Competition Law and Machine-Generated Exclusionary Conduct

1. Introduction

Machine-generated exclusionary conduct refers to anti-competitive conduct that is designed, implemented, or substantially carried out by algorithms, artificial intelligence (AI), automated pricing systems, recommendation engines, autonomous agents, or other computational systems, and which has the effect of weakening or excluding competitors.

Traditional competition law generally focuses on the conduct of businesses or economic undertakings. The fact that an algorithm makes or executes the decision does not ordinarily remove the conduct from competition-law scrutiny. The central questions remain:

• Who deployed or controlled the system?
• What market power does the undertaking possess?
• What conduct did the machine implement?
• Did the conduct exclude competitors or restrict competition?
• Was there an objective business justification or efficiency?
• What was the actual or likely competitive effect?

There is currently no fully developed separate body of case law specifically called “machine-generated exclusionary conduct.” Therefore, established cases concerning monopolization, abuse of dominance, foreclosure, refusal to deal, tying, predatory conduct, and digital-platform exclusion are particularly relevant by analogy.

2. Meaning of Machine-Generated Exclusionary Conduct

Machine-generated exclusionary conduct occurs where an automated system contributes materially to conduct that makes it more difficult for competitors to enter, survive, expand, or compete in a market.

Simple example

Suppose a dominant online marketplace uses an algorithm that automatically:

1. identifies competing sellers;
2. reduces their visibility;
3. promotes the platform's own products;
4. restricts access to important customer data; and
5. automatically changes search rankings when competitors become successful.

If these actions substantially disadvantage competitors without sufficient legitimate justification, they may raise competition-law concerns.

The important point is that automation does not automatically make conduct unlawful.

The legal analysis focuses on the economic conduct and its competitive effects, rather than simply on whether a human or machine made the decision.

3. Difference Between Ordinary and Machine-Generated Exclusion

Machine systems can therefore make exclusionary strategies faster, more precise, scalable and difficult to detect.

4. Main Forms of Machine-Generated Exclusionary Conduct

A. Algorithmic Self-Preferencing

A dominant platform may design an algorithm that systematically gives preference to its own products or services.

Example

A search or marketplace algorithm may place the platform's own service above competing services even where competing products would otherwise receive higher rankings.

Competition concerns may arise where the platform controls an important distribution channel and uses that control to disadvantage competitors.

B. Algorithmic Foreclosure

An algorithm may automatically restrict competitors' access to:

• customers;
• distribution channels;
• advertising opportunities;
• data;
• APIs;
• technical infrastructure;
• interoperability;
• search visibility.

This may create foreclosure effects.

Foreclosure does not necessarily mean that competitors are completely removed from the market. Even partial restriction can matter if the affected input or distribution channel is sufficiently important.

C. Algorithmic Predatory Pricing

An automated pricing system could repeatedly set prices below an appropriate measure of cost in order to weaken competitors.

The legal question would not simply be:

“Did the algorithm set a low price?”

Instead, authorities would examine matters such as:

• pricing strategy;
• costs;
• duration;
• market power;
• recoupment where relevant;
• competitive effects;
• legitimate promotional explanations.

D. Algorithmic Loyalty Mechanisms

Algorithms can automatically provide:

• preferential discounts;
• rebates;
• loyalty rewards;
• exclusive benefits;
• ranking advantages;

to customers who purchase predominantly or exclusively from the dominant undertaking.

Such systems can potentially make it harder for competitors to obtain customers.

E. Automated Refusal or Restriction of Access

An AI system could automatically determine that a competitor will receive:

• no API access;
• reduced data access;
• restricted interoperability;
• reduced platform functionality;
• limited advertising access.

If the undertaking possesses substantial market power, automated restrictions can raise refusal-to-deal or essential-input concerns depending on the applicable jurisdiction.

F. Algorithmic Tying and Bundling

A machine may automatically condition access to one service on the use of another service.

For example:

Access to a dominant cloud service → automatic requirement to use the provider's payment, identity, storage or analytics service.

This can create exclusionary effects when competitors cannot realistically compete with the tied product.

G. Algorithmic Discrimination Against Competitors

An algorithm can classify competitors according to:

• size;
• pricing;
• market share;
• customer activity;
• switching behaviour;
• strategic importance.

It could then automatically impose different:

• commissions;
• rankings;
• access conditions;
• advertising prices;
• technical limitations.

Such differentiation is not automatically unlawful, but discriminatory treatment can become relevant when used by a dominant undertaking to disadvantage rivals.

5. Why Machine-Generated Exclusion Is Particularly Important

5.1 Speed

Algorithms can make thousands or millions of decisions rapidly.

5.2 Scale

One automated system can affect an entire market simultaneously.

5.3 Continuous Adaptation

AI systems can change their behaviour according to competitor movements.

5.4 Opacity

Complex machine-learning models may make it difficult to determine why a particular competitor was disadvantaged.

5.5 Personalisation

Different competitors or consumers can receive different treatment.

5.6 Data Advantage

Dominant platforms may have enormous datasets unavailable to smaller competitors.

5.7 Network Effects

Automated exclusion may strengthen an existing network effect, making market entry increasingly difficult.

6. Competition-Law Framework

Machine-generated exclusionary conduct can potentially fall under several established competition-law doctrines.

United States

Potential legal frameworks include:

• Sherman Act §2 monopolization and attempted monopolization;
• Sherman Act §1 where coordination with others is involved;
• Clayton Act §3 in appropriate exclusive-dealing circumstances;
• other federal and state competition provisions.

European Union

Potentially relevant provisions include:

• Article 102 TFEU concerning abuse of a dominant position;
• Article 101 TFEU where algorithmic conduct involves coordination;
• the EU Digital Markets Act for designated gatekeepers and specified obligations.

India

Potentially relevant provisions include:

• Competition Act, 2002, Section 4 concerning abuse of dominant position;
• Sections 3 and 5–6 where the facts involve agreements or combinations;
• relevant regulations and digital-market enforcement developments.

The precise legal test depends upon the jurisdiction and the particular conduct.

7. Elements of an Exclusionary Competition Case

A. Relevant Market

Authorities normally identify the relevant:

• product/service market;
• geographic market.

Digital markets can create difficult questions concerning:

• zero-price services;
• multi-sided platforms;
• data;
• interoperability;
• ecosystems.

B. Market Power or Dominance

The existence of an algorithm alone does not establish unlawful exclusion.

Authorities may consider:

• market share;
• barriers to entry;
• network effects;
• switching costs;
• control over data;
• access to infrastructure;
• economies of scale;
• ecosystem advantages.

C. Exclusionary Conduct

The authority must identify the conduct that allegedly harms competition.

Examples include:

• self-preferencing;
• discriminatory access;
• refusal to supply;
• tying;
• predatory pricing;
• exclusive arrangements;
• technical restrictions;
• interoperability restrictions.

D. Competitive Effects

The investigation may consider whether the conduct:

• excludes competitors;
• raises rivals' costs;
• reduces consumer choice;
• prevents entry;
• reduces innovation;
• increases switching costs;
• strengthens dominance.

E. Objective Justification and Efficiencies

The undertaking may argue that the algorithm is justified because of:

• security;
• fraud prevention;
• quality control;
• privacy;
• cybersecurity;
• technical compatibility;
• consumer protection;
• efficiency.

Competition law should therefore distinguish legitimate automated optimization from exclusionary conduct.

8. Important Case Laws

Because the exact concept of machine-generated exclusion is relatively new, the following cases provide the principal legal foundations by analogy.

1. United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001)

Facts

Microsoft was found to have engaged in various exclusionary practices concerning the web-browser market and its relationship with the Windows operating system.

Principle

The case demonstrates that a dominant undertaking can violate competition law by using control over one important platform or product to disadvantage competing technologies.

Relevance to machine-generated exclusion

Modern platforms can use algorithms in ways analogous to Microsoft's use of technical and contractual mechanisms.

For example, an operating-system or platform algorithm could:

• restrict interoperability;
• disadvantage competing applications;
• alter default settings;
• suppress competing services.

The technical nature of the exclusion does not prevent competition-law scrutiny.

Lesson

Technological integration can become an exclusionary competition problem when used by a dominant undertaking to disadvantage competition.

2. United Brands Co. v. Commission, Case 27/76 (1978)

Facts

United Brands was found to have abused its dominant position through various commercial practices involving the banana market.

Principle

The case is a foundational EU authority concerning abuse of dominance and exclusionary conduct.

Relevance

A machine-generated commercial strategy may still constitute conduct attributable to the undertaking operating the system.

An AI system cannot ordinarily be treated as a legal shield separating the undertaking from its market conduct.

Lesson

Dominance and exclusion must be examined through the economic behaviour of the undertaking, regardless of whether modern technology is used to implement that behaviour.

3. Hoffmann-La Roche & Co. AG v. Commission, Case 85/76 (1979)

Facts

The case concerned loyalty-inducing arrangements adopted by a dominant undertaking.

Principle

The Court emphasized that a dominant undertaking has a special responsibility not to allow its conduct to impair genuine and undistorted competition.

Relevance

An AI-driven loyalty system could automatically:

• calculate rebates;
• target customers;
• modify incentives;
• identify customers likely to switch.

If such an automated system produces exclusionary loyalty effects, the underlying conduct can potentially be assessed under Article 102.

Lesson

Automation does not eliminate the special responsibilities associated with dominance.

4. Intel Corp. v. European Commission, Case C-413/14 P (2017)

Facts

Intel's rebate practices were examined under EU competition law.

Principle

The judgment emphasized the importance of examining the circumstances and potential foreclosure effects of rebates rather than relying solely on a formal characterization.

Relevant considerations can include:

• dominant position;
• market coverage;
• duration;
• conditions;
• share of the market affected;
• ability of competitors to compete.

Relevance to machine-generated exclusion

An AI system can automatically calculate individualized rebates or incentives.

This makes Intel particularly relevant to algorithmic commercial systems.

Example

An algorithm could automatically provide stronger discounts to customers who might otherwise switch to competitors.

The system's automated nature would not itself resolve whether the practice is lawful.

Lesson

Automated rebate systems can require effects-based examination where the applicable legal framework calls for it.

5. Google Search (Shopping), Case T-612/17 / Commission Decision

Facts

Google was found by the European Commission to have systematically given prominent placement to its own comparison-shopping service while demoting competing comparison-shopping services.

Principle

The case is highly important for digital-platform exclusion and self-preferencing.

Relevance to machine-generated exclusion

Search rankings are produced substantially through technological systems.

A platform algorithm can determine:

• visibility;
• ranking;
• traffic;
• prominence;
• consumer access.

Therefore, algorithmic ranking can become an important mechanism of exclusion.

Lesson

Where a dominant platform controls a major digital gateway, algorithmic ranking decisions can have substantial competitive significance.

6. Slovak Telekom a.s. and Deutsche Telekom AG, Joined Cases C-165/19 P and C-166/19 P (2021)

Facts

The cases concerned exclusionary conduct involving access to telecommunications infrastructure.

Principle

EU competition law recognizes that exclusionary practices concerning access to infrastructure can raise Article 102 concerns under appropriate circumstances.

Relevance

Modern digital ecosystems increasingly depend upon access to:

• APIs;
• cloud infrastructure;
• operating systems;
• app stores;
• data;
• technical interfaces.

An automated system that restricts such access can therefore create analogous competition concerns.

Lesson

Technical access restrictions can be economically significant even when they are implemented through sophisticated infrastructure rather than conventional contracts.

7. Android, Case T-604/18, Google and Alphabet v. Commission

Facts

The European Commission addressed Google's practices concerning the Android mobile ecosystem, including contractual arrangements associated with search and browser distribution.

Principle

The case illustrates how contractual and technological arrangements within a digital ecosystem can potentially reinforce market power and disadvantage competing services.

Relevance

Machine-generated exclusion can occur at multiple levels of an ecosystem:

Operating system → app distribution → search → advertising → data

An automated system may make the ecosystem even more difficult for rivals to challenge.

Lesson

Competition analysis of digital ecosystems may need to examine interactions between several connected products rather than viewing each automated decision in isolation.

8. Eastman Kodak Co. v. Image Technical Services, Inc., 504 U.S. 451 (1992)

Facts

Kodak's practices concerning replacement parts and servicing of its equipment were challenged as exclusionary.

Principle

The case recognized that market power can potentially exist in an aftermarket even where competition exists in a primary equipment market.

Relevance

Machine ecosystems can create similar aftermarket concerns.

For example:

AI device → proprietary software → proprietary data → proprietary repair/service ecosystem.

An undertaking might use automated restrictions to make customers dependent on its complementary services.

Lesson

Competition analysis may need to examine aftermarkets and ecosystem dependencies, not merely the primary product.

9. International Salt Co. v. United States, 332 U.S. 392 (1947)

Facts

International Salt tied the sale or lease of salt-processing machines to the purchase of salt.

Principle

The case became an important historical authority concerning tying arrangements.

Relevance

Modern AI systems can automatically implement tying arrangements.

For example:

access to AI infrastructure → mandatory use of a particular complementary service.

The machine may automatically enforce the condition.

Lesson

Technological automation can modernize traditional exclusionary practices without changing their fundamental competitive character.

10. Jefferson Parish Hospital District No. 2 v. Hyde, 466 U.S. 2 (1984)

Facts

The case concerned tying and exclusive arrangements involving hospital services.

Principle

The Supreme Court examined issues including separate products, market power and competitive effects.

Relevance

Automated platform ecosystems may similarly bundle separate products or services.

The fact that the bundling decision is generated algorithmically does not eliminate the need to examine:

• separate products;
• market power;
• coercion;
• competitive effects.

9. Attribution of Machine Conduct

One of the most difficult issues is:

Who is responsible when an AI system makes the exclusionary decision?

Possible actors include:

1. Developer

The developer creates the software.

2. Platform operator

The undertaking deploys the system in its commercial operations.

3. Business decision-maker

Management determines the objectives or parameters.

4. Data provider

A third party may supply data used by the system.

5. Autonomous system

The system may adapt its conduct without a specific human instruction for every individual transaction.

Competition law generally focuses on the undertaking and its conduct, rather than allowing the machine itself to become a convenient substitute for legal responsibility.

10. Intent Versus Effect

A major issue is whether the company intended to exclude competitors.

A company might argue:

“The AI independently produced this result.”

That statement does not automatically resolve the competition-law question.

Depending on the applicable legal standard, authorities may examine:

• purpose;
• design;
• foreseeable effects;
• implementation;
• market circumstances;
• actual effects.

A company may therefore face scrutiny where exclusion was not expressly programmed but was reasonably foreseeable from the system's design and deployment.

11. Algorithmic Opacity

Machine learning can make exclusion difficult to investigate.

Traditional software may follow:

IF X → THEN Y.

Machine learning may instead identify patterns through:

• training data;
• optimization functions;
• reinforcement;
• feedback loops;
• continuously updated parameters.

Consequently, investigators may need to examine:

• model objectives;
• training data;
• input variables;
• output patterns;
• logs;
• system changes;
• human instructions;
• performance metrics.

12. Raising Rivals' Costs Through AI

A particularly important theory is raising rivals' costs.

Suppose a dominant platform's algorithm:

1. restricts competitor access to data;
2. increases competitors' advertising costs;
3. reduces their visibility;
4. increases technical integration costs;
5. makes switching more difficult.

The competitor may remain legally present but become substantially less capable of competing.

This can be economically important even without complete exclusion.

13. Machine-Generated Exclusion and Network Effects

Digital markets often have strong network effects.

More users → more data → better algorithms → better service → more users.

A dominant undertaking can therefore potentially create a feedback loop:

Existing dominance
↓
More data
↓
Better machine optimization
↓
Higher user engagement
↓
More data
↓
Stronger market position
↓
Greater ability to disadvantage competitors

Competition authorities may need to examine whether such feedback mechanisms reinforce durable market power.

14. Data as an Exclusionary Instrument

Data can itself become a competitive advantage.

An undertaking may have:

• transaction data;
• consumer behaviour data;
• search data;
• location data;
• advertising data;
• supplier information.

An algorithm can use this information to identify emerging competitors before they become significant.

For example:

Platform observes competitor's rapid growth → algorithm identifies threat → ranking is reduced → competitor loses customers → growth slows.

Such conduct would require careful factual and economic analysis; merely possessing or analyzing competitor data is not automatically unlawful.

15. Legitimate Uses of Algorithms

Not every algorithmic disadvantage is anti-competitive.

Algorithms may legitimately:

• improve cybersecurity;
• detect fraud;
• prevent manipulation;
• remove counterfeit products;
• improve delivery;
• reduce prices;
• personalize services;
• improve product quality;
• protect privacy.

Therefore, competition law must distinguish:

Legitimate optimization

from

Exclusionary optimization.

This distinction is particularly important because a system can produce competitive disadvantages without having been designed primarily to eliminate competition.

16. Efficiency Defences

An undertaking may argue that the machine-generated practice creates legitimate efficiencies.

Possible efficiencies include:

• lower transaction costs;
• improved matching;
• faster delivery;
• reduced fraud;
• improved security;
• lower prices;
• better product quality;
• interoperability improvements;
• innovation.

Authorities may examine whether the claimed efficiency is:

1. genuine;
2. sufficiently connected to the conduct;
3. verifiable;
4. beneficial to consumers;
5. achievable through less restrictive means.

The precise legal test varies by jurisdiction and type of conduct.

17. Competition Concerns in Autonomous AI Agents

Future commercial systems may involve AI agents that can:

• negotiate contracts;
• purchase goods;
• set prices;
• select suppliers;
• allocate advertising;
• change distribution;
• negotiate discounts.

This creates a new problem.

An AI agent may independently decide:

“Competitor X is becoming a threat, therefore reduce its access.”

The legal system will need to determine how responsibility is allocated when no employee personally approves each individual decision.

The likely focus will remain on the undertaking's deployment, objectives, governance, controls and foreseeable operation of the system, rather than treating autonomous software as outside competition law.

18. Remedies

Competition authorities may use several remedies where legally available.

Structural remedies

• divestiture;
• separation of business units;
• reduction of ecosystem conflicts.

Behavioural remedies

• prohibition of discriminatory ranking;
• non-discrimination obligations;
• access requirements;
• interoperability;
• restrictions on self-preferencing.

Technical remedies

• API access;
• data portability;
• interoperability;
• independent auditing;
• algorithmic transparency.

Monitoring

Authorities may require:

• periodic reports;
• algorithm audits;
• compliance officers;
• documentation of model changes;
• testing for discriminatory outcomes.

19. Proposed Compliance Framework for Businesses

Businesses using AI in competitive markets should maintain:

1. Algorithmic competition assessment

Before deploying high-impact systems.

2. Human oversight

Important competitive decisions should have appropriate governance.

3. Audit trails

Maintain records of:

• model versions;
• instructions;
• parameters;
• important decisions.

4. Competitor-impact testing

Evaluate whether the system systematically disadvantages competitors.

5. Non-discrimination controls

Test rankings, prices and access conditions.

6. Periodic review

AI systems can change over time, so one-time compliance testing may be insufficient.

7. Legal accountability

Clearly identify the undertaking responsible for deployment and governance.

20. Key Case-Law Table

21. Major Legal Challenges

A. Proving Causation

It may be difficult to demonstrate that the algorithm caused the competitive harm.

B. Explaining the Model

Complex AI may not provide an easily understandable explanation for each decision.

C. Distinguishing Intentional From Emergent Conduct

The exclusionary effect may arise from machine learning rather than an explicit instruction.

D. Establishing Counterfactuals

Authorities may need to ask:

What would the market have looked like without the algorithmic restriction?

E. Rapid Technological Change

An investigation may take years while the relevant algorithm changes several times.

F. Cross-Border Operation

A single algorithm can affect markets in multiple jurisdictions simultaneously.

22. Future Development of Competition Law

Competition law is likely to increasingly examine the relationship between:

AI + data + algorithms + platforms + infrastructure + network effects.

Future enforcement may focus on:

• autonomous pricing;
• algorithmic self-preferencing;
• AI-enabled exclusion;
• automated interoperability restrictions;
• algorithmic loyalty schemes;
• AI-driven tying;
• automated refusal of access;
• machine-curated marketplaces;
• AI-controlled supply chains;
• autonomous commercial agents.

However, established competition principles remain important. The fact that technology is new does not necessarily require a completely new legal theory.

23. Conclusion

Machine-generated exclusionary conduct represents an important evolution of traditional exclusionary practices. AI and automated systems can make exclusion faster, more targeted, scalable and difficult to detect.

The central legal principle is that the use of a machine does not by itself make conduct lawful or unlawful. Competition analysis should examine:

• market power;
• relevant market;
• nature of the conduct;
• competitive effects;
• foreclosure;
• consumer impact;
• objective justification;
• efficiencies;
• responsibility for deployment and governance.

Cases such as Microsoft, United Brands, Hoffmann-La Roche, Intel, Google Shopping, Slovak Telekom, Google Android, Kodak, International Salt and Jefferson Parish provide important foundations for analysing these emerging forms of conduct.

The major future challenge will be ensuring that autonomous technological decision-making does not become a mechanism through which dominant undertakings can achieve exclusion that would be unlawful if carried out directly by humans.

Quick Revision Points

1. Machine-generated exclusion = automated conduct capable of disadvantaging competitors.
2. AI does not automatically remove conduct from competition law.
3. Market power remains an important consideration.
4. Algorithmic self-preferencing can create foreclosure concerns.
5. AI can facilitate predatory pricing, tying, loyalty schemes and discriminatory access.
6. Data and network effects can amplify exclusion.
7. Algorithmic opacity makes investigation difficult.
8. Responsibility generally remains connected to the undertaking deploying and governing the system.
9. Legitimate efficiency and security justifications must be distinguished from exclusionary strategies.
10. Microsoft, Intel, Google Shopping, Slovak Telekom and Google Android are particularly useful precedents for understanding digital/technology-driven exclusion.