European Standard-Setting for Competition Enforcement AI

1. Introduction

European standard-setting for Competition Enforcement AI refers to the development of legal, technical, procedural, and institutional standards governing the use of artificial intelligence in competition-law enforcement within Europe. It concerns how competition authorities and courts can use AI for:

• cartel detection;
• merger screening;
• market-definition analysis;
• abuse-of-dominance investigations;
• pricing and algorithmic-collusion detection;
• evidence discovery;
• economic modelling;
• prioritisation of enforcement cases;
• monitoring of remedies and commitments; and
• prediction of potentially anticompetitive conduct.

The European approach is not based on a single statute. Instead, it emerges from the interaction of EU competition law, procedural safeguards, the AI Act, GDPR, digital-market regulation, European standardisation, economic methodology, and case-law principles concerning evidence, proportionality, transparency and judicial review.

A central issue is that AI can improve enforcement capacity while simultaneously creating new risks: opaque decision-making, biased prioritisation, false positives, automated over-enforcement, confidentiality breaches, and excessive dependence on algorithmic outputs.

2. Legal and Institutional Framework

European competition enforcement is principally based on:

A. Articles 101 and 102 TFEU

• Article 101 TFEU prohibits anticompetitive agreements, decisions and concerted practices.
• Article 102 TFEU prohibits abuse of a dominant position.

AI can assist authorities in identifying patterns potentially falling within either provision.

B. Regulation 1/2003

Regulation 1/2003 provides the principal procedural framework for European Commission enforcement of Articles 101 and 102 TFEU.

AI may assist with:

• investigation selection;
• document analysis;
• economic screening;
• evidence organisation;
• dawn-raid material analysis;
• cartel detection.

However, AI cannot replace the Commission's statutory decision-making responsibilities.

C. Digital Markets Act

The DMA creates extensive obligations for designated gatekeepers and provides the Commission with substantial monitoring powers.

AI may be relevant to:

• algorithmic ranking;
• self-preferencing;
• interoperability;
• data access;
• personalised pricing;
• recommender systems;
• platform neutrality.

D. EU AI Act

The AI Act is particularly important because it establishes a horizontal European framework for AI governance.

For competition enforcement, its significance includes:

• risk management;
• transparency;
• human oversight;
• accuracy;
• robustness;
• cybersecurity;
• accountability.

The key principle is that AI used by public authorities should not become a substitute for legally accountable institutional judgment.

E. GDPR

Where competition authorities process personal data, GDPR principles can become relevant, particularly:

• purpose limitation;
• data minimisation;
• accuracy;
• transparency;
• security;
• automated decision-making safeguards.

3. What Does "European Standard-Setting" Mean?

European standard-setting can be divided into five layers.

1. Legal standards

These determine what conduct constitutes an infringement.

2. Evidentiary standards

These determine whether AI-generated evidence is sufficiently reliable to support enforcement.

3. Technical standards

These concern:

• explainability;
• auditability;
• model validation;
• data quality;
• cybersecurity;
• reproducibility.

4. Institutional standards

These determine who is responsible for AI-assisted enforcement.

5. Judicial standards

Courts must be capable of reviewing AI-assisted administrative decisions effectively.

Thus, European standard-setting is ultimately about ensuring that technological efficiency remains subordinate to legality, evidence and judicial accountability.

4. AI in Cartel Detection

One of the most important applications is automated cartel detection.

Competition authorities can analyse:

• prices;
• bidding patterns;
• market shares;
• capacity;
• transaction histories;
• communications;
• procurement data;
• geographic patterns.

Machine-learning models can identify unusual correlations.

For example:

If several competitors repeatedly submit almost identical bids, change prices simultaneously, or systematically rotate winning positions, an AI system may identify the pattern for further investigation.

However, correlation does not establish a cartel.

The European legal standard therefore requires a distinction between:

AI-generated suspicion → investigation → evidence → legal finding.

AI should normally operate as a screening instrument rather than an autonomous infringement adjudicator.

5. AI and Market Definition

AI can also assist in defining relevant markets.

Traditional market-definition analysis may involve:

• demand substitution;
• supply substitution;
• geographic boundaries;
• SSNIP analysis;
• diversion ratios;
• customer surveys;
• econometric evidence.

AI can process enormous datasets and identify substitution patterns that traditional methods may miss.

But automated market definition presents a danger.

A model could define markets according to historical purchasing behaviour rather than economically relevant substitution.

For example:

A consumer may historically purchase from one platform because of a contractual lock-in, even though another product would be a genuine substitute in a competitive market.

Therefore, AI-generated market boundaries must remain subject to economic and legal validation.

6. AI and Abuse of Dominance

AI can help identify potentially abusive behaviour under Article 102 TFEU.

Potential applications include detecting:

• discriminatory access;
• exclusionary pricing;
• self-preferencing;
• tying;
• refusal to supply;
• predatory pricing;
• loyalty-inducing mechanisms;
• discriminatory algorithms.

AI is particularly useful where dominant platforms process millions of transactions.

However, the algorithm must not simply equate:

large market share + unusual behaviour = abuse.

Dominance and abuse are separate legal questions.

7. Algorithmic Collusion

European enforcement faces an emerging problem where pricing algorithms may facilitate coordination without conventional human communication.

For example:

1. Competitor A adopts an AI pricing system.
2. Competitor B adopts another AI pricing system.
3. Both algorithms observe market prices.
4. Each algorithm adjusts prices.
5. Prices converge at supracompetitive levels.

The difficult question is whether this constitutes:

• an agreement;
• concerted practice;
• conscious parallelism;
• unilateral conduct; or
• lawful algorithmic adaptation.

European competition law traditionally requires a legal theory connecting the observed conduct to Article 101 or Article 102.

AI therefore creates pressure for new evidentiary standards without abandoning the existing legal elements of infringement.

8. Six Important Case Laws

1. Commission v Tetra Laval

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

The Court of Justice emphasised that complex economic assessments must rest upon sufficiently convincing evidence and that the Commission's reasoning must support its conclusions.

Relevance to Competition Enforcement AI

This principle is extremely important for AI-assisted enforcement.

An algorithm may produce a prediction that a merger will create anticompetitive effects. But the Commission cannot simply say:

"The model predicts harm."

The authority must demonstrate:

• what data were used;
• why the model is reliable;
• how the prediction relates to the legal theory of harm;
• whether alternative explanations were considered; and
• why the evidence satisfies the applicable legal standard.

Principle: AI prediction cannot replace reasoned economic proof.

2. Airtours v Commission

Case T-342/99, Airtours plc v Commission

The General Court famously required the Commission to establish the necessary conditions for collective dominance through sufficiently persuasive evidence.

The case is important because it demonstrates that complex economic conclusions require a coherent chain of evidence.

AI relevance

AI could identify:

• parallel pricing;
• market concentration;
• transparency;
• repeated interaction;
• capacity patterns.

But those indicators cannot automatically establish coordinated effects.

An AI model can identify candidate evidence, while the authority must establish the legal/economic theory.

3. Intel v Commission

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

The Court of Justice emphasised the importance of analysing the actual or potential exclusionary effects of rebates where appropriate.

AI relevance

AI enforcement systems could incorrectly classify pricing arrangements as abusive simply because they resemble historical patterns associated with exclusion.

The Intel framework illustrates the danger of classification without effects analysis.

AI should therefore assist authorities in analysing:

• pricing effects;
• foreclosure;
• coverage;
• duration;
• competitor access;
• economic incentives.

Principle: automated classification must not eliminate the substantive economic analysis required by competition law.

4. Google Shopping

Case C-48/22 P, Google and Alphabet v Commission

The Court of Justice addressed the Commission's findings concerning Google's treatment of comparison-shopping services.

AI relevance

The case is particularly relevant to AI-based platform enforcement because digital markets increasingly depend upon:

• ranking algorithms;
• search algorithms;
• recommendation systems;
• visibility mechanisms.

An enforcement AI may detect systematic ranking disparities.

But determining whether those disparities constitute unlawful self-preferencing or another form of abuse requires:

• market analysis;
• competitive-effects assessment;
• causal reasoning;
• consideration of legitimate explanations.

Principle: algorithmic evidence must be connected to a legally recognised theory of harm.

5. Estonian Competition Authority / Baltic-style Algorithmic Enforcement Principles

European competition authorities increasingly use computational methods for cartel screening and digital-market monitoring.

The emerging European approach can be understood through the broader case-law principle that competition authorities must establish infringements through evidence capable of judicial scrutiny.

This is especially important in algorithmic enforcement because statistical anomalies may have innocent explanations.

Examples include:

• common cost shocks;
• supply-chain disruption;
• identical public information;
• common input-price movements;
• regulatory changes.

Thus, AI-generated alerts should trigger investigation rather than automatically establish liability.

6. Schenker / Competition Evidence Principles

European competition jurisprudence concerning evidence and Article 101 demonstrates that competition authorities may rely on different categories of evidence, provided the overall evidentiary picture supports the infringement.

This principle has major implications for AI.

AI may combine:

• emails;
• metadata;
• transaction records;
• price histories;
• communications;
• bidding data;
• algorithmic outputs.

The danger is that an AI system could create a self-reinforcing evidentiary loop:

suspicious algorithm → suspicious data selection → model confirms suspicion → model output treated as evidence.

European enforcement standards should therefore require independent validation.

9. Evidentiary Reliability

A European standard for AI competition enforcement should require authorities to assess:

Data quality

Was the data:

• complete?
• accurate?
• representative?
• lawfully obtained?

Model reliability

Was the model:

• independently tested?
• validated?
• stress-tested?
• calibrated?

Error rates

Authorities should understand:

• false positives;
• false negatives;
• confidence intervals;
• model sensitivity.

Explainability

The authority should be able to explain why the AI produced a particular result.

Reproducibility

Where feasible, another qualified analyst should be able to reproduce the result.

10. Human Oversight

The most important European standard should be:

AI may assist competition authorities, but legally accountable officials must retain decision-making responsibility.

Human oversight should exist at several stages:

Data collection → AI screening → human verification → economic analysis → legal assessment → decision → judicial review

This prevents the creation of an:

AI → enforcement decision

pipeline.

Instead, Europe should maintain:

AI → evidence → human assessment → legally reasoned decision.

11. Procedural Fairness

AI-assisted enforcement creates significant procedural concerns.

An investigated company should potentially be able to challenge:

• data reliability;
• methodology;
• model assumptions;
• statistical methodology;
• causal inference;
• classification errors.

This does not necessarily mean that authorities must disclose proprietary source code in every case.

There must instead be a balance between:

investigative confidentiality + defence rights + effective judicial review.

12. Confidentiality and Trade Secrets

Competition investigations frequently involve confidential information.

AI systems may process:

• trade secrets;
• pricing information;
• customer databases;
• strategic plans;
• source code;
• personal data.

A European standard-setting framework should therefore include:

• access controls;
• encryption;
• audit logs;
• data minimisation;
• purpose limitation;
• segregation of sensitive datasets;
• controlled model training.

The risk is particularly high when public authorities use external AI providers.

13. Bias and Discriminatory Enforcement

AI can unintentionally reproduce historical enforcement biases.

For example, if an authority's historical investigations disproportionately focused on certain sectors, a machine-learning system trained on previous enforcement cases could learn that those sectors are more likely to violate competition law.

This creates a feedback loop:

historical enforcement → training data → algorithmic prioritisation → future enforcement → new training data.

European standards should therefore require periodic bias and sector-neutrality audits.

14. AI and Merger Control

AI can transform merger screening.

Authorities could use AI to identify:

• horizontal overlaps;
• vertical relationships;
• conglomerate effects;
• nascent competitors;
• data concentration;
• technology overlaps;
• potential killer acquisitions.

AI can also identify acquisitions that may not appear problematic through traditional market-share analysis.

This is particularly relevant to digital markets where:

today's small firm may become tomorrow's significant competitor.

Nevertheless, AI should not become a substitute for the Commission's substantive merger assessment.

15. AI and European Standardisation Bodies

Technical standardisation may increasingly involve organisations such as:

• CEN;
• CENELEC;
• ETSI.

Their standards can contribute to:

• trustworthy AI;
• cybersecurity;
• data governance;
• interoperability;
• algorithmic documentation;
• risk management.

Competition authorities can benefit from common technical standards because different national authorities may otherwise develop incompatible AI enforcement systems.

European standardisation therefore supports cross-border consistency.

16. European Competition Network

The European Competition Network (ECN) is particularly important.

National competition authorities and the European Commission increasingly confront the same digital platforms and algorithmic systems.

Common AI standards could facilitate:

• shared methodologies;
• common data formats;
• interoperable analytical tools;
• coordinated cartel screening;
• consistent risk assessment;
• cross-border investigations.

This reduces the possibility that:

Germany, France, Italy, Spain and the Commission apply fundamentally different algorithmic enforcement standards to the same platform.

17. Risks of Automated Competition Enforcement

A. False positives

An algorithm may identify lawful conduct as suspicious.

B. False negatives

Sophisticated cartels may evade detection.

C. Automation bias

Officials may trust algorithmic recommendations excessively.

D. Opacity

Black-box systems may make decisions difficult to explain.

E. Data bias

Poor historical data can produce systematically distorted results.

F. Strategic manipulation

Companies may deliberately modify data patterns to evade AI detection.

G. Adversarial behaviour

Market participants may learn how enforcement algorithms work and adjust their conduct accordingly.

18. Proposed European Standard

A robust European framework could adopt a seven-stage AI Competition Enforcement Standard:

Stage 1 — Lawful data acquisition

Verify that data are legally obtained.

Stage 2 — Data governance

Ensure accuracy, security and appropriate use.

Stage 3 — Algorithmic screening

Use AI to identify anomalies and patterns.

Stage 4 — Independent validation

Human economists and investigators verify the results.

Stage 5 — Legal qualification

Determine whether the evidence satisfies Articles 101 or 102 TFEU or another applicable rule.

Stage 6 — Procedural safeguards

Give investigated parties appropriate opportunities to contest the evidence.

Stage 7 — Judicial review

Ensure courts can meaningfully scrutinise the reasoning.

19. AI Enforcement and the Principle of Proportionality

European public law requires proportionality.

Therefore, authorities should ask:

1. Is AI use appropriate?
2. Is it necessary?
3. Is the degree of automated surveillance proportionate?
4. Is the data collection excessive?
5. Are less intrusive methods available?

A system that continuously monitors every transaction of every company could theoretically improve detection but may be disproportionate.

Thus:

enforcement efficiency ≠ unlimited algorithmic surveillance.

20. Relationship Between AI Act and Competition Law

The AI Act and competition law pursue different objectives.

They nevertheless intersect.

For example, a competition authority using a high-risk AI system may have to consider AI-governance obligations while simultaneously applying Articles 101 or 102 TFEU.

21. Key Legal Principle

The most important principle for European competition-enforcement AI is:

AI may improve the discovery and analysis of evidence, but it cannot independently create the legal conclusion that an undertaking infringed competition law.

The final determination must remain:

• legally reasoned;
• economically substantiated;
• procedurally fair;
• explainable;
• reviewable by courts.

22. Conclusion

European standard-setting for Competition Enforcement AI is developing around a human-accountable, evidence-based and technologically neutral model.

The central challenge is to obtain the benefits of AI—speed, pattern recognition, large-scale data processing and predictive analysis—without allowing algorithmic systems to become unreviewable enforcement authorities.

The emerging European model can therefore be expressed as:

AI detection → statistical/economic analysis → human verification → legal qualification → procedural safeguards → reasoned decision → judicial review.

The most important case-law lessons from Tetra Laval, Airtours, Intel and Google Shopping, together with broader European evidentiary principles, are that complex economic conclusions must be supported by convincing evidence and legally adequate reasoning.

Accordingly,