European Standard-Setting for Competition Enforcement AI

1. Introduction

European standard-setting for Competition Enforcement AI refers to the development of common technical, procedural, legal, and governance standards for using artificial intelligence in the detection, investigation, assessment, and enforcement of European competition law.

AI can assist competition authorities in:

detecting cartels and bid-rigging;

screening mergers and acquisitions;

identifying exclusionary conduct by dominant firms;

analysing large volumes of documents and communications;

detecting algorithmic price coordination;

defining relevant markets;

assessing effects and efficiencies;

monitoring commitments and remedies;

prioritising investigations; and

continuously monitoring systemic digital platforms.

European standard-setting is particularly important because competition enforcement is increasingly data-driven and algorithm-assisted, while EU competition law remains grounded in legal principles such as legality, proportionality, due process, transparency, judicial review, and protection of confidential information.

The central question is therefore:

How can European competition authorities use AI consistently and effectively without allowing automated systems to replace legally accountable competition-law judgment?

2. Meaning of European Standard-Setting

European standard-setting in this context has several dimensions.

A. Technical standards

Authorities may develop common standards concerning:

data formats;

interoperability;

algorithmic auditing;

model validation;

evidentiary reliability;

explainability;

reproducibility;

cybersecurity;

data provenance; and

audit logs.

B. Enforcement standards

Common approaches may be developed for:

cartel detection;

merger screening;

abuse-of-dominance investigations;

market definition;

economic analysis;

dawn-raid evidence processing;

algorithmic evidence;

digital-platform monitoring.

C. Procedural standards

AI-assisted enforcement must remain compatible with:

rights of defence;

access to the file;

confidentiality;

privilege;

reasoned decisions;

judicial review;

impartiality;

proportionality.

D. Institutional standards

European institutions and national competition authorities need common rules regarding:

human oversight;

responsibility for AI outputs;

procurement of AI systems;

independent validation;

model governance;

documentation;

cross-border cooperation.

3. Legal Foundations

European competition-enforcement AI operates within several overlapping legal frameworks.

3.1 Articles 101 and 102 TFEU

Article 101 prohibits anti-competitive agreements, decisions and concerted practices.

Article 102 prohibits abuse of a dominant position.

AI can assist authorities in determining whether conduct falls within these provisions, but the ultimate legal classification remains an exercise of public enforcement authority.

3.2 Regulation 1/2003

Regulation 1/2003 provides the principal procedural framework for EU competition enforcement.

It gives the European Commission investigative powers involving:

requests for information;

inspections;

interviews;

evidence gathering;

decisions;

fines.

AI can substantially increase the Commission's capacity to process evidence obtained through these powers.

However, AI does not enlarge the underlying statutory powers.

This distinction is crucial:

AI may improve the exercise of an enforcement power, but it cannot independently create an enforcement power.

3.3 European Competition Network

The European Competition Network (ECN) provides a major institutional setting for harmonisation.

If different national authorities use different AI methodologies, the same conduct could potentially be:

classified differently;

prioritised differently;

investigated differently; or

evaluated using inconsistent economic models.

European standards can therefore improve consistency of enforcement across Member States.

4. Why Standardisation Is Necessary

4.1 Preventing algorithmic divergence

Suppose one authority uses an AI system that treats a particular pricing pattern as a cartel indicator while another authority does not.

Without common standards, enforcement could become dependent upon:

which authority's algorithm happens to analyse the conduct.

Standardisation reduces this risk.

4.2 Improving evidentiary reliability

AI-generated findings may contain:

false positives;

false negatives;

biased training data;

correlation errors;

model drift;

incomplete datasets.

Competition authorities therefore require standards for determining when an AI output is sufficiently reliable to support further investigation.

4.3 Protecting procedural fairness

An undertaking must be able to challenge evidence relied upon against it.

If an authority simply says:

"The AI system identified anti-competitive conduct",

that may be inadequate.

The authority should ordinarily be able to explain:

what data were used;

what methodology was employed;

what question the model answered;

the degree of uncertainty;

whether human validation occurred;

how the output influenced the decision.

5. AI Applications in Competition Enforcement

5.1 Cartel detection

AI can identify suspicious patterns involving:

parallel pricing;

bid rotation;

market allocation;

communication patterns;

suspicious tender participation;

unusual bidding sequences.

Machine-learning systems can identify relationships that conventional screening might miss.

But parallel conduct alone does not establish an infringement of Article 101.

5.2 Merger enforcement

AI can help authorities screen transactions according to:

market concentration;

overlapping products;

customer switching patterns;

patent portfolios;

pricing data;

innovation indicators;

vertical relationships.

It can therefore function as an early-warning system.

However, automated merger screening should not automatically determine whether a transaction is lawful.

5.3 Abuse of dominance

AI can assist in detecting:

discriminatory ranking;

self-preferencing;

exclusionary pricing;

tying;

refusal to supply;

interoperability restrictions;

discriminatory access to data.

This is particularly relevant to large digital platforms.

5.4 Market definition

AI can analyse:

consumer searches;

transaction data;

substitution patterns;

geographic purchasing;

product characteristics;

customer behaviour.

However, legal market definition cannot simply be reduced to statistical similarity.

6. European Case Laws Relevant to AI-Based Enforcement

There is not yet a large body of European case law directly concerning AI systems used by competition authorities. Consequently, the appropriate approach is to use established EU competition, digital-market, procedural, and judicial-review cases to establish the legal principles governing future AI enforcement.

Case Law 1: United Brands v Commission

Case: United Brands Company and United Brands Continentaal BV v Commission, Case 27/76.

Principle

The Court of Justice established the importance of identifying:

relevant product markets;

geographic markets; and

substitutability.

Relevance to enforcement AI

AI may assist market-definition analysis by processing enormous quantities of consumer and pricing data.

However, United Brands demonstrates that market definition ultimately involves a legal-economic assessment, not merely computational classification.

Standard-setting implication

European AI systems should therefore distinguish between:

Data inference → economic analysis → legal conclusion.

The AI should not collapse these stages into one automated determination.

7. Case Law 2: Hoffmann-La Roche v Commission

Case: Hoffmann-La Roche & Co. AG v Commission, Case 85/76.

Principle

The Court established the foundational concept of abuse of dominance and particularly addressed loyalty-inducing rebate practices.

AI relevance

AI enforcement systems may identify:

loyalty rebates;

conditional discounts;

customer exclusion;

pricing patterns.

But the system must not assume that a particular pricing pattern automatically constitutes abuse.

The legal context remains essential.

Standard-setting implication

AI enforcement models should incorporate contextual legal assessment rather than pattern recognition alone.

8. Case Law 3: AKZO Chemie v Commission

Case: AKZO Chemie BV v Commission, Case C-62/86.

Principle

The Court developed important principles concerning predatory pricing and the significance of cost benchmarks.

AI relevance

Modern AI systems could analyse millions of:

prices;

costs;

transactions;

customer-specific discounts.

This can significantly improve enforcement screening.

Problem

A model could incorrectly infer predation from low prices without properly understanding:

cost structure;

strategic context;

market conditions;

efficiencies.

Standard-setting implication

AI tools should therefore identify risk indicators, rather than automatically classify conduct as predatory pricing.

9. Case Law 4: Intel v Commission

Case: Intel Corporation Inc. v Commission, Case C-413/14 P.

Principle

The Court emphasised the importance of assessing the circumstances and effects of rebate practices where relevant.

The case is especially important because it demonstrates that competition enforcement cannot always rely on formal categorisation alone.

AI relevance

AI systems could be trained to detect:

conditional rebates;

exclusivity incentives;

customer foreclosure;

pricing patterns.

But the Intel litigation illustrates why enforcement systems must permit effects-based economic analysis.

Standard-setting implication

AI-generated risk scores should not substitute for:

competitive-effects analysis;

evidence assessment;

consideration of efficiencies;

legal reasoning.

10. Case Law 5: Google Shopping

Case: Google and Alphabet v Commission, Case T-612/17.

Principle

The General Court examined Google's conduct concerning comparison-shopping services and the relationship between:

dominance;

search results;

self-preferencing;

competitive foreclosure;

effects on competition.

AI relevance

This is particularly significant for AI enforcement because modern digital markets are themselves increasingly controlled through algorithms.

Authorities may use AI to examine:

ranking systems;

recommendation systems;

search results;

personalised offers;

platform access;

visibility discrimination.

Standard-setting implication

European enforcement AI should be capable of distinguishing:

algorithmic optimisation from algorithmic exclusion.

A ranking difference by itself should not automatically become an infringement finding.

11. Case Law 6: Intel and the Evidentiary Problem of Automated Analysis

The broader Intel litigation is also important for AI because it demonstrates the importance of economic evidence and procedural scrutiny.

An AI model may identify an apparent exclusionary effect, but the authority must still establish the legal elements required by EU competition law.

This creates an important principle:

An AI prediction is evidence for investigation; it is not itself the legal infringement.

12. Case Law 7: Commission v Tetra Laval

Case: Commission v Tetra Laval BV, Case C-12/03 P.

Principle

The Court imposed significant requirements regarding the assessment of complex merger theories.

AI relevance

AI may be particularly attractive in merger enforcement because merger cases involve enormous datasets and complex predictions.

However, Tetra Laval reinforces the importance of rigorous evidence where an authority relies upon predictions about future market behaviour.

Standard-setting implication

AI-assisted merger decisions should require:

validated assumptions;

sensitivity analysis;

alternative scenarios;

uncertainty assessment;

human review.

13. Case Law 8: Commission v Alrosa

Case: Commission v Alrosa Company Ltd, Case C-441/07 P.

Principle

The Court emphasised proportionality in relation to commitments imposed in competition proceedings.

AI relevance

AI may be used to monitor compliance with commitments.

For example, an automated system could continuously assess whether a dominant platform:

provides access;

respects interoperability;

applies non-discriminatory conditions;

complies with behavioural remedies.

Standard-setting implication

AI monitoring must itself comply with proportionality.

An authority should not collect unlimited data merely because automated monitoring makes such collection technically possible.

14. Core European Standards for Competition Enforcement AI

A coherent European framework should contain at least the following standards.

14.1 Human-in-the-loop standard

AI should assist rather than replace legally accountable officials.

A high-risk enforcement decision should involve:

AI detection;

expert verification;

legal assessment;

evidentiary review;

institutional decision.

14.2 Explainability standard

Authorities should maintain sufficient documentation to explain:

input data;

methodology;

relevant variables;

model limitations;

confidence levels;

validation procedures.

The required degree of explanation should increase with the seriousness of the enforcement consequence.

14.3 Auditability standard

Every significant AI-assisted enforcement process should create an audit trail showing:

model version;

date of deployment;

data used;

changes to the model;

human interventions;

output;

subsequent decision.

14.4 Reproducibility standard

Where practicable, another qualified authority or court should be capable of understanding how the result was produced.

This is especially important where AI-generated analysis becomes part of the evidentiary record.

15. Data Governance

Competition-enforcement AI depends heavily upon data.

European standards should therefore address:

Data quality

Data must be:

accurate;

relevant;

sufficiently complete;

appropriately structured.

Data provenance

Authorities should know:

Where did the data come from?

Data integrity

The authority should be able to demonstrate that data were not improperly altered.

Confidentiality

Commercially sensitive information must remain protected.

Personal data

Competition authorities must also account for applicable European data-protection requirements where personal data are processed.

16. AI and Rights of Defence

This is one of the most important areas.

An undertaking facing enforcement action should not be placed in the position of defending itself against an unexplained algorithm.

Procedural safeguards should therefore include:

meaningful disclosure of relevant methodology;

access to evidence;

opportunity to challenge AI-generated conclusions;

disclosure of material limitations;

human reconsideration;

judicial review.

The principle can be expressed as:

No significant competition sanction should depend exclusively upon an opaque automated determination that cannot meaningfully be challenged.

17. AI Bias in Competition Enforcement

AI systems may reproduce biases contained in training data.

For example, a cartel-detection system trained primarily on historical cartel cases may over-identify industries with characteristics similar to previously prosecuted cartels.

This creates a feedback loop:

Historical enforcement data → AI training → repeated targeting → new enforcement data → reinforced model bias.

European standards should therefore require:

bias testing;

independent validation;

periodic recalibration;

false-positive analysis;

monitoring of sectoral disparities.

18. Standardisation Across the ECN

The European Competition Network could promote common methodologies concerning:

A. Cartel screening

Common definitions of suspicious indicators.

B. Algorithmic pricing

Common analytical methodologies for distinguishing:

conscious parallelism;

algorithmic coordination;

unilateral optimisation;

concerted practices.

C. Digital-platform monitoring

Common metrics concerning:

self-preferencing;

interoperability;

access;

ranking;

data discrimination.

D. AI-assisted evidence

Common standards for:

authenticity;

reliability;

metadata;

chain of custody;

model-generated evidence.

19. Relationship with European AI Regulation

Competition-enforcement AI does not exist independently of the broader European AI regulatory environment.

The EU's AI regulatory framework increasingly emphasises:

risk management;

transparency;

human oversight;

technical documentation;

accuracy;

robustness;

cybersecurity;

accountability.

Competition authorities therefore have an opportunity to develop a specialised competition-enforcement layer on top of general AI governance.

The result could be:

General AI standards

↓

European public-sector AI standards

↓

Competition-authority AI standards

↓

ECN interoperability standards

↓

Authority-specific implementation

20. Standardisation and Algorithmic Cartels

One of the most difficult problems is determining whether AI systems themselves facilitate coordination.

Suppose competing firms use pricing algorithms that independently learn to maintain higher prices.

The legal problem becomes:

Is the outcome merely autonomous algorithmic adaptation, or does it constitute coordination attributable to the undertakings?

European standard-setting should require authorities to examine:

human instructions;

algorithm design;

data sharing;

communication between firms;

optimisation objectives;

monitoring arrangements;

predictability of algorithmic responses.

AI should therefore help establish causal and organisational links, rather than merely identifying parallel prices.

21. Standardisation and Merger AI

AI could create a European merger-screening infrastructure.

A system might calculate a risk score based upon:

Market concentration + overlap + entry barriers + innovation effects + data concentration + network effects + vertical integration.

However:

Risk score ≠ presumptive illegality.

The score should trigger human investigation rather than automatically determine the merger outcome.

22. Standardisation and Remedies

AI can also assist in remedy design.

For example, authorities may use simulations to predict:

market-share changes;

switching;

entry;

pricing;

interoperability effects;

innovation outcomes.

But remedy design must remain consistent with proportionality.

The Alrosa principle is particularly important here.

23. Institutional Accountability

European standard-setting should clearly allocate responsibility.

Developer

Responsible for:

technical design;

security;

documentation;

model limitations.

Competition authority

Responsible for:

lawful deployment;

validation;

oversight;

evidence assessment.

Investigating officer

Responsible for:

interpreting outputs;

corroborating evidence;

ensuring procedural fairness.

Decision-maker

Responsible for:

final legal determination.

This prevents the problematic situation where responsibility disappears into the phrase:

"The algorithm said so."

24. Proposed European Governance Model

A useful model would consist of six layers:

Layer 1 — Legal standards

Articles 101 and 102 TFEU, merger law, procedural rules and fundamental rights.

Layer 2 — Evidentiary standards

Reliability, authenticity, provenance and reproducibility.

Layer 3 — Technical standards

Model accuracy, interoperability, cybersecurity and auditability.

Layer 4 — Procedural standards

Human review, disclosure, rights of defence and reasoned decisions.

Layer 5 — Institutional standards

ECN cooperation, cross-border validation and common methodologies.

Layer 6 — Judicial standards

Effective judicial review of AI-assisted enforcement.

25. Advantages

European standard-setting can provide:

Consistency across Member States.

Efficiency in processing enormous datasets.

Early detection of cartels and exclusionary conduct.

Better evidence management.

Improved digital-market monitoring.

Reduced duplication among authorities.

More sophisticated economic analysis.

Greater institutional cooperation.

26. Risks

There are equally important dangers.

Automation bias

Officials may over-trust algorithmic outputs.

False positives

Legitimate competitive conduct may be investigated unnecessarily.

False negatives

Sophisticated anti-competitive conduct may escape detection.

Opacity

Undertakings may be unable to challenge conclusions.

Data bias

Historical enforcement patterns may influence future enforcement.

Institutional dependency

Authorities may become dependent upon private AI vendors.

Security risks

Sensitive investigative datasets could become targets for cyberattacks.

Regulatory fragmentation

Different European authorities may develop incompatible AI methodologies.

27. Recommended European Standard

A future European standard for competition-enforcement AI could be based upon the following principles:

28. Overall Assessment

European standard-setting for competition-enforcement AI should not be understood as standardising the substantive outcome of competition cases.

Rather, it should standardise the quality, reliability and governance of AI-assisted enforcement.

The most appropriate European model is therefore:

Common technical standards + common evidentiary standards + common procedural safeguards + human legal judgment + judicial review.

The case law from United Brands, Hoffmann-La Roche, AKZO, Intel, Google Shopping, Tetra Laval and Alrosa demonstrates a broader principle: competition enforcement involves contextual economic and legal judgment that cannot safely be reduced to automated pattern recognition.

AI can become an extremely powerful investigative and analytical instrument, but European competition law should preserve the distinction between:

AI detection → AI analysis → human verification → legal assessment → accountable enforcement decision.

That distinction is likely to become one of the central safeguards of European digital competition enforcement.