Competition Law and Institutional Redesign for AI-Driven Competition Enforcement

1. Introduction

AI-driven competition enforcement refers to the use of artificial intelligence, machine learning, natural-language processing, network analysis, anomaly detection and automated data-processing tools by competition authorities to detect, investigate and analyse potentially anticompetitive conduct.

Institutional redesign goes further. It concerns changing the organisation, procedures, powers, expertise and safeguards of competition authorities so that they can effectively regulate markets in which AI systems themselves influence:

prices;

output;

market allocation;

advertising;

search rankings;

procurement;

consumer targeting;

investment;

innovation;

competitor interaction.

The central question is therefore not simply whether competition authorities should use AI, but whether competition institutions designed for traditional markets remain capable of enforcing competition law in algorithmic and AI-intensive markets.

2. Why Institutional Redesign Is Necessary

Traditional competition enforcement generally relies on:

complaint → investigation → evidence → economic analysis → decision → appeal.

AI-driven markets can operate much faster.

An algorithm may:

change prices thousands of times a day;

process millions of transactions;

personalise offers;

rank competitors dynamically;

coordinate through machine-learning systems;

optimise simultaneously across multiple markets.

Consequently, the evidence required for enforcement may exist in:

source code;

model weights;

APIs;

logs;

training datasets;

prompts;

model outputs;

cloud infrastructure;

automated decision records.

This requires competition authorities to develop institutional capacity beyond conventional legal and economic expertise.

3. Main Elements of Institutional Redesign

Institutional redesign can involve:

AI-specialist investigative units

algorithmic auditing capabilities

real-time market monitoring

data-science departments

technical evidence protocols

AI-assisted merger review

algorithmic cartel detection

regulatory sandboxes

cross-border cooperation

procedural safeguards for AI-generated evidence

The objective should be to strengthen enforcement while preserving due process and judicial review.

4. AI Changes the Nature of Competition Evidence

Traditional evidence may include:

contracts;

emails;

meeting records;

prices;

sales figures.

AI markets generate additional evidence:

model-training records;

system prompts;

algorithmic instructions;

API calls;

automated pricing logs;

recommendation outputs;

model version histories;

data pipelines;

machine-generated communications.

Competition authorities therefore need institutional procedures for collecting and preserving algorithmic evidence.

5. AI-Assisted Cartel Detection

AI can identify unusual patterns in procurement and pricing data.

For example, an enforcement system could detect:

suspiciously similar bids;

repeated price movements;

unusual bid rotation;

identical timing patterns;

coordinated capacity reductions.

The system could then identify cases requiring human investigation.

The important institutional principle is:

AI should generally identify enforcement leads; legal liability should remain subject to evidence and human decision-making.

6. Algorithmic Collusion

AI can complicate the traditional distinction between:

explicit collusion; and

independent parallel conduct.

Two algorithms may independently discover that coordinated pricing maximises profits without human executives explicitly communicating.

This creates difficult questions:

Is there an agreement?

Is there a concerted practice?

Is autonomous adaptation sufficient?

Who is legally responsible?

What evidence establishes intention?

Does deployment of a known coordination-capable algorithm create responsibility?

Institutional redesign must therefore equip authorities to investigate algorithmic causation and human responsibility.

7. Tacit Coordination and AI

Traditional tacit coordination depends upon firms observing each other's conduct.

AI makes this process much faster.

Algorithms can:

observe competitor prices;

predict competitor reactions;

adjust prices;

observe the result;

optimise future decisions.

This can create highly responsive coordination.

Competition authorities therefore need tools capable of distinguishing:

legitimate algorithmic optimisation

from

algorithmically facilitated coordination.

8. AI and Abuse of Dominance

AI can reinforce dominance through several mechanisms.

A dominant platform may control:

data;

computing infrastructure;

cloud services;

foundation models;

application interfaces;

distribution channels;

advertising infrastructure.

It may then use that position to:

preference its own services;

restrict competitors' access;

impose discriminatory conditions;

tie products;

exploit data advantages;

restrict interoperability.

Institutional redesign therefore requires authorities to analyse multi-layer ecosystems, rather than isolated product markets.

9. AI and Merger Control

AI changes merger analysis because a start-up's current revenue may be a poor indicator of its competitive importance.

An AI start-up might possess:

unique datasets;

specialised researchers;

important algorithms;

intellectual property;

strategic partnerships;

an emerging technology.

Its acquisition by a dominant company could eliminate future competition even before the start-up has significant revenues.

Competition authorities therefore increasingly need to assess:

innovation potential + data assets + talent + technology + ecosystem position.

10. Case Law 1 — Microsoft Corp. v Commission

The Microsoft case is highly relevant to AI-driven enforcement because it demonstrates the competition significance of technological ecosystems.

The case concerned Microsoft's dominance in PC operating systems and its refusal to provide interoperability information needed by competing work-group server operating systems.

The European courts upheld important elements of the Commission's reasoning.

Institutional significance

An authority dealing with AI platforms may face similar questions involving:

APIs;

model interfaces;

interoperability protocols;

cloud access;

technical documentation.

Principle

Competition authorities require sufficient technical expertise to determine whether control over an interface can prevent effective competition.

11. Case Law 2 — Google Shopping

The Google Shopping case concerned Google's treatment of comparison-shopping services within its search ecosystem.

The European Commission found that Google had abused its dominant position by favouring its own comparison-shopping service.

Institutional significance

AI-powered search and recommendation systems create similar issues.

An authority may need to determine whether an AI system:

ranks its own services preferentially;

suppresses competitors;

uses data obtained from competitors;

changes ranking rules selectively.

Principle

Effective competition enforcement may require technical examination of ranking and recommendation systems, not merely contractual documents.

12. Case Law 3 — Google Android

The Google Android case involved Google's conduct concerning Android devices, applications, search and ecosystem distribution.

The case illustrates how control over an operating-system ecosystem can be leveraged into adjacent markets.

Institutional significance

AI ecosystems may similarly involve:

foundation model → cloud → app store → assistant → advertising → data.

Competition authorities therefore require institutional structures capable of analysing ecosystem-wide leverage.

Principle

Market power in one technological layer can influence competition in connected markets.

13. Case Law 4 — Qualcomm

The Qualcomm litigation involved competition concerns concerning licensing practices and mobile communications technology.

The case illustrates the importance of understanding:

intellectual property;

technical standards;

licensing structures;

downstream competition.

Institutional significance

AI enforcement may similarly involve specialised technologies and IP rights.

Authorities need experts capable of analysing the interaction between:

technology + licensing + standards + competition.

Principle

Complex technology markets require competition analysis capable of understanding technical and commercial structures simultaneously.

14. Case Law 5 — Intel v Commission

The Intel litigation concerned alleged exclusionary rebates and the assessment of effects in a market involving powerful technology firms.

The Court of Justice emphasised the importance of analysing the actual or potential effects of exclusionary conduct in appropriate circumstances.

Institutional significance

AI-driven markets can involve complex pricing, discounts and contractual incentives.

An authority should therefore have economic and computational tools capable of testing:

foreclosure;

switching costs;

customer segmentation;

rival viability;

pricing effects.

Principle

Institutional capacity must support economically rigorous assessment rather than relying solely on formal contractual classification.

15. Case Law 6 — Bronner v Mediaprint

Bronner concerned access to a newspaper distribution system and the essential-facilities doctrine.

The Court adopted a demanding approach to compulsory access.

Institutional significance

AI infrastructure can create similar questions concerning:

cloud computing;

datasets;

APIs;

computing resources;

foundation-model access.

Authorities must determine whether the resource is genuinely indispensable or merely commercially advantageous.

Principle

AI enforcement institutions need rigorous legal and economic criteria before requiring access to privately controlled technological infrastructure.

16. Case Law 7 — IMS Health

IMS Health concerned refusal to license an intellectual-property-protected data structure.

The Court considered circumstances in which refusal to license could amount to abuse.

Institutional significance

AI systems often depend on:

proprietary datasets;

model architectures;

training resources;

APIs.

Institutional redesign must therefore combine competition-law expertise with intellectual-property and technical expertise.

Principle

AI-related access remedies require careful consideration of both competitive necessity and legitimate IP interests.

17. Case Law 8 — United States v Microsoft

The US Microsoft case involved exclusionary conduct concerning the development and distribution of competing technologies.

The case is relevant to AI because it illustrates how a powerful incumbent can use control over an established technological platform to influence emerging technological competition.

Institutional significance

AI enforcement may require authorities to investigate whether incumbent platforms use:

operating systems;

browsers;

cloud infrastructure;

app distribution;

APIs;

default settings

to disadvantage emerging AI competitors.

Principle

Institutional enforcement must be capable of analysing how technological ecosystems influence future competition.

18. AI-Driven Market Monitoring

Traditional competition authorities often rely on periodic investigations.

AI allows continuous monitoring.

A competition authority could create systems that monitor:

price movements;

procurement bids;

mergers;

market shares;

advertising auctions;

platform rankings;

algorithmic changes.

This could transform enforcement from:

reactive enforcement

into:

continuous competition surveillance.

However, continuous surveillance raises important legal and privacy issues.

19. Algorithmic Auditing Units

Competition authorities could establish dedicated Algorithmic Competition Audit Units.

Their responsibilities might include:

Technical auditing

Examine algorithms and system architecture.

Data analysis

Detect suspicious patterns.

Simulation

Model possible competitive effects.

Reproducibility

Reconstruct algorithmic decisions.

Monitoring

Track changes after remedies.

Expert evidence

Provide technical evidence to investigators and courts.

Such units should operate alongside, rather than replace, traditional lawyers and economists.

20. AI and Merger Simulation

AI can assist merger review through simulations.

For example, authorities could model:

price effects;

innovation incentives;

customer switching;

entry;

product substitution;

network effects.

AI can process large datasets more rapidly than conventional manual analysis.

But model outputs should remain evidence supporting human economic judgment, not automatic merger decisions.

21. AI and Digital Evidence

Institutional redesign should establish clear rules concerning:

authenticity;

chain of custody;

reproducibility;

explainability;

source-code access;

model-version preservation.

Suppose an authority alleges that an algorithm engaged in discriminatory pricing.

The undertaking might argue:

"That behaviour was not programmed."

The authority may then need to determine whether the behaviour arose from:

explicit instructions;

training data;

reinforcement learning;

model optimisation;

external inputs.

This requires specialised technical investigation.

22. Explainability

An AI enforcement tool may identify:

"High probability of coordinated bidding."

But the authority must still be able to explain:

what data were used;

which variables mattered;

how the conclusion was reached;

whether alternative explanations exist.

This is particularly important when AI-generated evidence is used against undertakings.

23. Human Oversight

Institutional redesign should preserve human responsibility.

A useful model is:

AI detection → expert validation → legal investigation → adversarial process → institutional decision → judicial review.

This reduces the danger of:

false positives;

hidden assumptions;

algorithmic bias;

automation bias.

The competition authority remains legally responsible for its decision.

24. AI and Regulatory Sandboxes

Competition authorities could establish controlled environments where innovative businesses can test:

pricing algorithms;

interoperability systems;

AI recommendation systems;

procurement technologies.

A sandbox can allow authorities to understand competitive effects before a technology becomes widespread.

However, sandboxes should not provide an implicit exemption from competition law.

25. Cross-Agency Institutional Cooperation

AI markets often cross traditional regulatory boundaries.

Competition authorities may need cooperation with:

data-protection authorities;

telecommunications regulators;

financial regulators;

consumer-protection authorities;

cybersecurity authorities;

intellectual-property offices.

For example:

AI platform → data → cloud → payments → advertising

may involve several regulatory regimes.

Institutional redesign should therefore provide lawful mechanisms for information sharing and coordinated investigations.

26. International Cooperation

AI markets are inherently global.

An AI company may:

develop models in one country;

train them in another;

use cloud infrastructure elsewhere;

sell services globally.

Competition authorities therefore need:

cross-border evidence mechanisms;

cooperation agreements;

coordinated merger review;

shared technical expertise;

compatible investigative methodologies.

Without cooperation, firms may face fragmented enforcement while global platforms operate across jurisdictions.

27. Institutional Independence

AI enforcement creates a risk of excessive technological dependence.

If a competition authority relies heavily on:

private AI vendors;

cloud providers;

external consultants;

proprietary analytical systems,

the regulator may become dependent upon the very ecosystem it regulates.

Institutional redesign should therefore preserve:

independent technical capacity;

transparent procurement;

auditability;

staff expertise;

control over regulatory data.

28. AI and Procedural Fairness

AI-assisted enforcement must respect procedural rights.

An undertaking should ordinarily be able to understand the substantive case against it.

Relevant safeguards include:

disclosure of material evidence;

opportunity to respond;

independent review;

access to expert examination;

protection of confidential information;

judicial review.

An authority should not rely on an opaque model merely by stating:

"The algorithm identified an infringement."

The AI output is evidence, not a substitute for legal reasoning.

29. False Positives

AI systems may identify conduct as suspicious when it is actually legitimate.

For example, competitors may independently increase prices because:

input costs increased;

demand increased;

supply declined.

A machine-learning model could mistake parallel price movements for coordination.

Therefore:

Statistical correlation ≠ proof of anticompetitive agreement.

Human investigators must establish the legal elements required by the applicable competition law.

30. AI and Competition Intelligence

Competition authorities can use AI to construct market-intelligence systems.

Such systems could map:

ownership relationships;

board connections;

technology dependencies;

supply chains;

patents;

licensing relationships;

venture investments;

API dependencies;

data flows.

This could reveal concentration that conventional market-share statistics overlook.

31. AI and Ecosystem Mapping

Modern AI markets frequently involve multiple interconnected markets.

A competition authority could create an ecosystem dependency map:

Chips
↓
Cloud
↓
Foundation model
↓
API
↓
Applications
↓
Distribution
↓
Consumers

Such mapping could reveal strategic bottlenecks.

The authority could then investigate whether control over one layer allows an undertaking to exclude competitors elsewhere.

32. AI and Institutional Memory

Competition authorities conduct investigations over many years.

AI can help preserve institutional knowledge concerning:

previous decisions;

market definitions;

economic models;

remedies;

compliance commitments;

recurring conduct.

This can improve consistency.

But institutional memory systems must also account for changes in technology and market conditions.

33. AI-Driven Remedies Monitoring

AI can also monitor compliance with remedies.

Suppose a dominant platform is required to provide non-discriminatory API access.

A monitoring system could continuously compare:

access speed;

API functionality;

rejection rates;

fees;

downtime;

treatment of affiliated and unaffiliated companies.

This can make behavioural remedies more measurable.

34. Risks of AI-Driven Enforcement

AI itself creates institutional risks.

Automation bias

Officials may trust algorithmic recommendations too much.

Model bias

Training data may distort results.

Opacity

Authorities may not understand proprietary models.

False positives

Legitimate competitive conduct may be investigated.

False negatives

Sophisticated collusion may remain undetected.

Privacy

Large-scale data analysis may involve sensitive information.

Cybersecurity

Competition authorities become attractive targets for cyberattacks.

35. Institutional Structure for AI Competition Enforcement

A possible institutional architecture could contain five layers.

Layer 1 — Legal Unit

Interprets:

competition statutes;

precedents;

procedural rules.

Layer 2 — Economic Unit

Analyses:

market definition;

market power;

incentives;

effects.

Layer 3 — AI/Technical Unit

Examines:

algorithms;

APIs;

models;

datasets;

system architecture.

Layer 4 — Data Intelligence Unit

Conducts:

network analysis;

anomaly detection;

continuous monitoring.

Layer 5 — Remedies and Compliance Unit

Monitors:

interoperability;

non-discrimination;

data access;

algorithmic commitments.

This institutional model promotes interdisciplinary enforcement.

36. Indian Competition Commission Perspective

For the Competition Commission of India, AI-driven enforcement could potentially complement existing investigative capabilities.

Potential institutional developments include:

specialist digital-markets teams;

algorithmic investigation capabilities;

technical evidence laboratories;

AI-based procurement-cartel detection;

digital merger-screening tools;

ecosystem mapping;

cross-regulator coordination.

However, these capabilities should operate within the statutory framework of the Competition Act, 2002 and applicable procedural safeguards.

37. AI and Competition Advocacy

Competition authorities can also use AI to identify structural problems before they become enforcement cases.

For example, an authority could analyse:

concentration trends;

investment flows;

patent ownership;

start-up acquisitions;

interoperability restrictions;

procurement participation.

The result could be competition advocacy directed at policymakers.

38. Difference Between AI-Assisted and AI-Automated Enforcement

This distinction is fundamental.

AI-assisted enforcement

AI:

searches;

identifies patterns;

predicts;

classifies;

prioritises cases.

Humans make the legal decision.

AI-automated enforcement

An algorithm determines:

which firm is investigated;

whether conduct is unlawful;

what remedy applies.

The second model presents substantially greater concerns regarding accountability and procedural fairness.

For competition enforcement, the AI-assisted model provides a clearer institutional basis because legal responsibility remains with accountable officials.

39. Six Institutional Design Principles

A future AI-capable competition authority should follow at least six principles:

1. Technical competence

Recruit data scientists, AI engineers and algorithmic economists.

2. Legal accountability

AI output should not replace legal reasoning.

3. Explainability

Material AI-assisted findings should be capable of meaningful explanation.

4. Human oversight

Important decisions should remain subject to expert review.

5. Independence

Authorities should not become dependent upon private AI providers.

6. Adaptability

Institutions should be able to update their tools as AI technology changes.

40. Overall Assessment

The central institutional challenge is that AI changes both the object and the method of competition enforcement.

The object changes because competition increasingly occurs through:

algorithms;

platforms;

datasets;

APIs;

cloud infrastructure;

foundation models.

The method changes because enforcement itself can use:

machine learning;

automated data analysis;

network analysis;

algorithmic auditing.

The case law from Microsoft, Google Shopping, Google Android, Qualcomm, Intel, Bronner and IMS Health demonstrates why technical expertise, economic analysis and legal judgment must operate together when competition depends upon technological infrastructure.

Conclusion

Institutional redesign for AI-driven competition enforcement should not mean replacing competition authorities with algorithms. It should mean creating competition institutions capable of understanding and supervising algorithmically organised markets.

The emerging institutional model can be summarised as:

AI detection + economic analysis + technical investigation + legal judgment + procedural safeguards + judicial review.

Such a framework allows competition authorities to detect complex conduct more efficiently while preserving the fundamental principles of competition law: evidence, accountability, transparency, proportionality, due process and independent human decision-making.