Competition Law and Competition Governance in Algorithmically Coordinated Societies

1. Introduction

An algorithmically coordinated society is one in which increasingly large portions of economic decision-making are performed, assisted, or coordinated by algorithms, artificial intelligence, automated platforms, data systems, and machine-learning models.

In conventional markets, competitors generally make independent decisions about:

• prices;
• output;
• supply;
• discounts;
• inventory;
• advertising;
• wages;
• market entry;
• customer allocation; and
• investment.

In algorithmically coordinated markets, these decisions may instead be influenced by systems that continuously observe competitors, process enormous quantities of data, predict market responses, and automatically modify commercial behaviour.

This creates a fundamental competition-law question:

When competitors coordinate through software rather than through direct human communication, when does technological coordination become unlawful coordination?

Existing competition law generally does not give algorithms a special exemption. The European Commission's horizontal cooperation guidelines expressly recognize that algorithms can facilitate "collusion by code" and state that firms cannot escape responsibility merely because an algorithm determines prices.

The harder question concerns autonomous algorithmic coordination—situations in which competing algorithms independently learn that maintaining higher prices or avoiding aggressive competition is profitable, without an express agreement between their human operators.

2. Meaning of Algorithmic Coordination

Algorithmic coordination can occur through several models.

A. Algorithm as a messenger

Human competitors make an unlawful agreement and then use algorithms to implement it.

Example: Competitors agree to maintain a minimum price and configure repricing software accordingly.

This is the easiest category for traditional competition law.

B. Algorithm as a monitoring mechanism

Competitors independently program algorithms to monitor competitors' prices and immediately respond to deviations.

The algorithm may make an existing cartel:

• faster;
• more stable;
• easier to enforce;
• more difficult to detect; and
• more costly to deviate from.

C. Common algorithm or hub-and-spoke coordination

Several competitors use the same platform, software provider or pricing intermediary.

The intermediary may collect information from competing firms and generate recommendations for each participant.

This creates potential hub-and-spoke concerns.

D. Algorithmic information exchange

Algorithms can collect and process:

• current prices;
• future prices;
• inventory;
• production capacity;
• customer behaviour;
• demand forecasts;
• discounts;
• margins; and
• commercially sensitive strategic information.

The continuous exchange of such information can reduce strategic uncertainty between competitors.

E. Autonomous algorithmic coordination

This is the most difficult category.

Two algorithms may independently learn that aggressive price competition reduces profits and consequently converge upon stable prices.

There may be:

• no telephone call;
• no email;
• no meeting;
• no written agreement; and
• no explicit instruction to collude.

The resulting economic effect may nevertheless resemble coordination.

This creates a major evidentiary and doctrinal challenge because traditional cartel law generally looks for some form of agreement, concerted practice, communication, or conscious coordination.

3. Why Algorithmic Coordination Is Different

Algorithms can fundamentally alter the competitive environment.

3.1 Continuous observation

Traditional competitors may observe competitors' prices periodically.

An algorithm can monitor them continuously.

3.2 Extremely rapid retaliation

If Firm A reduces its price, an algorithm may detect the reduction and respond within seconds.

This can discourage price reductions.

3.3 Reduced uncertainty

Algorithms can predict how competitors are likely to react.

This can make coordination more sustainable.

3.4 Self-learning

Machine-learning systems may identify strategies that their programmers did not explicitly specify.

3.5 Scale

A single algorithm can coordinate behaviour across:

• thousands of products;
• millions of transactions;
• multiple geographic markets; and
• numerous sellers.

3.6 Personalised pricing

Algorithms can potentially offer different prices to different consumers based on data concerning:

• purchasing behaviour;
• location;
• browsing history;
• willingness to pay;
• time;
• device;
• loyalty status; and
• demand conditions.

This creates additional concerns concerning algorithmic price discrimination and potentially exploitative conduct.

4. Legal Framework

A. Agreements and concerted practices

The traditional cartel provisions of competition law remain central.

India

Section 3 of the Competition Act, 2002 prohibits agreements causing or likely to cause an appreciable adverse effect on competition.

Section 3(3) particularly addresses horizontal arrangements such as:

• price fixing;
• limiting production;
• market allocation; and
• bid rigging.

The central issue in algorithmic cases is whether conduct occurring through software can constitute an agreement or concerted practice.

European Union

Article 101 TFEU addresses agreements, decisions of associations of undertakings and concerted practices that restrict competition.

United States

Section 1 of the Sherman Act addresses agreements restraining trade.

United Kingdom

Chapter I of the Competition Act 1998 prohibits agreements and concerted practices having the object or effect of preventing, restricting or distorting competition.

Thus, technology changes the mechanism of coordination, but not necessarily the legal prohibition.

5. Six Major Case Laws and Enforcement Developments

1. United States v. David Topkins — United States

Nature: Algorithm-assisted price-fixing.

This is one of the clearest examples of traditional cartel conduct being implemented through algorithms.

Online sellers of posters sold products through Amazon Marketplace. The participants agreed to fix prices and used computer code and pricing algorithms to implement the arrangement.

The U.S. Department of Justice prosecuted David Topkins, who pleaded guilty to price fixing. The case demonstrated that the use of sophisticated pricing algorithms does not transform an otherwise conventional cartel into lawful conduct.

Principle

An algorithm cannot legalise an underlying agreement to fix prices.

The important distinction is that the algorithm was essentially the instrument of the cartel, rather than an independent actor creating the cartel.

Significance

Topkins establishes the simplest liability model:

Human agreement → algorithmic implementation → cartel liability.

6. Trod Ltd / GB eye — United Kingdom

The UK's Competition and Markets Authority investigated two Amazon Marketplace sellers, Trod Ltd and GB eye Ltd.

The parties agreed not to undercut one another on specified products sold through Amazon UK.

They implemented the arrangement through automated repricing software.

The CMA imposed a £163,371 fine on Trod, while GB eye received immunity after reporting the cartel and cooperating with the investigation.

Principle

The CMA treated the software as a mechanism for implementing the underlying agreement.

The case therefore demonstrates that:

Automated execution does not remove human responsibility for a cartel.

Importance for algorithmic society

It illustrates the "messenger algorithm" model:

Competitors agree → software implements agreement → traditional cartel law applies.

7. Eturas UAB and Others v Lietuvos Respublikos konkurencijos taryba — CJEU

Case C-74/14

This case involved an online travel-booking system.

The platform administrator communicated a technical restriction concerning discounts available through the system. The issue was whether participating travel agencies could be treated as engaging in a concerted practice merely because the platform communicated the restriction.

The CJEU held that receipt of the system message alone was not automatically sufficient to establish participation in a concerted practice. However, awareness of the restriction together with circumstances indicating acceptance could be relevant to establishing participation, subject to the relevant evidentiary analysis.

Principle

The case is particularly important because it moves beyond a straightforward written cartel.

It raises the question:

Can participation in a common technological environment, combined with knowledge and conduct, establish concerted action?

Significance

Eturas is important for modern platform governance because competition authorities cannot simply assume:

"The computer sent the message, therefore the companies did nothing."

At the same time, technological participation cannot automatically substitute for proof of the legally required form of coordination.

8. Samir Agrawal v. Competition Commission of India — India

This case directly confronted algorithmic pricing in ride-hailing platforms.

The allegation concerned Ola and Uber and the proposition that their algorithms effectively determined fares accepted by drivers, potentially facilitating coordination among drivers.

The CCI rejected the initial information, and the matter ultimately reached the NCLAT.

The NCLAT examined whether the platform's algorithmic pricing could constitute a hub-and-spoke arrangement.

The decision emphasized the distinction between:

1. algorithmically determining prices using data; and
2. competitors actually agreeing or communicating through the platform to fix prices.

The case therefore illustrates an important evidentiary problem: similar or algorithmically determined prices do not automatically prove collusion.

Principle

A competition authority must distinguish:

parallel algorithmic behaviour

from

concerted algorithmic behaviour.

Importance for India

The case is particularly relevant to:

• digital platforms;
• gig economy;
• dynamic pricing;
• ride-hailing;
• AI pricing systems;
• hub-and-spoke theories; and
• Section 3 of the Competition Act.

9. United States v. RealPage — Algorithmic Rental Pricing

RealPage represents a more advanced form of algorithmic coordination.

In 2024, the U.S. Department of Justice and state plaintiffs sued RealPage, alleging that landlords supplied competitively sensitive information to RealPage's pricing software and that the software generated rental-price recommendations based on information from competing landlords. The complaint alleged violations of Sections 1 and 2 of the Sherman Act.

The alleged information included matters such as:

• rental prices;
• lease terms;
• projected vacancies; and
• other competitively sensitive information.

The case is important because the alleged mechanism was not simply:

"Competitors agreed on a price."

Instead, the theory involved:

competitors → sensitive information → common algorithm → pricing recommendations → coordinated market behaviour.

In November 2025, the DOJ announced a proposed settlement requiring RealPage to cease using competitors' non-public competitively sensitive information in runtime pricing and imposing restrictions on the use of certain data for model training.

Significance

RealPage demonstrates the emerging regulatory approach toward:

• common algorithms;
• competitively sensitive data;
• automated pricing recommendations;
• algorithmic monitoring;
• AI-assisted coordination; and
• structural dependence on a common pricing intermediary.

10. RealPage Landlord Proceedings — Greystar, LivCor, Willow Bridge and Others

The RealPage litigation expanded beyond the software provider itself.

The DOJ alleged that participating landlords shared competitively sensitive information and used common algorithmic pricing mechanisms.

Several proposed settlements subsequently imposed restrictions concerning:

• competitors' sensitive information;
• algorithmic pricing systems;
• third-party pricing software;
• monitoring;
• participation in industry meetings; and
• algorithmic features capable of aligning competitor prices.

For example, the proposed settlement with Greystar addressed the use of competitors' sensitive information and algorithmic pricing.

A 2026 proposed settlement involving Willow Bridge similarly contemplated restrictions on algorithms using competitors' competitively sensitive data and the possibility of court-appointed monitoring where uncertified third-party pricing algorithms are used.

Significance

This development shifts competition governance from simply asking:

"Did competitors communicate?"

toward asking:

"What data entered the algorithm, whose data entered it, what rules governed the algorithm, and how did firms respond to its recommendations?"

11. Pinnacle Property Management — 2026 Development

A further development occurred in September 2026, when the DOJ filed a proposed consent decree concerning Pinnacle Property Management in the broader algorithmic rental-pricing enforcement action.

The allegations concerned the use of competitors' competitively sensitive information through RealPage's algorithms and discussions concerning pricing strategies and software parameters.

The proposed decree would restrict the use of certain anticompetitive algorithms and competitors' sensitive information and provide for monitoring of certain third-party pricing systems.

Importance

This demonstrates that algorithmic competition governance is increasingly moving toward technology-specific compliance obligations, rather than relying exclusively on traditional fines after an infringement.

12. Comparative Analysis of the Six Core Cases

13. The Central Legal Distinction: Parallelism vs Coordination

This is perhaps the most important doctrinal problem.

Suppose:

• Firm A's algorithm charges ₹100.
• Firm B's algorithm charges ₹100.
• Both algorithms subsequently raise prices to ₹120.

That fact alone does not necessarily establish an unlawful agreement.

There could be legitimate explanations:

• identical cost structures;
• common market data;
• similar demand conditions;
• rational competitive responses;
• common publicly available information; or
• independent algorithmic optimisation.

Therefore:

Algorithmic price similarity is evidence requiring investigation, not automatically proof of cartelisation.

The stronger case arises where there is evidence of:

• communication;
• shared confidential information;
• common pricing instructions;
• coordinated software configurations;
• agreements to use particular algorithms;
• deliberate monitoring;
• common intermediary control;
• restrictions against deviation; or
• conscious acceptance of a coordinated strategy.

14. New Category: "Collusion by Code"

Modern competition governance increasingly recognizes that coordination can occur through software.

The European Commission's horizontal guidelines expressly discuss algorithms that facilitate coordination and explain that firms cannot avoid liability merely because their pricing decisions are generated by algorithms.

This produces an important rule:

Traditional cartel

Human agreement → human implementation

Algorithmic cartel

Human agreement → algorithm → automated implementation

More complex coordination

Competitor data → common intermediary → algorithm → coordinated recommendations

Most difficult scenario

Independent algorithms → machine learning → strategic convergence

The first three can often be addressed using existing competition law.

The fourth raises much more difficult questions about the meaning of "agreement" and "concerted practice."

15. Competition Governance in Algorithmically Coordinated Societies

Competition governance must therefore move beyond traditional enforcement.

A. Algorithmic transparency

Authorities may need access to:

• model architecture;
• pricing rules;
• training data;
• input variables;
• output records;
• version histories;
• audit logs;
• human overrides; and
• software configuration.

However, transparency must be balanced against:

• trade secrets;
• cybersecurity;
• privacy; and
• intellectual property.

16. Algorithmic Audit Requirements

High-risk pricing algorithms may require periodic competition audits.

An audit could ask:

1. What data does the system collect?
2. Does it collect competitors' non-public information?
3. Does the system recommend coordinated prices?
4. Does it penalise price reductions?
5. Does it monitor competitors?
6. Does it automatically respond to competitors?
7. Can managers override recommendations?
8. Is there an audit trail?
9. Who designed the algorithm?
10. Who controls its parameters?

This would transform competition compliance from an exclusively legal exercise into a technical governance process.

17. Data Governance

Data is the raw material of algorithmic coordination.

Competition authorities therefore increasingly need to examine:

• data pooling;
• data exclusivity;
• real-time competitor data;
• data portability;
• data access;
• information exchanges;
• common databases; and
• data intermediaries.

The RealPage proceedings illustrate why data governance and competition law are increasingly interconnected.

18. Governance of Common Algorithm Providers

A major future issue is the algorithmic intermediary.

Suppose 500 competing businesses use the same pricing provider.

The provider could potentially become an economic "hub."

Competition governance may therefore require rules concerning:

• what competitor information the intermediary may collect;
• how information is aggregated;
• whether real-time competitor information can be used;
• whether recommendations are personalised;
• whether one customer's data affects another customer's recommendation;
• whether the provider can monitor compliance; and
• whether the provider can punish deviations.

19. Human Accountability

One of the most important governance principles is:

Delegating a commercial decision to an algorithm should not automatically delegate legal responsibility.

A company cannot necessarily defend an unlawful strategy by saying:

"The AI made the decision."

The European Commission has expressly stated that firms remain responsible where algorithms operate under their direction or control.

This principle can be expressed as:

Human control → algorithmic execution → corporate responsibility.

20. Algorithmic Compliance Programmes

Modern competition compliance programmes should include a technological component.

Traditional compliance

• employee training;
• competition manuals;
• dawn-raid procedures;
• legal review.

Algorithmic compliance

Additionally:

• model testing;
• data-flow mapping;
• algorithmic competition audits;
• logging;
• red-team testing;
• independent review;
• model-change approval;
• sensitive-data controls;
• automated alerts; and
• periodic competition-risk assessments.

21. Competition Law and AI Developers

AI developers may increasingly become relevant actors in competition investigations.

Potential questions include:

• Did the developer know the algorithm would coordinate prices?
• Was the system trained using competitors' confidential data?
• Was the model designed to reward parallel pricing?
• Did the developer instruct users to follow recommendations?
• Did the provider monitor users' compliance?
• Did the provider discourage deviation?
• Did the provider facilitate information exchange?

This does not mean every AI provider is a cartel participant.

The legal analysis must still establish the relevant statutory elements.

22. Algorithmic Governance and Dominance

Algorithmic coordination is not limited to cartels.

A dominant platform could potentially use algorithms to:

• favour its own services;
• discriminate against rivals;
• exclude competitors;
• rank products selectively;
• restrict interoperability;
• exploit data advantages;
• degrade rival access;
• personalise offers strategically; or
• impose discriminatory conditions.

Thus, competition governance has two interconnected dimensions:

Coordination problem

Competitors coordinate through algorithms.

Exclusion problem

A dominant algorithmic platform uses algorithms to exclude competitors.

23. Essential Facilities and Algorithmic Infrastructure

Some algorithmic systems can become economically indispensable.

Examples may include:

• major app stores;
• payment infrastructure;
• cloud platforms;
• advertising exchanges;
• digital identity systems;
• interoperability interfaces;
• critical datasets;
• marketplace infrastructure.

Competition authorities may therefore face questions concerning:

• access;
• interoperability;
• discriminatory access;
• API restrictions;
• data portability;
• self-preferencing; and
• refusal to supply.

24. Algorithmic Markets and Consumer Welfare

Algorithmic coordination can affect consumers through:

Higher prices

Algorithms can stabilise prices above competitive levels.

Reduced discounts

Automated monitoring may make discounting easier to detect and punish.

Reduced innovation

If firms become comfortable with coordinated market conditions, incentives to innovate may decline.

Reduced entry

Entrants may find it difficult to compete against established algorithmic ecosystems.

Personalised discrimination

Different consumers may receive different offers based upon algorithmically inferred willingness to pay.

Reduced choice

Algorithmic ranking can influence which products consumers actually see.

25. Evidentiary Problems

Traditional cartel evidence often includes:

• emails;
• meetings;
• telephone calls;
• contracts;
• messages;
• witness testimony.

Algorithmic cases require additional evidence.

Digital evidence may include:

• source code;
• API logs;
• model weights;
• configuration files;
• model-training records;
• datasets;
• system messages;
• version histories;
• server logs;
• internal algorithm documentation; and
• communications between developers and business personnel.

This means competition authorities increasingly require technical forensic capabilities.

26. The "Black Box" Problem

Machine-learning systems can sometimes produce outputs that are difficult to explain.

This creates a significant legal problem.

Suppose an algorithm consistently produces coordinated prices.

The authority may ask:

Why?

The company may respond:

"We do not know; the model learned it."

That response creates a governance challenge.

A competition system cannot rely exclusively upon the proposition that unlawful conduct becomes lawful merely because its causal mechanism is difficult to explain.

27. Autonomous Tacit Coordination

This is arguably the hardest future competition-law problem.

Imagine:

• Algorithm A independently learns that price reductions are unprofitable.
• Algorithm B independently reaches the same conclusion.
• Neither company communicates with the other.
• Both algorithms continuously monitor market prices.
• Both learn that maintaining a high price maximises expected profits.
• Prices converge and remain elevated.

There may be economic coordination without conventional legal communication.

Existing competition law may struggle because:

Economic coordination and legally prohibited coordination are not necessarily identical concepts.

The distinction is crucial.

28. Should Autonomous Coordination Automatically Be Illegal?

There is currently no simple universal answer.

Competition law traditionally distinguishes between:

Independent conduct

Generally lawful.

Agreement or concerted practice

Potentially unlawful.

Explicit cartel

Clearly prohibited in most jurisdictions.

The autonomous algorithm problem sits between these categories.

Competition authorities therefore need to avoid treating every instance of parallel algorithmic behaviour as cartelisation while also ensuring that technology cannot become a shield for deliberately engineered coordination.

29. Proposed Governance Architecture

A future competition-governance framework could contain six layers.

Layer 1 — Ex ante design

Competition assessment before deploying high-risk algorithms.

Layer 2 — Data governance

Restrictions on the use of competitors' confidential information.

Layer 3 — Algorithmic testing

Testing for coordinated outcomes and anti-competitive incentives.

Layer 4 — Continuous monitoring

Monitoring algorithmic outputs after deployment.

Layer 5 — Human accountability

Identifying responsible executives, developers and business units.

Layer 6 — Enforcement

Possible remedies including:

• fines;
• behavioural commitments;
• algorithm modification;
• data restrictions;
• access remedies;
• independent monitoring;
• interoperability requirements; and
• in appropriate circumstances, structural remedies.

The recent RealPage-related settlements illustrate the movement toward algorithm-specific behavioural restrictions and monitoring.

30. Regulatory Challenges for India

India faces particularly important questions because of the growth of:

• digital marketplaces;
• UPI-based commerce;
• ride-hailing;
• food-delivery platforms;
• e-commerce;
• fintech;
• digital advertising;
• AI;
• cloud computing; and
• algorithmic pricing.

The Competition Commission of India may increasingly need to examine whether algorithmic systems:

1. facilitate horizontal coordination;
2. create hub-and-spoke structures;
3. facilitate information exchange;
4. reinforce platform dominance;
5. discriminate against competitors;
6. foreclose market access;
7. facilitate personalised pricing; or
8. create anti-competitive network effects.

The Samir Agrawal litigation provides an important Indian foundation for analysing the relationship between algorithmic pricing and the concept of a hub-and-spoke arrangement.

31. Algorithmic Governance and the Future of Cartel Law

The traditional cartel model can be represented as:

Competitor A ↔ Competitor B

Future algorithmic markets may instead look like:

Competitor A → Algorithm ← Competitor B

or:

Competitor A → Common Platform ← Competitor B

or:

Competitor A → Data Pool → AI Model ← Data Pool ← Competitor B

The law therefore has to identify where the coordination actually occurs.

The relevant actor might be:

• the competitor;
• the platform;
• the software provider;
• the data intermediary;
• the algorithm designer;
• the corporate user; or
• several of these actors simultaneously.

32. Important Doctrinal Principles

The emerging case law and enforcement experience support several important propositions.

Principle 1 — Technology neutrality

Competition law generally applies irrespective of whether coordination occurs offline or digitally.

Principle 2 — Algorithm is not a legal shield

A company cannot automatically escape liability merely because software implemented the conduct. Topkins and Trod are particularly clear examples.

Principle 3 — Algorithmic similarity is not automatically collusion

Samir Agrawal demonstrates the importance of distinguishing algorithmically generated parallel conduct from legally cognisable coordination.

Principle 4 — Sensitive data is central

RealPage demonstrates the increasing importance of examining what competitively sensitive information enters an algorithm.

Principle 5 — Common intermediaries create special risks

Where competitors depend on the same pricing or decision-making intermediary, hub-and-spoke theories become increasingly relevant.

Principle 6 — Technical governance is becoming competition governance

Modern enforcement may require monitoring:

data + code + model + output + human decision-making.

33. Future Case-Law Questions

Courts and competition authorities are likely to confront questions such as:

1. Can an AI system itself become the mechanism of a cartel?
2. When does use of common software constitute a hub-and-spoke arrangement?
3. Is deliberate adoption of an algorithm known to facilitate coordination sufficient?
4. What level of human knowledge is required?
5. Can an algorithm's internal learning process constitute evidence of coordination?
6. Can companies be liable for foreseeable algorithmic coordination?
7. What constitutes competitively sensitive training data?
8. Can algorithm providers be treated as intermediaries?
9. When should competition authorities require algorithmic audits?
10. Can algorithmic coordination justify structural remedies?

These questions will increasingly determine the future development of competition law.

34. Conclusion

Competition law in an algorithmically coordinated society cannot remain confined to traditional concepts of meetings, contracts and explicit communications.

The fundamental concern remains the preservation of independent competitive decision-making.

The leading examples show an evolutionary progression:

Topkins — algorithm implements an explicit cartel.

Trod/GB eye — automated repricing implements an agreement.

Eturas — common technological infrastructure raises questions of knowledge and concerted practice.

Samir Agrawal — algorithmic pricing does not automatically establish a hub-and-spoke cartel.

RealPage — common algorithms and competitors' sensitive data create a much more sophisticated coordination problem.

Pinnacle and related 2025–2026 proceedings — competition remedies are increasingly reaching into the design, data inputs, monitoring and governance of algorithmic pricing systems.

The emerging principle can therefore be expressed as:

Competition law must regulate not merely what firms agree to do, but also the technological systems through which competitive decisions are generated, communicated, monitored and implemented.

The ultimate challenge is to preserve the distinction between legitimate algorithmic competition—where technology makes markets faster and more efficient—and algorithmically facilitated coordination, where technology reduces competitive independence. The former can enhance competition; the latter can reproduce the economic effects of a cartel even when the mechanism looks radically different from a traditional cartel.