Competition Law and Machine-Owned Enterprises and Antitrust

1. Introduction

Machine-owned enterprises may be understood as enterprises in which artificial intelligence, autonomous software agents, smart contracts, robotics, or machine-controlled systems perform functions traditionally carried out by human managers—such as pricing, purchasing, contracting, allocation of resources, investment, market entry, negotiation, and strategic decision-making.

At present, competition law generally does not recognise an AI system or autonomous machine as an independent legal person equivalent to a corporation. The legal responsibility ordinarily remains with the human, company, platform, owner, developer, operator, or beneficiary behind the system.

The competition-law problem becomes more difficult when a machine can independently make commercially significant decisions. For example:

Company A and Company B each deploy autonomous pricing agents. Neither company instructs its agent to collude. The two agents independently learn that maintaining a high price maximises long-term profits and repeatedly maintain supra-competitive prices.

Traditional antitrust law asks whether there was an agreement, concerted practice, abuse of dominance, or other legally attributable conduct. The machine-autonomy scenario creates a difficult question: who is legally responsible when the economically harmful decision was generated by a machine rather than expressly instructed by a human?

Recent scholarship describes this emerging phenomenon as the possible transition from conventional monopoly structures toward “machine-opoly” or autonomous-agent market power.

2. Meaning of a Machine-Owned Enterprise

The expression can cover several different models.

A. AI-managed enterprise

A conventional company remains legally incorporated, but AI systems make most commercial decisions.

Example:

• AI determines prices;
• AI selects suppliers;
• AI allocates inventory;
• AI negotiates contracts;
• AI determines advertising expenditure.

The company remains the legal undertaking.

B. Autonomous-agent enterprise

An autonomous AI agent can:

1. receive capital or digital assets;
2. identify commercial opportunities;
3. enter transactions;
4. negotiate prices;
5. interact with other agents;
6. reinvest revenues; and
7. modify its strategies without direct human instructions.

C. DAO or smart-contract enterprise

A decentralised autonomous organisation can operate through:

• blockchain protocols;
• smart contracts;
• token-based governance;
• automated treasury management; and
• algorithmic decision-making.

The competition question is whether the apparent decentralisation actually conceals identifiable persons or businesses exercising control.

D. Machine-to-machine markets

The most radical model is a market in which AI agents are the principal decision-makers on both sides.

For example:

AI Agent A purchases electricity from AI Agent B, while both independently negotiate prices and quantities.

The machines may transact without a human intervening in individual transactions.

3. Why Machine-Owned Enterprises Create Antitrust Problems

Traditional competition law assumes that enterprises are operated by humans who can:

• communicate;
• form agreements;
• make decisions;
• understand market consequences; and
• be held responsible for unlawful conduct.

Autonomous systems complicate each assumption.

Principal problems

4. Applicable Competition-Law Framework

A. Agreement and concerted practice

The first issue is whether autonomous conduct can constitute an agreement or concerted practice.

In India, Section 3 of the Competition Act, 2002 is particularly relevant because "agreement" is defined broadly enough to encompass an arrangement, understanding or action in concert.

The central difficulty is:

Can an algorithm's independent optimisation process satisfy the legal requirement of concerted conduct?

The answer will depend upon the factual architecture of the system.

If competitors deliberately configure their machines to coordinate, the traditional agreement doctrine can apply relatively comfortably.

The more difficult case is genuinely autonomous machine learning without human communication.

5. Algorithmic Collusion

Algorithmic collusion can broadly be divided into four models.

1. Messenger model

Humans agree to fix prices and machines merely implement the agreement.

This is the easiest case legally.

2. Hub-and-spoke model

Competitors use a common platform or algorithm as an intermediary.

The platform becomes the "hub", while competitors are the "spokes".

3. Predictable-agent model

A company knows that its algorithm will react predictably to competitors' algorithms.

The algorithm becomes a mechanism for implementing coordination.

4. Autonomous-learning model

Independent AI systems learn from market interactions and converge toward coordinated outcomes without an express human agreement.

This is the most difficult category.

Academic research has specifically examined the possibility that reinforcement-learning agents could independently converge on supra-competitive pricing and develop punishment/reward strategies resembling cartel behaviour.

6. Case Laws

1. United States v. Topkins

United States v. Topkins, No. CR 15-00201 (N.D. Cal. 2015) is one of the foundational algorithmic-pricing cases.

Several online sellers allegedly agreed to fix prices for products sold through Amazon Marketplace and used automated pricing software to implement the arrangement.

Competition principle

The importance of Topkins is that use of software does not transform conventional price fixing into lawful conduct.

The critical legal conduct was the human agreement. The algorithm was the mechanism through which the agreement was implemented.

Relevance to machine-owned enterprises

If a machine-owned enterprise is deliberately programmed to implement an unlawful agreement, the fact that the actual price-setting is performed automatically should not immunise the underlying conduct.

2. Eturas UAB v. Lietuvos Respublikos konkurencijos taryba

Case C-74/14, Eturas

The CJEU considered a platform through which travel agencies operated and where the platform introduced a restriction affecting the maximum discount that could be offered.

Principle

The case is important for determining when participation in a technological system can contribute to a finding of concerted practice.

A platform message or technological mechanism by itself does not automatically establish liability; the surrounding circumstances and knowledge of participating undertakings matter.

Importance for autonomous AI

Suppose several businesses use the same AI platform and the platform automatically changes their pricing rules.

The legal question becomes:

Did the businesses knowingly participate in or accept a coordinated restriction?

Eturas therefore provides an important conceptual bridge between conventional cartel law and machine-mediated coordination. The CJEU's framework focuses on concertation, subsequent market conduct and causation.

3. Trod Ltd v Competition and Markets Authority

Trod Ltd and GB Eye Ltd v CMA, UK Competition Appeal Tribunal (2016)

The case concerned online sellers and price coordination.

Principle

Online marketplaces do not create a competition-law safe harbour.

An agreement between competitors can remain unlawful even where technology is used to monitor and enforce prices.

Relevance

Machine-owned enterprises may use:

• automated monitoring;
• automated repricing;
• AI-based competitor tracking; and
• automatic enforcement.

Trod demonstrates that technological implementation does not eliminate the underlying competition-law analysis.

4. Samir Agrawal v Competition Commission of India

Samir Agrawal v Competition Commission of India, (2021) 3 SCC 136

This is particularly important for Indian competition law.

The case concerned allegations relating to algorithmic pricing by Ola and Uber and whether the platforms facilitated a hub-and-spoke arrangement among drivers.

The CCI and subsequent proceedings considered whether algorithmic pricing amounted to price fixing or a hub-and-spoke cartel. The Supreme Court ultimately dealt with the matter principally on the question of locus/standing, while the underlying algorithmic-pricing controversy remains highly significant for competition-law analysis.

Principle

The case illustrates that:

Algorithmic price uniformity is not automatically equivalent to unlawful collusion.

There must be evidence capable of satisfying the statutory requirements for concerted conduct.

Importance for machine enterprises

An autonomous pricing machine producing identical or similar prices does not necessarily establish an antitrust violation.

The authority would need to investigate:

• who designed the algorithm;
• what data it receives;
• whether competitors share data;
• whether a common pricing mechanism exists;
• whether coordination was intended;
• whether competitors knew about the mechanism; and
• whether the system facilitates anti-competitive coordination.

5. Meru Travel Solutions Pvt Ltd v Uber India Systems Pvt Ltd

Meru Travel Solutions Pvt Ltd v Uber India Systems Pvt Ltd, CCI, 14 July 2021.

The case concerned allegations of predatory pricing and dominance involving Uber in the Delhi-NCR radio-taxi market.

Relevance

Although not a pure "machine-owned enterprise" case, it demonstrates an important distinction:

Algorithmic conduct can generate competition concerns under both Sections 3 and 4 of the Competition Act.

An AI-controlled enterprise could therefore face two different kinds of scrutiny:

• coordination with competitors under Section 3; and
• unilateral exclusionary conduct under Section 4.

6. United States v RealPage

The RealPage litigation is particularly relevant to modern algorithmic pricing.

The U.S. Department of Justice alleged that RealPage's rental-pricing software used competitively sensitive information and facilitated coordination among landlords. The matter was subsequently resolved through a settlement restricting aspects of RealPage's use of competitors' information in pricing recommendations.

Principle

The important issue is not simply that software recommends prices.

The competition concern arises where software architecture enables competitors to:

• share sensitive information;
• reduce independent pricing decisions;
• coordinate through a common system; or
• outsource strategic pricing to a common algorithm.

Machine-enterprise significance

If several autonomous enterprises delegate pricing to the same external AI, that AI could potentially become an antitrust "hub".

7. Uber Technologies / Spencer Meyer litigation

The U.S. litigation involving Uber's algorithmic pricing raised the theory that an intermediary platform could facilitate price coordination among independent drivers.

Although the litigation did not establish a general rule that algorithmic pricing itself constitutes a cartel, the theory became important in subsequent scholarship concerning hub-and-spoke algorithmic coordination.

It illustrates a central distinction:

A platform imposing its own pricing model is not necessarily equivalent to competitors agreeing among themselves to fix prices.

That distinction is particularly important when assessing autonomous enterprises.

8. Apple Inc. v Pepper

Apple Inc. v Pepper, 587 U.S. 273 (2019)

This U.S. Supreme Court case concerned Apple's App Store and antitrust standing.

Although it was not an AI case, it is relevant to machine-controlled enterprises because it illustrates how competition law can analyse a powerful digital intermediary whose technological architecture controls access between different sides of a market.

Relevance

An AI-operated enterprise controlling:

• access;
• payments;
• rankings;
• interoperability;
• data;
• distribution; and
• transactions

could similarly become an important intermediary whose conduct affects downstream competition.

9. Ohio v American Express

Ohio v American Express Co., 585 U.S. 529 (2018)

The U.S. Supreme Court examined competition in a two-sided transaction platform.

Relevance to autonomous enterprises

Many machine-operated businesses will be two-sided or multi-sided:

AI platform → sellers → consumers

or

AI procurement platform → suppliers → industrial customers.

Competition authorities may therefore have to analyse effects across both sides of the platform rather than looking at one market in isolation.

10. United States v Google

The Google antitrust litigation provides an important modern example of competition concerns involving digital infrastructure, distribution and default arrangements.

The significance for machine-owned enterprises lies in the possibility that an autonomous enterprise could use control over a critical technological ecosystem to reinforce its position through:

• defaults;
• distribution;
• interoperability restrictions;
• data advantages;
• exclusive arrangements; and
• ecosystem effects.

The fact that the competitive mechanism is technologically automated would not prevent traditional dominance principles from applying.

7. Machine Ownership Does Not Necessarily Create Separate Legal Personality

This is one of the most important propositions.

Suppose an AI system:

owns cryptocurrency → enters contracts → buys computing capacity → hires other AI agents → earns revenue → reinvests profits.

That does not automatically mean that the AI itself is the legal undertaking.

Competition authorities would likely examine the legal and economic structure behind it.

Possible responsible entities include:

1. AI developer;
2. AI owner;
3. operator;
4. platform;
5. controlling shareholder;
6. DAO participants;
7. smart-contract developers;
8. beneficiaries; or
9. incorporated entity through which the machine operates.

Thus, the first question is not:

"Did the machine violate antitrust law?"

It is:

"Which legally responsible undertaking's conduct is represented by the machine's behaviour?"

8. Autonomous Pricing Without Human Collusion

This is the hardest theoretical problem.

Assume:

• Firm A uses AI-1.
• Firm B uses AI-2.
• Neither communicates with the other.
• Both algorithms observe market prices.
• Both learn that lowering prices causes retaliation.
• Both eventually maintain high prices.

There may be:

No email + No meeting + No contract + No direct communication.

Yet consumers may experience higher prices.

Traditional cartel law can struggle because many legal systems require some form of agreement or concerted conduct.

The problem has been described as the distinction between explicit algorithmic collusion and autonomous/tacit machine coordination.

9. Is Autonomous Machine Collusion Automatically Illegal?

Not necessarily.

This distinction is essential.

Scenario A — Human agreement

"We agree to maintain a price of ₹1,000 and program our AI accordingly."

This presents a conventional cartel problem.

Scenario B — Shared algorithm

Competitors give the same AI provider access to competitively sensitive data and permit the system to determine prices for them.

This creates substantial hub-and-spoke concerns.

Scenario C — Independent learning

Two independent AI systems independently learn that price reductions are unprofitable.

This is legally more difficult.

Scenario D — Deliberately engineered autonomous coordination

Companies know their AI systems will coordinate and deliberately deploy them to produce that result.

The evidence of human involvement may substantially change the legal analysis.

10. Abuse of Dominance by Machine-Owned Enterprises

Autonomous enterprises may also create Section 4 / Article 102-type problems.

A machine-controlled dominant undertaking could potentially engage in:

A. Algorithmic exclusion

The AI automatically lowers prices whenever a new competitor enters the market.

B. Self-preferencing

The AI ranking system systematically promotes products belonging to the enterprise.

C. Discriminatory access

The AI provides different API access conditions to competing businesses.

D. Automated tying

The AI makes access to one service conditional upon purchasing another.

E. Predatory pricing

An AI dynamically prices below an appropriate cost benchmark to eliminate competitors.

F. Refusal to deal

An autonomous system automatically refuses access to essential data, infrastructure or interoperability.

11. Machine-Owned Enterprises and Market Definition

Traditional market definition may become difficult.

An AI company may simultaneously compete in:

• software;
• cloud computing;
• advertising;
• data;
• financial services;
• logistics;
• robotics;
• professional services.

The relevant market could therefore be:

Product market

• AI agents;
• AI infrastructure;
• autonomous decision-making services;
• AI-powered procurement;
• AI-powered financial services.

Input markets

• computing power;
• training data;
• foundation models;
• semiconductor capacity;
• cloud infrastructure.

Downstream markets

• transportation;
• retail;
• financial services;
• healthcare;
• logistics.

A machine enterprise controlling a critical upstream input may acquire substantial downstream leverage.

12. Data as a Source of Machine Market Power

Autonomous enterprises may have an important competitive advantage through continuous data acquisition.

An AI system can collect:

• prices;
• consumer preferences;
• supplier behaviour;
• competitor movements;
• transaction history;
• inventory;
• geographic information;
• demand patterns.

The more transactions the system performs, the more data it may obtain.

This creates a potential feedback loop:

More transactions → more data → better AI → better predictions → more transactions → greater market power.

This is sometimes described as a data/network-effect feedback mechanism.

13. Network Effects and Machine Economies

Machine-owned enterprises can potentially exhibit unusually strong network effects.

For example:

More autonomous agents use Platform X → Platform X obtains more transaction data → its AI becomes more effective → other agents prefer Platform X → competing platforms receive less data → entry becomes more difficult.

Competition authorities may therefore examine:

• economies of scale;
• economies of scope;
• data advantages;
• switching costs;
• interoperability;
• multi-homing;
• access to computing resources;
• network effects.

14. Merger Control

Machine-owned enterprises create novel merger-control questions.

A conventional merger may involve:

Company A buys Company B.

But AI markets may involve:

Company A acquires a small AI developer with a highly valuable model, dataset, autonomous agent or engineering team.

Traditional turnover thresholds may underestimate the competitive importance of the target.

Authorities may therefore examine:

• future competitive significance;
• innovation pipelines;
• data assets;
• compute access;
• intellectual property;
• AI models;
• user networks;
• potential competition.

This is particularly important where an incumbent acquires an emerging AI technology before it becomes a conventional revenue-generating competitor.

15. Killer Acquisitions and Machine Enterprises

An established machine-operated platform might acquire several small autonomous-agent companies.

Even where each target has:

• low revenue;
• few employees;
• limited present market share,

the acquisitions could potentially eliminate future sources of competition.

The relevant question becomes:

What competitive constraint would the target have imposed in the future?

This makes innovation competition especially important.

16. Essential Facilities and AI Infrastructure

Machine-owned enterprises may become dependent upon infrastructure controlled by another AI enterprise.

Potential bottlenecks include:

• GPUs;
• cloud computing;
• foundation models;
• AI APIs;
• specialised datasets;
• model evaluation infrastructure;
• robotic operating systems;
• autonomous-agent marketplaces.

A dominant infrastructure provider could potentially use control of such inputs to disadvantage downstream competitors.

The competition analysis would depend upon the applicable jurisdiction's rules concerning:

• refusal to deal;
• essential facilities;
• discriminatory access;
• interoperability;
• foreclosure;
• leveraging.

17. Vertical Integration

An AI-controlled enterprise could operate vertically across an entire chain.

For example:

AI chip → cloud → foundation model → AI agent → marketplace → logistics → consumer service.

This creates potential concerns where control at one level is used to reinforce market power at another level.

The relevant theories may include:

• foreclosure;
• tying;
• bundling;
• self-preferencing;
• discriminatory access;
• exclusive dealing;
• interoperability restrictions.

18. Machine-Owned Enterprises and Labour Markets

Autonomous enterprises could also change labour-market competition.

If AI systems replace large portions of conventional labour, competition authorities may need to examine:

• concentration among employers;
• algorithmic wage-setting;
• worker allocation;
• automated recruitment;
• labour-platform restrictions;
• non-compete mechanisms;
• sharing of wage information.

An AI system that independently determines compensation across a large labour platform could create issues analogous to algorithmic price coordination on the consumer side.

19. Antitrust Liability and Attribution

A useful framework is:

Step 1 — Identify the machine

What AI or autonomous system made the decision?

Step 2 — Identify its legal owner

Who owns or controls it?

Step 3 — Identify the undertaking

Which economic entity conducts the business?

Step 4 — Examine human involvement

Was the conduct:

• instructed;
• anticipated;
• authorised;
• tolerated;
• monitored; or
• deliberately engineered?

Step 5 — Examine market effects

Did the behaviour:

• raise prices;
• restrict output;
• exclude competitors;
• reduce innovation;
• discriminate;
• prevent entry?

Step 6 — Apply the relevant competition rule

Possible provisions include:

• cartel prohibition;
• abuse of dominance;
• merger control;
• vertical restraints;
• discriminatory access;
• exclusionary conduct.

20. Evidence in Machine-Antitrust Cases

Traditional evidence may be inadequate.

Competition authorities may need access to:

• source code;
• model architecture;
• training data;
• system prompts;
• API logs;
• model outputs;
• reinforcement-learning records;
• audit trails;
• transaction histories;
• deployment instructions;
• model versions;
• system changes.

A major evidentiary question will be:

Can the authority prove why the AI took a particular competitive decision?

This makes algorithmic explainability and auditability increasingly important.

21. Compliance Duties for Machine-Owned Enterprises

Businesses deploying autonomous commercial systems should consider:

1. Competition-by-design

Competition risks should be assessed before deployment.

2. Algorithmic audit

Regularly test whether the system:

• coordinates prices;
• discriminates against competitors;
• excludes entrants;
• shares sensitive information.

3. Data controls

Prevent the autonomous system from receiving unnecessary competitor-sensitive information.

4. Human oversight

Maintain meaningful human supervision over high-risk competitive decisions.

5. Logging

Preserve:

• inputs;
• outputs;
• model versions;
• decisions;
• modifications.

6. Kill-switch mechanisms

A company should be able to stop an autonomous system producing unlawful outcomes.

7. Independent testing

Test the system against simulated cartel and exclusion scenarios.

22. Possible Antitrust Remedies

Traditional fines may remain available, but autonomous systems create additional remedial possibilities.

Behavioural remedies

Authorities could require:

• removal of certain pricing inputs;
• restrictions on competitor data;
• interoperability;
• algorithmic audits;
• disclosure requirements.

Structural remedies

In extreme cases:

• divestiture;
• separation of business units;
• separation of data assets;
• infrastructure access obligations.

Technological remedies

Authorities might require:

redesign of an algorithm so that it cannot use competitors' confidential information to generate prices.

The RealPage resolution illustrates how competition remedies can target the architecture and information inputs of pricing systems, rather than merely imposing a financial penalty.

23. Important Legal Distinction: Machine Autonomy ≠ Corporate Independence

A machine can be technologically autonomous without being legally independent.

For example:

AI makes 99% of the decisions.

That does not necessarily mean:

AI = undertaking.

Conversely:

Human managers make only 1% of the decisions.

That does not necessarily eliminate corporate responsibility.

Competition law traditionally focuses on the economic undertaking and its conduct, rather than simply on which physical entity pressed the button.

24. Comparative Position

25. Six Core Case Laws at a Glance

The first six provide the strongest starting point for an examination answer specifically focused on machine-operated enterprises and antitrust. The algorithmic-pricing cases demonstrate that competition law is already being applied to automated decision systems, even though a fully autonomous machine enterprise raises questions that existing case law has not definitively resolved.

26. Emerging Doctrine: From Algorithmic Competition to Machine Competition

The future problem is broader than algorithmic pricing.

A genuinely autonomous enterprise could potentially decide:

Market entry → acquisition → pricing → procurement → advertising → hiring → contracting → investment → exit

without human intervention in individual decisions.

Competition law may consequently need to move from asking only:

"What did the company instruct its employees to do?"

toward asking:

"What autonomous commercial system did the undertaking deploy, what market conditions did it create, and who legally controlled or benefited from that system?"

This does not mean that autonomous machine behaviour should automatically be treated as unlawful. Lawful optimisation, dynamic pricing and automated contracting can produce substantial efficiencies. The central antitrust distinction remains between competitive automation and automation that facilitates unlawful coordination or exclusion.

27. Conclusion

Machine-owned enterprises represent a major conceptual challenge for antitrust law because technological autonomy can become separated from human decision-making while economic power remains concentrated in identifiable business ecosystems.

The most important principles are:

1. AI is not automatically a separate legal undertaking.
2. The owner, operator or controlling enterprise may remain legally responsible.
3. Algorithms can facilitate conventional cartels.
4. Common platforms can create hub-and-spoke concerns.
5. Algorithmic price similarity alone does not necessarily establish collusion.
6. Autonomous learning creates the difficult problem of tacit machine coordination.
7. Dominant AI systems may create exclusion, tying, self-preferencing and access problems.
8. Data, compute, models and network effects can become sources of market power.
9. AI acquisitions may raise innovation and potential-competition concerns.
10. Competition compliance increasingly requires algorithmic auditing, data controls, logging and human oversight.

The central legal challenge is therefore not simply "Can a machine violate antitrust law?" It is:

How should competition law attribute, investigate and remedy economically significant conduct generated by increasingly autonomous commercial machines?