Competition Law and Strategic Competition Policy for Autonomous Markets

1. Introduction

The expression “autonomous markets” can be understood in modern competition policy as markets in which important commercial decisions are increasingly made or executed by algorithms, AI systems, autonomous agents, platforms, or automated decision-making systems rather than directly by human managers.

Examples include:

AI-driven pricing;

autonomous procurement;

algorithmic trading;

automated advertising;

dynamic pricing;

autonomous supply-chain management;

AI purchasing agents;

automated recommendation systems;

autonomous logistics;

digital marketplaces; and

agentic AI systems capable of negotiating or transacting on behalf of users.

This creates a fundamental competition-law question:

When market decisions are delegated to autonomous systems, how should competition law preserve rivalry when the firms' behaviour may be coordinated, discriminatory, exclusionary, or concentrated without traditional human-to-human communication?

The OECD's recent work specifically identifies agentic AI, algorithmic collusion, data access, model restrictions, vertical integration and market contestability as emerging competition-policy issues. (OECD)

2. Meaning of an Autonomous Market

An autonomous market is not necessarily a market without human participants.

Rather, it is a market where significant competitive decisions are delegated to systems capable of:

collecting market information;

predicting demand;

changing prices;

selecting suppliers;

negotiating terms;

allocating resources;

ranking products;

targeting consumers;

responding to competitors; and

potentially taking actions without obtaining fresh human approval.

For example:

Human objective → AI agent → market observation → autonomous decision → transaction

The competition-law difficulty is that the final market behaviour may not correspond neatly to a conventional managerial decision.

3. Why Autonomous Markets Challenge Competition Law

Traditional competition law generally asks:

Who made the decision, what agreement existed, and what conduct resulted?

Autonomous markets introduce additional questions:

Who designed the algorithm?

Who trained the AI system?

Who supplied its data?

Who set its objectives?

Who controlled its parameters?

Did different AI systems communicate?

Did they independently reach similar outcomes?

Can the firm predict what its agent will do?

Who is legally responsible for the agent's conduct?

These questions create what can be called the attribution problem.

The OECD has specifically identified attribution of liability as an area requiring further competition-policy development for agentic AI. (OECD)

4. Strategic Competition Policy

A strategic competition policy for autonomous markets should pursue six principal objectives:

1. Contestability

New firms must be able to enter.

2. Competitive neutrality

Autonomous systems should not be designed to systematically favour incumbents without legitimate justification.

3. Interoperability

AI agents and digital platforms should be capable, where appropriate, of interacting with competing systems.

4. Data access

Critical data advantages should not become unjustifiable barriers to entry.

5. Algorithmic accountability

Businesses should retain responsibility for competition-sensitive decisions delegated to algorithms.

6. Innovation

Competition policy should avoid regulating autonomous technology so heavily that innovation itself is suppressed.

5. Major Competition Concerns

A. Autonomous Algorithmic Collusion

This is one of the most important concerns.

Suppose:

Firm A uses Agent A;

Firm B uses Agent B.

Both agents continuously observe the market and independently discover that maintaining high prices produces greater profits.

Neither company directly communicates with the other.

The agents repeatedly maintain elevated prices.

This creates the difficult question:

Can autonomous coordination amount to a competition-law violation without traditional human communication?

The OECD distinguishes between explicit algorithm-facilitated collusion, hub-and-spoke coordination and more autonomous forms of algorithmic coordination. (OECD)

6. Case Law 1 — United States v. Topkins

Facts

Online sellers agreed to fix prices for posters sold through Amazon Marketplace.

Pricing software was used to implement the agreed pricing strategy.

Legal significance

The case is important because the algorithm was not treated as an independent source of legality.

The underlying human agreement remained relevant.

Principle

Delegating implementation of a cartel to software does not transform cartel conduct into lawful independent pricing.

Relevance to autonomous markets

This is the simplest model:

Human agreement → algorithmic implementation → anti-competitive outcome

Competition law can apply using traditional cartel principles.

7. Case Law 2 — Eturas UAB v Lietuvos Respublikos Konkurencijos Taryba

Facts

Eturas operated an online travel-booking platform.

A system message was sent to travel agencies indicating a restriction concerning discounts that could be offered through the platform.

CJEU approach

The Court considered when knowledge of an electronically communicated restriction, combined with subsequent market behaviour, could establish participation in a concerted practice.

Principle

Electronic systems can provide evidence of concertation, but the existence of an automated message alone does not automatically establish liability.

Importance

The case is particularly relevant to autonomous markets because it demonstrates that:

digital communication can constitute evidence of coordination even where competitors do not negotiate through conventional face-to-face meetings.

8. Case Law 3 — Trod Ltd and GB Eye Ltd v CMA

Facts

Two online sellers agreed to maintain minimum prices for posters sold through Amazon Marketplace.

They used automated repricing software to implement their agreement.

Competition significance

The UK Competition and Markets Authority treated the arrangement as a competition-law infringement.

Principle

Automation does not eliminate liability where businesses deliberately configure technology to implement an anti-competitive agreement.

Autonomous-market relevance

It establishes an important distinction:

Autonomous execution ≠ autonomous legal responsibility.

Where humans deliberately configure a system to produce anti-competitive results, traditional competition law remains applicable.

9. Case Law 4 — Samir Agrawal v Competition Commission of India

Facts

The case concerned allegations that Ola and Uber's algorithmic pricing mechanisms facilitated price coordination among drivers.

The allegation was essentially that drivers were deprived of independent price-setting because the platform's algorithm determined the relevant fare.

The CCI did not find a prima facie case on the facts presented, and the matter ultimately reached the Supreme Court through the appellate process. The reported proceedings discuss the distinction between algorithmic pricing and the agreement necessary to establish cartel conduct. (Indian Kanoon)

Importance

The case is highly relevant to autonomous markets because it demonstrates that:

algorithmic uniformity alone is not necessarily proof of an unlawful agreement.

Strategic lesson

Competition authorities must distinguish:

independent algorithmic optimisation;

platform-imposed pricing;

coordinated algorithmic behaviour; and

genuine cartel arrangements.

10. Case Law 5 — United States v. Apple

Context

The US antitrust litigation concerning Apple's conduct in digital markets illustrates the broader problem of ecosystem control.

Issues surrounding Apple's ecosystem include:

app distribution;

payment mechanisms;

developer access;

interoperability;

platform rules.

Autonomous-market relevance

Autonomous agents increasingly depend upon digital ecosystems.

For example:

AI agent → app store → payment system → cloud service → merchant.

If one platform controls several of these stages, it may possess opportunities to:

restrict access;

favour its own services;

impose discriminatory conditions;

increase switching costs.

The strategic lesson is that competition policy should examine ecosystem architecture, not simply individual transactions.

11. Case Law 6 — Google Shopping

Facts

The European Commission found that Google systematically favoured its own comparison-shopping service in general search results.

The General Court largely upheld the Commission's decision.

Principle

Control over a major digital gateway can permit a dominant undertaking to favour its own downstream service.

Autonomous-market relevance

AI agents may increasingly decide:

which products users see;

which suppliers are contacted;

which payment service is selected;

which hotel is booked;

which cloud provider is used.

If the agent is controlled by a dominant platform, agentic self-preferencing becomes a potential future competition concern.

12. Case Law 7 — Microsoft

Facts

Microsoft was found to have abused its dominant position through conduct involving interoperability information and product tying.

Principle

Control over a technological ecosystem can be leveraged into adjacent markets.

Relevance to autonomous markets

Imagine an AI agent that operates exclusively within one company's ecosystem:

AI agent → proprietary operating system → proprietary cloud → proprietary payment system.

If competitors cannot effectively interact with the agent, interoperability restrictions may become a competitive bottleneck.

13. Case Law 8 — Deutsche Telekom

Facts

Deutsche Telekom was found to have engaged in a margin squeeze concerning telecommunications access.

Relevance

Autonomous markets require infrastructure.

AI agents depend upon:

networks;

cloud computing;

data centres;

APIs;

payment infrastructure.

If a vertically integrated infrastructure provider uses its position to make downstream competition commercially difficult, traditional competition principles concerning exclusionary conduct remain relevant.

14. Case Law 9 — Bronner v Mediaprint

Facts

A newspaper publisher sought access to a rival's distribution infrastructure.

Principle

Refusal to supply becomes an abuse only under demanding circumstances, including where the facility is indispensable and no realistic alternative exists.

Autonomous-market significance

This principle could become important where AI agents depend upon:

proprietary APIs;

data infrastructure;

cloud systems;

payment networks;

digital identity systems.

Not every useful technological resource is automatically an essential facility.

15. Case Law 10 — Intel v Commission

Facts

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

The CJEU required attention to the capability of the rebates to foreclose an equally efficient competitor.

Autonomous-market relevance

AI platforms may provide:

preferential access;

computing discounts;

developer incentives;

API rebates;

cloud credits.

A dominant AI ecosystem could potentially use such commercial mechanisms to restrict rival systems.

The relevant question is not simply:

"Was a discount offered?"

but:

"Could the conduct foreclose effective competitors?"

16. Autonomous Agents and Hub-and-Spoke Competition

A particularly important model is:

Competitor A

↓

Common AI provider

↓

Competitor B

↓

Competitor C

The common AI provider may:

collect competitor information;

optimise prices;

recommend pricing;

observe demand;

predict competitor behaviour.

This can create a hub-and-spoke risk.

The crucial question is whether the common system merely provides an independent technological service or becomes a mechanism through which competitors coordinate competitively sensitive behaviour.

17. Autonomous Collusion Without Human Agreement

The most difficult hypothetical is:

Agent A → observes Agent B

Agent B → observes Agent A

↓

Both independently learn that cooperation produces higher profits

↓

Both maintain supra-competitive prices

No human explicitly agrees.

This is sometimes described as autonomous algorithmic collusion.

Current competition policy does not simply equate price convergence with unlawful collusion. The OECD notes that the actual legal treatment remains an emerging area, with relatively few completed antitrust cases specifically involving AI-enabled autonomous conduct. (OECD ONE)

Therefore evidence would be needed concerning:

system design;

information flows;

training;

instructions;

communications;

incentives;

predictability;

human involvement.

18. Autonomous Pricing and Consumer Harm

AI agents can dynamically alter prices based on:

demand;

inventory;

competitor prices;

consumer behaviour;

location;

purchasing history.

Potential concerns include:

Personalised pricing

Different consumers receive different prices.

Algorithmic discrimination

Prices are adjusted using sensitive or proxy information.

Coordinated pricing

Agents respond to each other in ways that soften competition.

Price transparency

Consumers may not understand why prices change.

Competition law must distinguish legitimate dynamic pricing from conduct that actually restricts competition.

19. Autonomous Purchasing Agents

A new competition issue is the rise of consumer AI agents.

Instead of consumers directly comparing:

Amazon vs Walmart vs local retailer,

an AI agent may independently:

search products;

compare prices;

evaluate reviews;

negotiate;

select a seller;

complete payment.

This could potentially increase competition because the agent reduces search costs.

But the agent itself could become a new gatekeeper.

For example:

80% of consumers use Agent X → Agent X determines which sellers receive visibility.

The competition problem then shifts from:

platform dominance

to:

agent-mediated gatekeeping.

20. Autonomous Agents as New Gatekeepers

An autonomous agent may control the customer's purchasing journey.

Potential forms of self-preferencing include:

ranking affiliated merchants first;

recommending the agent's own products;

using proprietary data to disadvantage competitors;

restricting competing payment methods;

steering users toward affiliated services.

Thus:

AI agent + network effects + proprietary data = potential new gatekeeper structure.

21. Data Concentration

Autonomous markets are highly data-intensive.

Competitive advantages may derive from:

transaction data;

real-time price information;

consumer preferences;

behavioural data;

supplier information;

historical transactions.

If one firm controls the data necessary to develop competitive autonomous agents, entry barriers can increase.

Potential remedies include:

data portability;

interoperability;

access obligations in appropriate circumstances;

data-sharing arrangements;

privacy-compatible datasets.

22. Autonomous Markets and Interoperability

Interoperability may become one of the most important strategic competition tools.

Imagine:

Agent A

cannot communicate with

Platform B

because Platform B deliberately restricts its API.

If Platform B is dominant, the restriction could increase switching costs and reinforce ecosystem power.

However, interoperability mandates must account for:

cybersecurity;

privacy;

intellectual property;

system integrity.

23. Autonomous Markets and Vertical Integration

A large AI company might control:

Chips → Cloud → Foundation Model → AI Agent → Marketplace.

Vertical integration may create efficiencies.

But it can also create foreclosure opportunities.

For example, the company could:

favour its own model;

restrict rival models;

give its agent privileged access;

restrict competing agents;

bundle cloud and AI services.

The strategic competition framework should therefore assess the entire value chain.

24. Autonomous Markets and Merger Control

Traditional turnover-based merger thresholds may miss acquisitions of small AI companies.

A startup may have:

little revenue;

exceptional technology;

valuable data;

talented researchers;

rapidly growing users.

A strategic merger framework should therefore consider:

innovation;

potential competition;

data;

technology;

future market entry;

ecosystem effects.

The objective is not to prohibit innovation-enhancing acquisitions automatically, but to ensure that potentially significant competitive constraints are not overlooked.

25. Autonomous Markets and Consumer Welfare

Competition policy must consider multiple dimensions of consumer welfare.

Price

Will autonomous systems reduce or increase prices?

Quality

Will AI improve service quality?

Choice

Will consumers have meaningful alternatives?

Privacy

Will autonomous agents exploit personal data?

Innovation

Will independent AI developers continue to enter?

Reliability

Will automated decisions produce systemic failures?

Thus consumer welfare should not be reduced to immediate price effects alone.

26. Autonomous Markets and Small Businesses

Autonomous systems can potentially lower entry barriers.

A startup can use AI for:

accounting;

marketing;

logistics;

procurement;

customer support;

product design.

This can reduce minimum efficient scale and make smaller firms more competitive. The OECD has specifically identified AI's potential to lower entry barriers, reduce minimum efficient scale and support product differentiation. (OECD)

But dependence upon a few AI providers can create new barriers.

For example:

Startup → dependent on one cloud provider → dependent on one model provider → dependent on one AI marketplace.

27. Strategic Regulatory Framework

A suitable competition-policy framework can be structured as follows:

Stage 1 — Identify the autonomous system

What decisions does the system make?

Stage 2 — Identify the market

Which product, geographic and technological markets are affected?

Stage 3 — Identify the controller

Who:

owns;

designs;

trains;

deploys; and

supervises

the system?

Stage 4 — Examine information flows

What data does the system receive from:

competitors;

customers;

suppliers;

platforms?

Stage 5 — Analyse market power

Consider:

network effects;

data;

switching costs;

scale;

compute;

APIs;

ecosystem control.

Stage 6 — Examine conduct

Look for:

collusion;

self-preferencing;

exclusion;

tying;

discriminatory access;

refusal to deal;

predatory conduct.

Stage 7 — Assess effects

Consider:

prices;

innovation;

entry;

quality;

consumer choice;

resilience.

Stage 8 — Select proportionate remedies

Potential remedies include:

interoperability;

portability;

transparency;

access;

behavioural commitments;

structural remedies where justified.

28. Compliance Framework for Businesses

Companies deploying autonomous systems should maintain:

Algorithmic competition audit

Regularly test whether algorithms produce exclusionary or coordinated outcomes.

Human oversight

Maintain meaningful supervision over competition-sensitive decisions.

Data governance

Identify whether competitor-sensitive information is being used.

Model governance

Document:

training;

instructions;

objectives;

constraints.

Competition controls

Prevent agents from receiving instructions that could facilitate:

price fixing;

market allocation;

customer allocation;

bid coordination.

Monitoring

Test whether autonomous systems behave differently from their intended parameters.

29. Strategic Competition Policy Matrix

30. Important Distinction: Automation Is Not Automatically Anticompetitive

A crucial legal principle is:

Autonomous decision-making is not itself an antitrust violation.

AI can generate substantial competitive benefits.

It may:

lower prices;

improve forecasting;

reduce transaction costs;

improve inventory management;

increase product variety;

lower entry costs;

improve consumer search.

The competition issue arises from how the technology is designed, deployed and used, not simply from the fact that it is autonomous.

31. Six Major Case-Law Principles

32. Future Direction of Competition Policy

The future competition framework for autonomous markets is likely to move from:

"Did humans agree?"

toward a more sophisticated inquiry:

Who designed the autonomous decision system, what information did it receive, what objectives was it given, what competitive effects could reasonably be anticipated, and what degree of control or responsibility remained with the undertaking?

This does not mean abandoning traditional competition law.

Rather, existing concepts—agreement, concerted practice, dominance, foreclosure, tying, refusal to deal, essential facilities and merger control—must be applied to technologically different mechanisms.

Current OECD analysis similarly indicates that many AI competition problems are extensions of familiar theories of harm, while attribution and genuinely autonomous behaviour present newer challenges. (OECD)

33. Conclusion

Strategic competition policy for autonomous markets requires competition law to adapt to a world in which algorithms and AI agents increasingly act as economic decision-makers.

The principal concerns are:

autonomous algorithmic collusion;

hub-and-spoke coordination;

AI-enabled price discrimination;

agentic self-preferencing;

data concentration;

technological lock-in;

interoperability restrictions;

AI-platform dominance;

autonomous purchasing gatekeepers;

exclusionary vertical integration; and

acquisitions of emerging AI competitors.

The cases of Topkins, Eturas, Trod/GB Eye and Samir Agrawal demonstrate that automation does not remove traditional competition-law responsibility. Google Shopping, Microsoft, Deutsche Telekom, Bronner and Intel provide broader principles concerning digital gatekeepers, interoperability, infrastructure and exclusionary conduct.

The central proposition is:

Competition law should regulate the competitive effects of autonomous systems without treating autonomy itself as unlawful. The objective should be to preserve contestability, independent decision-making, innovation, interoperability and consumer choice while ensuring that firms cannot use autonomous technology as a mechanism for cartelisation, exclusion or the entrenchment of market power.

Importantly, the current case law remains more developed for algorithm-assisted conduct than for genuinely autonomous AI agents acting without meaningful human direction; that distinction is likely to become increasingly important as agentic systems are deployed more widely. (OECD)