AI-Driven Transformation of Antitrust Enforcement Institutions

Introduction

Artificial intelligence is changing antitrust enforcement from a predominantly human-led, retrospective and case-specific model toward a potentially continuous, data-driven and technologically assisted model.

Traditional competition authorities generally investigate after receiving complaints, market information, leniency applications, merger notifications or other evidence. AI can allow authorities to identify suspicious patterns much earlier by processing enormous datasets involving prices, transactions, rankings, algorithms, procurement records, corporate ownership, communications and platform behaviour.

The transformation therefore concerns not merely the use of AI as an investigative tool, but the institutional architecture of antitrust enforcement itself:

From complaint-driven enforcement → continuous market surveillance → algorithmic detection → human investigation → technologically informed remedies.

The OECD has specifically identified AI-enabled algorithmic pricing as creating new enforcement challenges for competition authorities, while emphasizing the need for authorities to adapt their investigative and analytical capabilities.

I. Meaning of AI-Driven Transformation of Antitrust Institutions

AI-driven transformation means the integration of:

• machine learning;
• natural-language processing;
• graph analytics;
• anomaly detection;
• automated document review;
• algorithm auditing;
• predictive analytics;
• network analysis;
• data scraping;
• digital forensics;
• large language models;
• automated merger screening; and
• continuous market monitoring

into the work of competition authorities.

The institutional transformation can occur at five levels:

II. Why Antitrust Institutions Need Transformation

AI changes the competitive environment in several ways.

1. Speed

Algorithms can alter prices, rankings and allocation decisions thousands or millions of times.

A conventional investigation may therefore examine conduct after the competitive environment has already changed.

2. Scale

A digital platform can process:

• millions of transactions;
• billions of search results;
• consumer-level data;
• real-time prices;
• supplier information; and
• algorithmic recommendations.

Human investigators cannot manually examine such volumes.

3. Opacity

Modern AI systems can make decisions through complex models whose internal logic is difficult to reconstruct.

This creates an evidentiary problem:

How does an authority prove an antitrust violation when the relevant decision is produced by an algorithm rather than an easily identifiable human instruction?

4. Continuous adaptation

AI systems learn from market information and can modify their behaviour.

Consequently, an authority may have to investigate not merely:

“What did the company decide?”

but:

“How was the decision-making system designed, trained, updated and constrained?”

III. Transformation of Institutional Functions

A. AI-Based Antitrust Intelligence

Competition authorities can establish permanent AI-based monitoring systems.

Such systems could monitor:

• prices;
• discounts;
• margins;
• market shares;
• bidding patterns;
• supplier switching;
• customer allocation;
• algorithmic recommendations;
• platform rankings;
• exclusivity arrangements;
• merger activity;
• ownership networks; and
• interlocking directorates.

The objective is early identification of unusual competitive patterns, rather than waiting for a formal complaint.

China's 2026 internet-platform antitrust compliance guidance expressly recognizes algorithmic coordination, dynamic pricing, traffic allocation and algorithmic monitoring as competition-law risks.

IV. AI and Cartel Detection

AI can transform cartel enforcement.

Traditional cartel detection relies heavily on:

• whistleblowers;
• leniency;
• emails;
• meetings;
• suspicious communications;
• bidding patterns.

AI can supplement these mechanisms by identifying statistical patterns such as:

• unusually parallel prices;
• identical price movements;
• synchronized bidding;
• suspicious market allocation;
• repeated bid rotation;
• abnormal margins;
• coordinated capacity reductions.

Important limitation

Parallel algorithmic behaviour is not automatically a cartel.

Competition authorities must distinguish:

1. conscious agreement;
2. facilitated coordination;
3. unilateral algorithmic adaptation;
4. tacit coordination;
5. legitimate parallel conduct.

This distinction is especially important because AI may independently produce similar outputs from similar market information.

V. AI and Hub-and-Spoke Coordination

AI can create new versions of hub-and-spoke structures.

A platform may sit between competing suppliers and possess information concerning:

• prices;
• inventories;
• demand;
• discounts;
• customers;
• capacity.

The platform's algorithm may then recommend prices to competing businesses.

The institutional question becomes:

When does technological facilitation become legally relevant coordination?

The Indian case of Samir Agrawal v. Competition Commission of India is particularly important.

The allegation concerned algorithmic pricing by Ola and Uber and the possibility that the platforms could operate as a hub coordinating drivers. The NCLAT ultimately rejected the allegation on the facts, emphasizing the absence of the necessary agreement or concerted action.

Institutional lesson

AI enforcement cannot simply equate:

algorithm + similar prices = cartel.

Authorities need evidence concerning:

• communication;
• information exchange;
• contractual arrangements;
• algorithmic design;
• knowledge;
• implementation;
• control; and
• competitive effects.

VI. AI and Digital Evidence

AI dramatically expands the volume of evidence available to authorities.

Evidence may include:

• source code;
• model documentation;
• API records;
• training data;
• prompts;
• logs;
• model updates;
• pricing outputs;
• recommendation histories;
• internal communications;
• cloud records;
• metadata;
• dashboards;
• A/B testing results.

Competition institutions therefore require specialized digital-forensics divisions.

The traditional investigator may need to work alongside:

• data scientists;
• software engineers;
• algorithm auditors;
• economists;
• cybersecurity specialists;
• statisticians; and
• AI governance specialists.

VII. AI-Based Merger Screening

AI can transform merger control.

Authorities can construct databases containing:

• corporate ownership;
• subsidiaries;
• investments;
• minority shareholdings;
• patents;
• datasets;
• cloud infrastructure;
• AI models;
• computing capacity;
• key employees;
• strategic partnerships.

AI can identify relationships that might not be obvious from a conventional merger notification.

This is particularly relevant to AI markets because competitive power may arise from control over:

chips + compute + cloud + data + models + distribution + applications.

The OECD has identified AI infrastructure, including advanced computing resources and chips, as an area requiring close competition monitoring and a combination of enforcement and advocacy tools.

VIII. AI and Market Definition

Traditional market definition frequently relies upon:

• SSNIP analysis;
• substitution;
• consumer surveys;
• price data;
• economic modelling.

AI can supplement these methods through:

• real-time demand analysis;
• consumer switching data;
• search behaviour;
• transaction-level substitution patterns;
• product similarity analysis;
• natural-language analysis of consumer preferences.

However, AI should not automatically replace economic judgment.

Market definition remains a legal-economic exercise requiring interpretation of competitive constraints.

IX. AI and Abuse of Dominance

AI may help authorities detect:

Self-preferencing

A dominant platform may rank its own products differently.

Discriminatory algorithms

Different businesses may receive different visibility or access.

Predatory pricing

AI can identify sustained below-cost pricing patterns.

Excessive pricing

Automated analysis can compare pricing across markets.

Refusal of access

AI can identify systematic denial of API, data or platform access.

Tying and bundling

AI can identify relationships between services and purchasing behaviour.

The European Commission's DMA enforcement illustrates the movement toward continuous digital-platform supervision. In July 2026, the Commission announced findings concerning Google's self-preferencing in Search and restrictions on steering in Google Play, imposing fines totaling €890 million.

X. AI and Algorithmic Self-Preferencing

AI can make self-preferencing more sophisticated.

Instead of explicitly stating:

“Rank our product first,”

a platform might optimize a recommendation system around objectives that systematically favour affiliated services.

The enforcement challenge therefore shifts from examining a written rule to examining:

• model objectives;
• training data;
• ranking criteria;
• optimization functions;
• feedback loops;
• performance metrics.

The EU's DMA framework has institutionalized certain obligations for gatekeepers, including requirements concerning fair and non-discriminatory treatment in ranking.

XI. AI and Data as a Source of Market Power

Data can become an antitrust asset.

The Bundeskartellamt's Facebook case is an important precedent.

The authority found that Facebook's combination of data from Facebook, Instagram, WhatsApp and third-party sources raised abuse-of-dominance concerns and imposed restrictions on combining such data without voluntary consent.

Institutional significance

Competition authorities increasingly need expertise in the interaction between:

data → AI capability → consumer targeting → advertising → market power.

The enforcement institution consequently becomes partly a data-governance institution.

XII. AI and Automated Regulatory Surveillance

The future competition authority may operate a permanent Competition Intelligence Platform.

It could continuously monitor:

Market Data     ↓ AI Screening     ↓ Anomaly Detection     ↓ Risk Classification     ↓ Human Review     ↓ Formal Investigation     ↓ Economic Analysis     ↓ Enforcement / Remedy     ↓ Continuous Monitoring

The crucial safeguard is that AI should generally operate as a detection and analytical mechanism, not as an autonomous adjudicator.

XIII. Transformation of Investigative Institutions

Traditional model

Complaint   ↓ Preliminary inquiry   ↓ Investigation   ↓ Economic analysis   ↓ Decision

AI-enabled model

Continuous market data        ↓ AI monitoring        ↓ Anomaly identification        ↓ Risk scoring        ↓ Human verification        ↓ Digital investigation        ↓ Economic analysis        ↓ Legal assessment        ↓ Enforcement        ↓ Algorithmic compliance monitoring

The authority consequently becomes continuously present in the market, rather than appearing only after a suspected infringement has occurred.

XIV. Six Important Case Laws

1. Eturas UAB v Lietuvos Respublikos konkurencijos taryba

CJEU, Case C-74/14 (2016)

Travel agencies used a common computerized booking system. The system administrator sent a message that resulted in an automatic restriction on discounts.

The CJEU considered whether conduct implemented through a common computer system could constitute a concerted practice and addressed the evidentiary implications of such automated coordination.

Significance

The case demonstrates that competition law must be capable of analysing technology-mediated coordination, not merely physical meetings.

2. AC-Treuhand AG v European Commission

CJEU, Case C-194/14 P (2015)

AC-Treuhand provided services facilitating cartel activity, including organizing meetings and handling commercially sensitive information.

The CJEU confirmed that an undertaking providing knowing and essential assistance to an anticompetitive arrangement can fall within the competition-law prohibition even though it is not itself a competitor in the cartelized market.

AI relevance

An AI platform or intermediary that knowingly facilitates coordination could raise analogous questions concerning:

• technological facilitation;
• data aggregation;
• algorithmic coordination;
• monitoring;
• automated implementation.

3. Samir Agrawal v Competition Commission of India

NCLAT, 2020

The case concerned allegations that Ola and Uber's algorithms facilitated price fixing among drivers.

The NCLAT rejected the hub-and-spoke allegation on the facts, emphasizing the absence of an agreement or meeting of minds necessary for establishing the alleged infringement.

AI relevance

This is particularly important for AI enforcement because it demonstrates the distinction between:

algorithmically determined prices
and
algorithmically facilitated collusion.

4. Bundeskartellamt – Facebook/Meta Data Case

Germany, 2019

The German competition authority restricted Facebook's combination of user data from different sources, treating the conduct within the framework of abuse of dominance.

AI relevance

AI increases the economic value of aggregated data. Therefore, enforcement institutions need to investigate not only price but also:

• data accumulation;
• interoperability;
• data combination;
• data portability;
• access to datasets.

5. United States v. RealPage Inc.

U.S. Department of Justice, 2024

The DOJ alleged that RealPage's algorithmic pricing system facilitated coordination among competing landlords by using competitively sensitive information supplied by participating landlords.

The case alleged violations of Sections 1 and 2 of the Sherman Act.

Institutional significance

RealPage illustrates a major change in enforcement:

The algorithm itself can become a central object of antitrust investigation.

Authorities therefore need technical capabilities to reconstruct:

• input data;
• model architecture;
• pricing recommendations;
• information flows;
• customer participation;
• algorithmic outputs.

6. Google Search / DMA Enforcement

The European Commission's Google Search proceedings under the DMA demonstrate another institutional transformation: enforcement can move from traditional ex-post antitrust litigation toward continuous obligations and regulatory supervision.

In 2026, the Commission adopted measures requiring Google to facilitate effective sharing of anonymised Search data with eligible search competitors, including AI chatbots offering search functionalities.

Significance

The institutional model becomes:

designation → obligation → monitoring → technical specification → compliance decision → continuing supervision.

This is materially different from a traditional one-off infringement proceeding.

XV. China's Emerging Institutional Model

China provides an especially clear example of institutional adaptation.

The 2026 SAMR Internet Platform Antitrust Compliance Guidelines expressly address:

• algorithmic collusion;
• dynamic pricing;
• algorithmic traffic allocation;
• ranking;
• AI-assisted coordination;
• discriminatory algorithms;
• algorithmic screening;
• continuous monitoring;
• audit records;
• algorithm transparency and explainability. 

The guidelines also encourage platforms to conduct targeted screening of pricing algorithms, recommendation systems, ranking logic and advertising strategies and to maintain audit records.

China's broader regulatory approach also emphasizes improved digital enforcement capabilities, including online discovery, case processing and digital regulatory infrastructure.

XVI. Institutional Problems Created by AI

1. False positives

AI may identify parallel behaviour that is entirely lawful.

Therefore:

Anomaly ≠ infringement.

2. False negatives

Sophisticated firms may deliberately design algorithms to conceal coordination.

An AI system trained on known cartel patterns could fail to recognize a new type of conduct.

3. Explainability

An authority must be able to explain why an investigation was initiated.

An opaque AI model creates procedural concerns if investigators cannot explain its reasoning.

4. Due process

Competition law enforcement involves potentially severe consequences:

• fines;
• structural remedies;
• behavioural restrictions;
• divestitures;
• damages;
• reputational consequences.

Therefore, automated detection cannot substitute for procedural safeguards.

5. Confidentiality

Competition authorities possess extremely sensitive information.

AI systems may process:

• trade secrets;
• source code;
• pricing information;
• merger documents;
• customer data.

Strong information-security controls are consequently essential.

XVII. Institutional Separation of AI Functions

A sophisticated competition authority should separate:

1. Detection AI

Identifies suspicious conduct.

2. Investigation AI

Organizes evidence and documents.

3. Economic AI

Performs statistical and econometric analysis.

4. Legal research AI

Assists investigators in locating relevant authorities.

5. Compliance AI

Monitors implementation of remedies.

6. Human adjudication

Makes legally accountable decisions.

This separation helps prevent:

algorithmic detection → automatic accusation → automatic punishment.

XVIII. AI and Leniency Programs

AI could also transform cartel leniency.

Authorities could identify likely cartel structures before receiving a leniency application.

This creates a potential institutional tension:

If firms know that authorities have powerful automated detection systems, the incentive structure surrounding voluntary disclosure may change.

Authorities therefore need to reconsider:

• leniency incentives;
• whistleblower protection;
• automated detection thresholds;
• cooperation mechanisms.

XIX. AI and Dawn Raids

Digital dawn raids increasingly require investigators to understand:

• cloud computing;
• encrypted messaging;
• enterprise software;
• source-code repositories;
• AI development environments;
• model logs;
• decentralized storage.

The modern dawn raid may therefore involve algorithmic evidence preservation, not merely seizure of paper or ordinary electronic files.

XX. AI and Merger Remedies

AI can also help authorities monitor whether companies comply with remedies.

For example, after a divestiture or access remedy, AI could continuously examine:

• API availability;
• ranking;
• access times;
• pricing;
• discrimination;
• customer switching;
• data access;
• interoperability.

This turns remedies from:

one-time obligations

into potentially:

continuously monitored obligations.

XXI. Need for Specialized Competition Institutions

AI-driven antitrust enforcement is likely to require institutional specialization.

A modern authority may need:

Competition lawyers
↓
Economists
↓
Data scientists
↓
AI engineers
↓
Cyber-forensics specialists
↓
Algorithm auditors
↓
Sector specialists

The competition authority consequently becomes a multidisciplinary institution rather than a predominantly legal-economic organization.

XXII. Human Oversight as the Central Principle

The most important institutional principle is:

AI should increase the investigative capacity of competition authorities without transferring ultimate legal responsibility to an algorithm.

Human officials should retain responsibility for:

• defining the legal theory;
• deciding whether evidence is sufficient;
• interpreting economic evidence;
• assessing intent where legally relevant;
• determining causation;
• evaluating efficiencies;
• selecting remedies;
• ensuring procedural fairness.

AI should primarily provide:

scale + speed + pattern recognition + evidence organization.

XXIII. Future Model of an AI-Enabled Competition Authority

A possible institutional structure is:

                    COMPETITION AUTHORITY                           │        ┌──────────────────┼──────────────────┐        │                  │                  │   AI Market Lab     Digital Forensics   Merger Analytics        │                  │                  │  Price monitoring    Source-code         Ownership/  Market shares       analysis             investment  Algorithms          Data extraction      networks        │                  │                  │        └──────────────────┼──────────────────┘                           │                    Human Investigators                           │                    Competition Lawyers                           │                      Chief Economist                           │                  Decision / Adjudication                           │                 AI Compliance Monitoring

XXIV. Key Legal Principles

The transformation should be governed by several principles:

1. Human accountability

A human authority must remain legally responsible.

2. Explainability

AI-generated investigative findings should be capable of meaningful explanation.

3. Auditability

Authorities should retain records of data, models and analytical processes.

4. Proportionality

AI surveillance should correspond to legitimate competition-enforcement objectives.

5. Data minimization

Authorities should avoid unnecessary collection of personal information.

6. Non-discrimination

AI tools should not systematically produce discriminatory investigative outcomes.

7. Reproducibility

Important analytical conclusions should be capable of independent verification.

8. Evidentiary validation

AI-generated findings should be independently corroborated before enforcement action.

XXV. Overall Legal Significance

AI-driven institutional transformation changes antitrust enforcement along three dimensions.

First: From reactive to proactive enforcement

Authorities can identify emerging risks before they become fully developed infringements.

Second: From transaction evidence to system evidence

The investigation may focus on the architecture of the algorithm, rather than merely a contract or email.

Third: From episodic enforcement to continuous supervision

Digital-market regulation increasingly permits authorities to monitor compliance continuously, as illustrated by the EU's DMA framework and China's algorithm-focused platform guidance.

Conclusion

AI-driven transformation of antitrust enforcement institutions represents a shift from traditional case-by-case enforcement toward technologically enabled, continuous and data-intensive competition governance.

The central institutional challenge is not simply teaching competition authorities to use AI. It is redesigning the authority so that it can understand markets in which algorithms themselves determine prices, rankings, access, allocation and competitive strategy.

The jurisprudence of Eturas, AC-Treuhand, Samir Agrawal, Facebook, RealPage and modern Google/DMA proceedings demonstrates the progressive movement from traditional human coordination toward technologically mediated competition concerns. The cases also show why AI detection cannot itself establish liability: legal concepts such as agreement, concerted practice, dominance, causation and competitive harm still require rigorous human and economic assessment.

Accordingly, the emerging institutional model is best understood as:

AI-assisted detection + digital investigation + advanced economics + human legal judgment + continuous compliance monitoring.