Competition Law and Strategic Foresight for Competition Policy in Autonomous Markets

Introduction

Strategic foresight in competition law means anticipating how markets, technologies, business models and sources of market power may develop before conventional antitrust indicators become fully visible. In autonomous markets, this becomes particularly important because algorithms, artificial intelligence, automated agents and platform infrastructures can increasingly make pricing, ranking, matching, investment, procurement and distribution decisions with limited human intervention.

Competition authorities therefore face a shift from asking only:

“Has competition already been harmed?”

towards also asking:

“What market structure, technological capability or governance mechanism could make competition difficult to restore in the future?”

The OECD has specifically identified algorithmic pricing as capable of producing both efficiency-enhancing competition and risks of coordination, information exchange and anticompetitive conduct. Its 2025 G7 work also identifies autonomous learning and algorithmic pricing as emerging competition-policy challenges.

Strategic foresight does not mean regulating every new technology merely because it could become powerful. It requires competition authorities to identify credible future scenarios, detect early indicators of market power and preserve contestability while allowing beneficial innovation.

1. Meaning of Autonomous Markets

An autonomous market is a market in which important competitive decisions are increasingly delegated to automated systems.

Examples include:

• algorithmic pricing;
• AI-based product recommendations;
• automated advertising auctions;
• autonomous procurement;
• algorithmic credit allocation;
• ride-hailing matching;
• dynamic electricity pricing;
• automated financial trading;
• AI-driven search rankings;
• platform seller rankings;
• automated inventory management;
• autonomous logistics;
• AI agents negotiating transactions.

The distinguishing characteristic is not simply the use of software. It is the delegation of commercially significant decision-making to systems capable of continuously processing data and adapting their behaviour.

This can produce a market in which:

Data → Algorithm → Prediction → Decision → Market response → New data → Algorithmic adaptation

The cycle can operate much faster than traditional human decision-making.

2. Why Strategic Foresight Matters

Traditional competition law is largely reactive.

A conventional investigation may proceed through:

1. market definition;
2. identification of market power;
3. identification of conduct;
4. assessment of effects;
5. determination of infringement;
6. imposition of remedies.

Autonomous markets complicate every stage.

A. Market boundaries may change rapidly

An AI platform may compete simultaneously in:

• search;
• advertising;
• cloud computing;
• data;
• payments;
• software;
• content;
• commerce.

Consequently, today's relevant market may not accurately describe tomorrow's competitive constraint.

B. Market power may come from infrastructure

A company may possess relatively little conventional physical infrastructure while controlling:

• data;
• APIs;
• operating systems;
• cloud infrastructure;
• model access;
• ranking systems;
• identity systems;
• payment rails;
• interoperability standards.

C. Conduct can become autonomous

An algorithm may automatically:

• increase prices;
• reduce discounts;
• rank products;
• exclude suppliers;
• allocate customers;
• optimise advertising;
• match competitors' prices.

The traditional distinction between intentional human conduct and automated conduct therefore becomes more difficult to apply.

3. Strategic Foresight as a Competition-Law Method

Strategic foresight should not replace conventional antitrust analysis. It should operate as an early-warning and policy-development layer.

A useful framework is:

Horizon Scanning

↓

Identify Emerging Technologies

↓

Identify Potential Bottlenecks

↓

Develop Alternative Market Scenarios

↓

Identify Early Warning Indicators

↓

Test Competitive Resilience

↓

Choose Proportionate Intervention

↓

Monitor Market Evolution

This allows authorities to intervene before an irreversible concentration of market power occurs.

4. Key Competition Risks in Autonomous Markets

A. Algorithmic Collusion

Autonomous pricing systems can make it easier for competing firms to observe and respond to each other's prices.

The OECD has recognised several risks:

• algorithms facilitating explicit agreements;
• hub-and-spoke coordination;
• information exchange;
• algorithmic monitoring;
• tacit coordination through autonomous learning. 

The crucial legal question becomes:

When does automated parallel behaviour remain independent competition, and when does it constitute unlawful coordination?

B. Algorithmic Hub-and-Spoke Arrangements

A common algorithmic intermediary can become the mechanism through which competitors coordinate.

For example:

Competitor A → shared algorithm → Competitor B

The algorithm may receive competitively sensitive information from multiple firms and generate recommendations that influence each participant's conduct.

The intermediary can therefore perform a function similar to a traditional coordinating hub.

5. Case Law

Case 1: United States v. David Topkins

This is one of the clearest early examples of algorithm-assisted price fixing.

Topkins and co-conspirators sold posters through Amazon Marketplace. They agreed to coordinate prices and implemented the agreement through pricing algorithms. The DOJ described the prosecution as its first criminal prosecution specifically involving an online marketplace conspiracy.

Competition-law significance

The case establishes an important foresight principle:

Technology does not immunise an otherwise conventional cartel.

The fact that the agreement was implemented through computer code did not transform price fixing into legitimate automated competition.

It demonstrates that authorities must investigate:

• algorithm design;
• instructions given to algorithms;
• communications between competitors;
• source code;
• pricing rules;
• data inputs.

Case 2: Trod Ltd / GB eye Ltd — Amazon Marketplace

The UK's Competition and Markets Authority investigated two sellers of posters and frames on Amazon Marketplace.

The businesses agreed not to undercut one another and used automated repricing software to implement the arrangement. The CMA imposed a fine on one seller.

Significance

This case illustrates the transition from:

human cartel → software-assisted cartel → potentially self-executing cartel.

Once programmed, the algorithm can continuously enforce the competitive restriction.

For future enforcement, this means that authorities should examine not only communications between executives but also the architecture of automated decision-making systems.

Case 3: Eturas UAB and Others v Lithuanian Competition Council

Case C-74/14, Court of Justice of the European Union, 2016

Eturas involved travel agencies using a common electronic booking system.

The system administrator communicated a restriction on discounts, and the booking system automatically limited discounts available to customers. The CJEU examined whether the circumstances could establish a concerted practice under Article 101 TFEU.

The Court also addressed the evidentiary question: mere receipt of an automated system message was not automatically sufficient to establish participation; the relevant presumption had to remain rebuttable.

Significance

Eturas is particularly important for autonomous markets because it demonstrates that:

• software architecture can facilitate coordination;
• automated restrictions can have competition-law consequences;
• evidence concerning user knowledge and participation remains important;
• automated behaviour cannot simply be attributed mechanically to every system participant.

Case 4: Samir Agrawal v. Competition Commission of India

The Indian litigation concerning Ola and Uber addressed allegations that algorithmic pricing facilitated price coordination among drivers.

The allegation was that the platforms' algorithms determined fares and restricted the ability of individual drivers to compete independently on price. The CCI did not find a prima facie case on the material before it, and the matter subsequently reached the appellate and Supreme Court proceedings.

Significance

The case demonstrates a critical distinction:

Algorithmic pricing ≠ automatically unlawful price fixing.

An authority must establish the legal elements necessary to connect automated pricing with a prohibited agreement or concerted practice.

For strategic foresight, however, the case is important because it highlights the need to understand:

• algorithmic fare determination;
• platform-driver relationships;
• platform governance;
• data inputs;
• pricing autonomy;
• contractual restrictions.

Case 5: Google Shopping

Google LLC and Alphabet Inc. v European Commission, Case C-48/22 P, 2024

The Google Shopping litigation concerned Google's treatment of its own comparison-shopping service in general search results.

The CJEU's 2024 judgment considered the relationship between Google's dominant general-search position and its treatment of competing comparison-shopping services, including questions concerning foreclosure capability, causal effects and the appropriate counterfactual analysis.

Significance for autonomous markets

The case illustrates how algorithmic ranking infrastructure can become a source of competitive power.

A platform controlling an essential gateway may influence competition through:

• ranking;
• visibility;
• recommendation;
• search prominence;
• traffic allocation.

The foresight question therefore becomes:

Who controls the algorithmic gateway through which competitors reach consumers?

This can be more important than merely measuring market share.

Case 6: United States v. RealPage

The RealPage litigation represents a particularly important modern development in algorithmic pricing.

The U.S. Department of Justice alleged that competing landlords supplied non-public, competitively sensitive information to RealPage, whose software used that information to generate rental-price recommendations. The government alleged violations of Sections 1 and 2 of the Sherman Act.

The case has subsequently generated extensive litigation and settlements involving participating property-management companies.

Significance

RealPage illustrates a more sophisticated problem than conventional cartel conduct:

Data → central algorithm → recommendations → independent firms → coordinated market outcomes

The competitive concern is therefore potentially embedded in the information architecture of the market.

For competition authorities, this means that investigation may need to focus on:

• what data firms provide;
• whether the data are competitively sensitive;
• who controls the algorithm;
• how recommendations are generated;
• whether firms retain meaningful independent decision-making;
• whether the system makes deviation from coordinated outcomes less attractive.

Case 7: United States v. Apple — E-books

In the Apple e-books litigation, the Second Circuit affirmed findings that Apple orchestrated a conspiracy among publishers concerning e-book prices. The case involved digital distribution and contractual mechanisms that restricted retailers' pricing freedom.

Significance

Although not itself an autonomous-AI case, it is relevant to strategic foresight because it demonstrates how digital architecture and contractual design can reshape price competition.

Competition authorities should therefore examine not merely algorithmic behaviour but the interaction between:

• algorithms;
• contracts;
• platform rules;
• access conditions;
• technical design.

6. Amazon and Algorithmic Governance

The U.S. FTC's case against Amazon provides another useful example of the broader issue.

The FTC and state plaintiffs allege that Amazon used interconnected strategies affecting seller pricing, search visibility, fulfilment and competition. Among the allegations is that Amazon's systems could penalise sellers offering lower prices elsewhere and that Amazon's search systems favoured its own products. The litigation remains ongoing.

This illustrates an emerging concept:

Algorithmic governance as a source of market power

A platform may not merely participate in a market.

It may design the rules through which other market participants compete.

That distinction is central to future competition policy.

7. From Market Share to Algorithmic Control

Traditional competition analysis frequently considers:

• market share;
• barriers to entry;
• concentration;
• customer power;
• switching costs.

Autonomous markets require additional indicators.

Possible indicators include:

8. Data as a Strategic Competition Asset

Autonomous markets depend heavily on data.

A platform can obtain:

More users → more data → better model → better service → more users → more data

This creates a data feedback loop.

The competition problem arises when the loop becomes difficult for rivals to replicate.

Potential concerns include:

• exclusive access to strategically important data;
• refusal to provide interoperability;
• discriminatory API access;
• data portability restrictions;
• tying data access to other services;
• acquisition of emerging data-driven competitors.

Strategic foresight therefore requires authorities to ask whether today's data advantage could become tomorrow's structural monopoly.

9. AI Agents and Autonomous Purchasing

A particularly significant future development is the AI agent acting on behalf of consumers.

Instead of a consumer manually comparing products:

Consumer → AI agent → search → recommendation → negotiation → purchase

This could reduce traditional search costs.

But it could also create a new intermediary layer.

If a small number of AI agents control:

• consumer discovery;
• recommendations;
• payments;
• purchasing;
• switching;

then competition may shift from competition for consumers to competition for access to AI agents.

This creates new questions concerning:

• neutrality;
• self-preferencing;
• transparency;
• interoperability;
• exclusive arrangements;
• access conditions.

10. Autonomous Collusion

One of the most difficult future problems is machine-generated coordination.

Suppose several competing firms use reinforcement-learning systems.

The systems may independently learn:

“If I increase price after my rival increases price, my long-term revenue improves.”

No human employee expressly agrees to coordinate.

This creates three possible categories:

1. Explicit coordination

Humans agree and algorithms implement the agreement.

Traditional antitrust principles remain strongly applicable.

2. Algorithm-assisted coordination

Human firms knowingly design systems that facilitate coordination.

Existing competition rules may also apply, depending on the evidence.

3. Emergent autonomous coordination

Independent AI systems learn parallel strategies without an express human agreement.

This is considerably more difficult.

The legal challenge is distinguishing:

lawful interdependent behaviour

from

conduct sufficiently attributable to firms to satisfy competition-law requirements.

The OECD has specifically identified autonomous learning as an area requiring competition-policy attention.

11. Strategic Foresight and Merger Control

Merger control becomes particularly important in autonomous markets because an acquisition may eliminate a future competitor before the competitive threat becomes visible in conventional market shares.

Authorities should examine:

• AI start-ups;
• data assets;
• foundation models;
• cloud infrastructure;
• specialised algorithms;
• developer ecosystems;
• distribution channels;
• intellectual property;
• talent;
• interoperability technologies.

A small company may have low current revenue but possess technology capable of becoming an important competitive constraint.

Thus:

Current turnover ≠ necessarily future competitive significance.

12. Killer Acquisitions and Innovation Competition

Strategic foresight requires attention to innovation competition.

An established platform may acquire a small AI company because:

1. the target is an emerging competitor;
2. its technology threatens an incumbent ecosystem;
3. its data could strengthen another competitor;
4. its technology could become a future substitute.

Competition policy must therefore consider future competitive constraints, not merely current substitution.

13. Interoperability as a Competition Tool

Autonomous ecosystems can become closed systems.

For example:

Operating system → AI assistant → cloud → payment → marketplace → identity

If every layer is controlled by one ecosystem, rivals may find it difficult to enter.

Interoperability can therefore become a competition-policy instrument.

Potential remedies include:

• API access;
• data portability;
• interoperability obligations;
• non-discrimination rules;
• technical access standards;
• switching mechanisms.

14. Transparency and Explainability

Competition authorities increasingly need technical understanding of algorithmic systems.

Important evidence may include:

• source code;
• model documentation;
• training datasets;
• model cards;
• system logs;
• API documentation;
• decision rules;
• audit trails;
• experimentation records;
• pricing histories.

However, transparency should not automatically mean disclosure of trade secrets or source code.

The objective is to enable effective competition enforcement while protecting legitimate innovation and confidentiality.

15. Competition by Design

A future-oriented approach may require competition by design.

This means considering competitive effects during the design of:

• platforms;
• AI systems;
• digital marketplaces;
• APIs;
• ranking systems;
• recommendation engines;
• interoperability standards.

Instead of waiting for a mature monopoly, authorities can identify design features likely to create durable exclusion.

Examples include:

• discriminatory rankings;
• artificial switching barriers;
• exclusive data access;
• self-preferencing;
• discriminatory interoperability;
• default arrangements;
• technical restrictions preventing multi-homing.

16. Dynamic Market Definition

Autonomous markets may require greater emphasis on dynamic competitive constraints.

Instead of asking only:

What products compete today?

authorities should also consider:

What technologies could become substitutes tomorrow?

Relevant evidence may include:

• innovation pipelines;
• venture investment;
• R&D capabilities;
• technological convergence;
• consumer switching;
• emerging business models;
• interoperability;
• potential entrants.

This does not eliminate conventional market definition. It supplements it with a forward-looking assessment.

17. Regulatory Sandboxes and Competition Monitoring

Strategic foresight can be implemented through controlled regulatory experimentation.

Authorities could establish:

Competition sandboxes

Businesses developing autonomous systems could voluntarily provide information regarding:

• algorithmic architecture;
• pricing mechanisms;
• data usage;
• interoperability;
• competitive safeguards.

This could allow authorities to identify risks before they mature into enforcement cases.

18. Early-Warning Indicators

Competition authorities could develop an Autonomous Market Competition Dashboard.

Structural indicators

• increasing concentration;
• declining entry;
• increasing switching costs;
• dependency on a single API;
• dependency on one cloud provider.

Algorithmic indicators

• increasingly identical pricing;
• rapid parallel price movements;
• reduced price dispersion;
• algorithmic retaliation against discounts;
• coordinated ranking changes.

Data indicators

• exclusive data access;
• increasing data accumulation;
• refusal of interoperability;
• discriminatory API access.

Innovation indicators

• acquisition of emerging competitors;
• declining independent R&D;
• increasing dependence on incumbent infrastructure.

19. Remedies for Autonomous Markets

Traditional fines may be insufficient where the competitive problem arises from technological architecture.

Possible remedies include:

Structural remedies

• divestiture;
• separation of business units;
• restrictions on acquisitions.

Behavioural remedies

• non-discrimination;
• access obligations;
• interoperability;
• data portability;
• restrictions on self-preferencing.

Algorithmic remedies

• independent audits;
• algorithmic monitoring;
• reporting requirements;
• audit trails;
• restrictions on certain data inputs.

Institutional remedies

• specialist technical teams;
• continuous market monitoring;
• cooperation with data-protection and technology regulators.

20. Principle of Proportionality

Strategic foresight must not become technological precaution without evidence.

Algorithms can generate substantial competitive benefits.

They can:

• reduce transaction costs;
• improve matching;
• lower prices;
• optimise inventories;
• increase consumer choice;
• enable smaller firms to compete;
• improve resource allocation.

The CMA has expressly noted that pricing algorithms can produce competitive benefits while also creating competition risks.

Therefore, the correct policy objective is not:

“Control autonomous technology.”

It is:

“Preserve competitive conditions while allowing beneficial autonomous innovation.”

21. Future Competition-Policy Model

A useful model for autonomous markets is:

Observe → Anticipate → Test → Monitor → Intervene

Observe
Identify technological and market developments.

↓

Anticipate
Develop alternative future market scenarios.

↓

Test
Examine whether emerging conduct could undermine competition.

↓

Monitor
Track early-warning indicators.

↓

Intervene
Use the least restrictive effective competition remedy.

This is more adaptable than relying exclusively on retrospective enforcement.

22. Six Core Legal Lessons from the Case Law

23. Challenges for Competition Authorities

1. Lack of technical expertise

Traditional economists and lawyers may not be able to independently reconstruct complex AI systems.

2. Explainability problems

Machine-learning systems may generate decisions that are difficult to interpret.

3. Speed

An algorithm can change market conditions within seconds.

4. Evidence preservation

Algorithmic systems may continuously update themselves.

5. Attribution

Determining responsibility for machine-generated behaviour may be difficult.

6. Cross-border enforcement

AI infrastructure, data and users may be located across multiple jurisdictions.

7. Regulatory overlap

Competition authorities may increasingly interact with:

• AI regulators;
• data-protection authorities;
• consumer-protection regulators;
• financial regulators;
• telecommunications regulators.

24. A Strategic-Foresight Test for Autonomous Markets

A competition authority could ask seven questions:

1. Who controls the infrastructure?

2. Who controls the data?

3. Who controls access to consumers?

4. Can competitors interoperate?

5. Can consumers and suppliers switch?

6. Can the algorithm independently coordinate or exclude?

7. Could today's technological advantage become tomorrow's structural market power?

If several answers point toward durable dependency, the market may require enhanced competition monitoring.

Conclusion

Strategic foresight represents an important evolution in competition policy for autonomous markets. Traditional competition law remains applicable to algorithmic markets, as demonstrated by Topkins, Trod/GB eye, Eturas, Samir Agrawal, Google Shopping and RealPage. At the same time, these cases reveal that future competition problems may increasingly arise not from conventional agreements alone but from algorithmic infrastructure, data concentration, ranking systems, interoperability restrictions, autonomous pricing and ecosystem governance.

The central policy challenge is therefore to move from an exclusively reactive antitrust model toward a system capable of detecting emerging competitive risks without suppressing technological innovation.

In autonomous markets, competition polic