. analysing communications;
7. detecting unusual bidding patterns;
8. reconstructing market behaviour;
9. identifying potentially exclusionary conduct; and
10. testing economic theories of harm.

Importantly, AI normally functions as an investigative and analytical tool, rather than replacing the legal decision-maker.

The OECD has observed that competition authorities are developing algorithmic-auditing and explainable-AI capabilities and increasingly need technical expertise to understand algorithmic conduct.

3. Why AI Is Important for Antitrust Enforcement

Traditional antitrust investigations can involve enormous quantities of information.

For example, an investigation may contain:

• millions of emails;
• years of pricing information;
• thousands of contracts;
• bidding records;
• internal company documents;
• source code;
• pricing algorithms;
• customer data;
• transaction records;
• platform logs; and
• communications between competitors.

Manually examining all of this information can be extremely difficult.

AI can help investigators identify patterns that deserve further human investigation.

For example:

10 competing companies may independently change their prices thousands of times every month. An AI system could identify unusual synchronization in those changes and flag the relevant period for investigators.

The AI finding itself would not necessarily prove a cartel. Investigators would still need to establish the legally relevant facts.

4. Major Uses of AI in Antitrust Investigations

A. Cartel Detection

One of the most important applications is cartel screening.

AI can examine:

• identical or near-identical price movements;
• suspicious bid rotation;
• unusual bidding patterns;
• geographic allocation;
• customer allocation;
• repeated winning patterns;
• sudden parallel price increases; and
• communication patterns.

The OECD has specifically discussed AI and other technologies as tools for improving ex-officio cartel detection.

Example

Suppose five construction companies repeatedly participate in public tenders.

An AI system could identify:

• Company A wins Tender 1;
• Company B wins Tender 2;
• Company C wins Tender 3;
• losing companies repeatedly submit unusually high bids;
• the same companies rotate their winning positions.

This does not automatically establish a cartel, but it can provide an investigative lead.

5. B. Algorithmic Pricing Investigation

AI can investigate how businesses determine prices.

Modern businesses may use automated systems that continuously monitor:

• competitors' prices;
• demand;
• inventory;
• customer behaviour;
• market conditions; and
• historical transactions.

Investigators may therefore need to determine:

1. Who designed the algorithm?
2. What instructions were given to it?
3. What data does it receive?
4. Does it receive competitors' information?
5. Who controls its pricing parameters?
6. Did competing businesses use the same software?
7. Did competitors communicate about the system?
8. Did the algorithm implement an existing agreement?

The legal significance depends on the underlying conduct—not merely on the fact that an algorithm was used.

The OECD's work distinguishes legitimate efficiency-enhancing algorithmic pricing from circumstances in which algorithms may facilitate restrictive conduct.

6. C. Document Review

AI can assist investigators in reviewing enormous documentary datasets.

For example, an authority could use machine-learning or language-processing systems to identify documents containing concepts relating to:

• price coordination;
• competitor communications;
• market allocation;
• customer allocation;
• exclusion of competitors;
• strategic agreements;
• common pricing systems; and
• internal discussions concerning competitors.

AI can therefore help investigators prioritise documents for human review.

However, because AI systems can produce false positives and false negatives, important evidence should remain subject to human verification.

7. D. Network Analysis

Antitrust investigations frequently involve relationships among multiple companies.

AI-assisted network analysis can map:

Company A → Distributor → Company B → Software Provider → Competitor

This can help investigators identify:

• common intermediaries;
• common technology providers;
• ownership relationships;
• common directors;
• communication networks;
• common suppliers; and
• possible hub-and-spoke arrangements.

This becomes particularly relevant where several competitors use the same pricing or marketplace technology.

8. E. Detection of Hub-and-Spoke Arrangements

AI can assist in identifying situations where a common intermediary may facilitate coordination between competitors.

For example:

Competitor A
↓
Common Pricing Platform
↓
Competitor B
↓
Competitor C

The investigation would then examine whether the common platform merely provides an ordinary commercial service or whether it facilitates unlawful coordination.

The UK competition authorities have identified common algorithmic systems and third-party pricing systems as potential mechanisms through which information exchange or coordination could occur.

9. F. Merger Investigation

AI can also assist merger review.

Authorities can use computational tools to examine:

• market shares;
• customer switching;
• pricing data;
• geographic markets;
• product similarity;
• internal documents;
• customer complaints;
• innovation indicators; and
• potential competitive overlaps.

AI may therefore help authorities identify whether a transaction requires deeper investigation.

However, the final assessment still requires legal and economic analysis.

10. G. Abuse of Dominance

AI can help investigate potential exclusionary conduct by dominant companies.

Potential areas include:

• self-preferencing;
• discriminatory ranking;
• refusal of access;
• tying;
• bundling;
• loyalty mechanisms;
• discriminatory algorithms;
• exclusionary pricing;
• manipulation of search results; and
• preferential treatment of the dominant firm's products.

The investigator may need to understand the actual operation of the algorithm, not merely its output.

11. H. AI-Assisted Economic Analysis

AI can help process large economic datasets.

Investigators may analyse:

• price elasticity;
• margins;
• demand substitution;
• customer switching;
• geographic patterns;
• bidding behaviour;
• market concentration; and
• effects of a suspected restriction.

The OECD's recent work emphasizes that AI's competitive effects are highly context-dependent and can differ according to firms' capabilities, sector characteristics, and access to enabling inputs.

12. Important Case Laws

There is an important qualification: there are still relatively few judicial decisions directly concerning modern generative-AI antitrust investigations. The existing case law largely concerns algorithms, electronic platforms, automated pricing, and digital evidence. These cases provide the legal foundations for investigating newer AI systems. The OECD likewise notes that actual antitrust cases involving newer generative-AI conduct remain limited.

Case 1 — United States v. David Topkins

Court: U.S. District Court for the Northern District of California
Year: 2015
Subject: Algorithmic price fixing

This is one of the clearest examples of algorithm-assisted cartel conduct.

Topkins and competitors selling posters on Amazon Marketplace agreed to coordinate prices. They used pricing algorithms to implement the agreement.

Topkins pleaded guilty to a Sherman Act price-fixing offence. The U.S. Department of Justice specifically described the use of pricing algorithms to implement the agreement.

Legal significance

The important principle is:

An algorithm does not make an otherwise unlawful agreement lawful.

If competitors first agree to fix prices and then use software to implement that agreement, traditional cartel principles can still apply.

Case 2 — Trod Ltd / GB eye Ltd

Authority: UK Competition and Markets Authority
Year: 2016
Subject: Online price fixing and automated repricing

Two competing online sellers of posters agreed not to undercut one another.

They used automated repricing software to implement the arrangement.

The CMA imposed a fine of more than £160,000.

Legal significance

The case demonstrates how competition authorities can investigate the relationship between:

Human agreement → pricing rules → algorithmic implementation → market prices

It is therefore particularly relevant to AI-assisted antitrust investigations.

Case 3 — Eturas

Case: Eturas UAB and Others v Lietuvos Respublikos konkurencijos taryba
CJEU: Case C-74/14
Year: 2016

This case concerned the E-TURAS online booking system used by travel agencies.

The system's administrator sent communications concerning a reduction in online booking discounts, and the system technically restricted the discounts available to participating agencies. The Lithuanian competition authority investigated the conduct.

Legal significance

The case is important because electronic platforms can become relevant evidence in proving coordination.

The fact that communication occurs through a digital system does not remove it from competition law.

For AI investigations, the principle is particularly useful because investigators may need to examine:

• platform messages;
• system settings;
• algorithmic instructions;
• automated restrictions; and
• user responses.

Case 4 — United States v. Airline Tariff Publishing Company

Court: U.S. District Court
Year: 1994
Subject: Computerised airline fare communication

The DOJ case concerned airline pricing and the use of an electronic fare dissemination system.

The government alleged that airlines used the system in ways that facilitated communication concerning fares and other competitive terms. The matter resulted in consent judgments.

Legal significance

This case predates modern AI but is highly relevant historically.

It demonstrates that competition law can apply where computerised information systems facilitate coordination.

The important investigative lesson is that authorities must examine not only direct communications but also the technological infrastructure through which market information is exchanged.

Case 5 — T-Mobile Netherlands

Case: T-Mobile Netherlands BV and Others v Raad van bestuur van de Nederlandse Mededingingsautoriteit
CJEU: Case C-8/08
Year: 2009
Subject: Concerted practices

The case concerned communications among competing mobile telephone operators concerning dealer remuneration.

The CJEU explained the legal approach to concerted practices and restrictions of competition by object.

Relevance to AI investigations

The case establishes an important background principle:

Competition authorities do not necessarily need to prove a fully developed written contract before investigating coordinated conduct.

In an AI environment, this becomes important because coordination could potentially be evidenced through a combination of:

• communications;
• algorithmic instructions;
• common technical settings;
• pricing behaviour; and
• other circumstantial evidence.

The existence of parallel algorithmic behaviour alone, however, should not automatically be treated as proof of an agreement.

Case 6 — AC-Treuhand

Case: AC-Treuhand AG v European Commission
CJEU: Case C-194/14 P
Year: 2015

AC-Treuhand concerned the role of a consultancy/association-type intermediary in cartel arrangements.

The CJEU confirmed that an undertaking that actively contributes to implementing or facilitating an anticompetitive agreement can potentially fall within Article 101 TFEU even if it is not itself operating in the same product market as the cartel participants.

Relevance to AI

This principle is particularly important for AI systems because a technology provider may sit between competing businesses.

For example:

Competitor A
↘
AI/pricing intermediary
↗
Competitor B

The legal question is not simply whether the intermediary is an AI company.

Investigators would need to examine its actual role in the alleged coordination.

Case 7 — Amazon Marketplace Investigation

Authority: UK OFT/CMA
Investigation: Amazon Marketplace
Subject: Price-parity arrangements

The UK's Office of Fair Trading investigated Amazon's Marketplace price-parity policy. Amazon subsequently ended the policy, and the investigation was closed on administrative-priority grounds without a finding of infringement.

Relevance

This demonstrates how platform rules can influence competition between marketplaces.

Modern AI investigations can go further by examining automated mechanisms that:

• monitor competitor prices;
• identify deviations;
• alter rankings;
• remove offers; or
• affect eligibility for platform features.

The important point is that automated enforcement mechanisms can themselves become relevant evidence.

13. Case Comparison

14. AI as an Investigative Tool vs AI as the Conduct

This distinction is extremely important.

Situation 1 — AI used by the authority

The competition authority uses AI to:

• search documents;
• identify suspicious pricing;
• detect cartel patterns;
• analyse transactions;
• classify evidence.

Here, AI is the investigative tool.

Situation 2 — AI used by the investigated company

A company uses AI to:

• set prices;
• rank competitors;
• allocate customers;
• recommend prices;
• monitor rivals.

Here, AI is part of the potentially investigated conduct.

Situation 3 — AI used by several competitors

Several competitors use the same AI pricing system.

The authority may need to determine:

• whether they independently selected the same software;
• whether the software receives common information;
• whether the provider facilitates coordination;
• whether competitors communicated with one another;
• whether the system automatically responds to rivals; and
• whether there is evidence of an agreement or concerted practice.

This is one of the most difficult areas of future antitrust enforcement.

15. Autonomous AI Collusion

A particularly difficult question is whether AI systems can independently develop pricing behaviour that results in higher prices without humans expressly agreeing to coordinate.

For example:

Firm A's AI → observes Firm B's price
↓
changes its own price

Firm B's AI → observes Firm A's price
↓
changes its own price

Over time, both systems may learn that maintaining higher prices produces greater returns.

This raises a fundamental legal question:

Can competition law establish an unlawful agreement when there is no conventional human communication or agreement?

There is currently no simple universal answer.

The CMA and OECD both recognize autonomous algorithmic/AI collusion as an emerging competition-law issue.

16. Evidence Problems in AI Investigations

AI creates several evidentiary difficulties.

1. Explainability

An AI model may produce a result without providing a legally satisfactory explanation of how it reached that result.

2. Attribution

Investigators must determine:

Who actually made the decision?

Was it:

• a director;
• an employee;
• a software developer;
• an AI provider;
• an automated system; or
• a combination of these?

3. False positives

An AI system may identify parallel pricing that is actually caused by:

• common costs;
• supply shortages;
• demand shocks;
• common market information; or
• legitimate competitive reactions.

4. False negatives

AI may fail to detect sophisticated coordination.

5. Data quality

Poor or incomplete data can produce misleading results.

6. Confidentiality

Antitrust investigations frequently contain highly sensitive business information. AI systems therefore create additional data-governance and confidentiality concerns.

17. Due Process and Procedural Safeguards

AI-assisted enforcement should remain subject to ordinary legal safeguards.

Authorities should consider:

• reliability of the model;
• quality of training data;
• reproducibility of results;
• audit trails;
• human verification;
• confidentiality;
• procedural fairness;
• disclosure obligations;
• protection of legally privileged material; and
• the investigated party's ability to challenge evidence.

An AI-generated probability or risk score should generally be treated as an investigative lead, not automatically as proof of an infringement.

18. Human Oversight

Human oversight remains essential.

A sensible enforcement structure is:

AI screening
↓
Human investigation
↓
Economic analysis
↓
Legal assessment
↓
Collection and verification of evidence
↓
Procedural safeguards
↓
Final decision by authorised officials/court

This prevents an automated model from effectively becoming the decision-maker.

The OECD has similarly emphasized that the appropriate investigative technique depends on the particular case and that sophisticated technical approaches are not always necessary.

19. Relationship with Traditional Competition Law

AI does not automatically require a completely new body of antitrust law.

Existing competition principles can still address many forms of conduct.

Article 101 TFEU / cartel rules

Relevant to:

• price fixing;
• market allocation;
• information exchange;
• coordinated conduct.

Article 102 TFEU / abuse of dominance

Relevant to:

• exclusionary algorithms;
• discriminatory access;
• self-preferencing;
• tying;
• exploitative conduct.

Merger control

Relevant to:

• acquisitions of AI companies;
• concentration of datasets;
• control over computing infrastructure;
• vertical integration across AI markets.

National competition laws

Domestic antitrust statutes may similarly apply to algorithmic conduct depending on the jurisdiction.

20. Future Direction of AI-Assisted Antitrust Enforcement

Future investigations are likely to involve increasing use of:

• machine-learning cartel screening;
• natural-language processing;
• automated document review;
• graph/network analysis;
• anomaly detection;
• algorithmic auditing;
• synthetic-data testing;
• pricing simulations;
• source-code analysis;
• agentic AI systems; and
• continuous market monitoring.

The CMA has stated that it is developing technical capabilities involving AI and agentic systems for detecting potential consumer- and competition-law breaches.

The OECD's 2026 work also identifies continuing structural competition concerns across AI markets, including concentration at important layers of the AI value chain.

21. Key Legal Principles

The main principles can be summarised as follows:

1. AI does not create an exemption from competition law.
2. Traditional cartel rules can apply where algorithms implement human agreements.
3. Digital communications can constitute important evidence of coordination.
4. Third-party technology providers may become relevant where they facilitate coordination.
5. Parallel algorithmic behaviour is not automatically proof of an unlawful agreement.
6. AI-generated investigative leads require human verification.
7. Authorities must distinguish legitimate algorithmic efficiencies from anticompetitive conduct.
8. Explainability and attribution are major evidentiary challenges.
9. AI can be used both to detect cartels and to investigate algorithmic exclusion.
10. Existing competition-law concepts remain highly relevant, but autonomous AI behaviour creates unresolved legal questions.

22. Conclusion

AI-assisted antitrust investigation represents an evolution in enforcement rather than a complete replacement of traditional competition law.

Competition authorities can use AI to process huge datasets, detect suspicious pricing, identify cartel patterns, analyse communications, examine algorithms, and prioritise investigations. At the same time, companies may themselves use AI to implement or facilitate anticompetitive conduct.

The existing cases—particularly Topkins, Trod/GB eye, Eturas, Airline Tariff Publishing, T-Mobile Netherlands, and AC-Treuhand—show important principles for dealing with technologically mediated coordination. However, modern generative and agentic AI raises newer questions concerning autonomous decision-making, explainability, attribution, and proof of agreement.