Competition Law and Competition Concerns in Machine-Designed Protocols

1. Introduction

Machine-designed protocols are rules, decision systems, algorithms, or technical protocols that are designed, optimized, or continuously modified by software or artificial intelligence rather than being entirely specified by human decision-makers.

Examples include:

• AI-designed pricing and bidding protocols;
• automated allocation and matching rules;
• platform ranking and recommendation protocols;
• machine-generated access and interoperability rules;
• autonomous trading protocols;
• algorithmic procurement mechanisms;
• blockchain or smart-contract governance protocols;
• AI-designed standards and technical interfaces;
• automated rules governing participants in digital ecosystems.

Competition law does not generally exempt conduct merely because a machine designed, recommended, implemented, or modified the rule. The central legal questions remain whether the protocol produces an agreement, coordinated conduct, exclusionary conduct, discriminatory access, abuse of dominance, or another restriction of competition.

The CMA has specifically recognised that algorithms can facilitate explicit coordination, create hub-and-spoke information exchange, and potentially generate autonomous tacit coordination.

2. Meaning of Machine-Designed Protocols

A conventional protocol is normally designed as:

Human objective → human rules → software implementation → market outcome.

A machine-designed protocol may instead operate as:

Objective/data → machine optimisation → machine-generated rules → automated implementation → market feedback → further machine modification.

The important competition-law issue is therefore the allocation of decision-making power between humans and machines.

A protocol can be:

A. Human-designed and machine-operated

Humans establish the rules and software merely executes them.

B. Machine-assisted

Humans establish objectives but AI determines parameters or strategies.

C. Machine-designed

The AI itself develops the operative rules or protocol.

D. Continuously machine-adaptive

The protocol changes in response to competitors, consumers, market conditions, or data.

The last two categories create particularly difficult competition questions because the final market behaviour may not have been expressly contemplated by any individual human actor.

3. Relevant Competition-Law Framework

A. Anti-competitive agreements

Machine-designed protocols may facilitate:

• price fixing;
• output restrictions;
• market allocation;
• customer allocation;
• bid coordination;
• information exchange;
• resale-price coordination;
• restrictions on interoperability.

In India, Section 3 of the Competition Act, 2002 is particularly relevant.

The crucial issue is not whether an algorithm physically communicated with another algorithm. The issue is whether the underlying conduct can legally be attributed to undertakings and whether there is an agreement, arrangement, understanding, or concerted practice having an anti-competitive effect.

B. Abuse of dominant position

Where a dominant platform uses a machine-designed protocol, Section 4 of the Competition Act may become relevant.

Possible forms include:

1. self-preferencing;
2. discriminatory ranking;
3. discriminatory access;
4. exclusionary interoperability rules;
5. tying;
6. refusal to deal;
7. exploitative pricing;
8. leveraging of data advantages;
9. exclusion of competing protocols.

The European Commission's Google Shopping litigation illustrates the importance of examining how a dominant platform's own technical systems can favour its affiliated service over competing services.

4. Major Competition Concerns

4.1 Algorithmic Collusion

The most obvious concern is that machine-designed protocols could make coordination easier.

Suppose competing firms independently deploy AI systems instructed to:

maximise long-term profit while responding rapidly to competitors.

The systems may learn that aggressive price competition is less profitable than maintaining elevated prices.

This raises the difficult distinction between:

• express collusion;
• algorithm-assisted collusion;
• algorithm-mediated coordination;
• autonomous tacit coordination.

The CMA has expressly identified autonomous tacit collusion as a possible competition concern.

However, autonomous coordination does not automatically establish a conventional cartel. Competition authorities must still establish the applicable legal elements.

5. Hub-and-Spoke Machine Protocols

A particularly important structure occurs when several competitors use the same algorithmic intermediary.

For example:

Hotel A →
Hotel B → Common AI pricing provider → Market prices
Hotel C →

The common system may receive:

• prices;
• inventories;
• occupancy;
• future pricing plans;
• capacity;
• demand forecasts.

The algorithm may then recommend prices to competing firms.

This can create a hub-and-spoke theory of coordination.

The concern is not merely the software itself. It is whether the common technological intermediary facilitates information exchange or coordinated conduct between competitors.

The CMA has specifically identified common pricing software and third-party algorithmic systems as potential mechanisms for information exchange and coordination.

6. Machine-Designed Protocols and Information Exchange

Machine-designed protocols can radically increase the speed and precision of information exchange.

Traditional information exchange:

Competitor A sends information → Competitor B receives it.

Machine-mediated exchange:

Data feed → algorithm → prediction → automated response.

Information can therefore become commercially actionable almost instantaneously.

Potentially sensitive information includes:

• future prices;
• inventory;
• production capacity;
• customer segmentation;
• bidding intentions;
• discounts;
• costs;
• capacity utilisation.

This may reduce the uncertainty that normally makes competitive markets function.

7. Self-Learning Protocols

Self-learning systems create a more difficult problem.

An AI may be instructed:

maximise profits subject to legal and commercial constraints.

The AI may discover a strategy involving:

• parallel pricing;
• market segmentation;
• strategic capacity reduction;
• exclusion of particular rivals;
• discriminatory access;
• retaliation against aggressive competitors.

The company may argue that:

"The machine independently discovered the strategy."

Competition law may nevertheless examine whether the undertaking designed, deployed, supervised, or benefited from the system.

The absence of a human instruction saying "collude" does not necessarily eliminate competition-law risk.

8. Machine-Designed Protocols as Digital Infrastructure

Protocols can also become infrastructure.

Consider an AI-designed protocol governing:

• payment interoperability;
• digital identity;
• cloud access;
• app communication;
• EV charging;
• financial APIs;
• blockchain transactions.

If a dominant undertaking controls the protocol, it may have the ability to determine:

who can participate, under what conditions, and on what terms.

This creates possible essential-facility, refusal-to-deal, interoperability, and discrimination concerns.

9. Protocol Lock-In

Machine-designed protocols may produce technological lock-in.

A platform could develop a protocol that:

1. attracts users;
2. attracts developers;
3. generates more data;
4. improves the protocol;
5. increases switching costs;
6. attracts still more users.

This creates a feedback loop:

Users → Data → Better protocol → More users → More data

The resulting network effects can strengthen market power.

The UK digital-regulator research recognises that platforms can use accumulated personal data to improve algorithmic systems, increasing engagement and reinforcing network effects.

10. Discriminatory Machine-Designed Protocols

An AI protocol may treat market participants differently.

Examples:

• higher API fees for rival platforms;
• slower access for competing applications;
• preferential ranking for affiliated businesses;
• different interoperability standards;
• discriminatory allocation of computing resources;
• different transaction-validation rules.

The difficulty is distinguishing:

Legitimate optimisation

from

Exclusionary discrimination.

Competition authorities may therefore examine:

• purpose;
• effect;
• market power;
• technical justification;
• economic rationale;
• availability of alternatives;
• foreclosure of competitors.

11. Six Important Case Laws

Because machine-designed protocols are a relatively new phenomenon, there are few reported judgments dealing with that exact terminology. The following cases establish principles that are highly relevant to algorithmic and machine-designed protocols.

Case 1: Samir Agrawal v. Competition Commission of India

Supreme Court of India, 2021, 3 SCC 136

This is one of the most important Indian authorities concerning algorithmic pricing.

The allegation concerned the pricing mechanism used by ride-hailing platforms and the argument that algorithmic pricing effectively restricted the ability of drivers to compete independently.

The Supreme Court ultimately dismissed the information on the facts and statutory framework applicable to the case.

Importance

The case demonstrates that:

• algorithmic pricing does not automatically constitute price fixing;
• the operation of an algorithm must be analysed within the statutory concept of an agreement;
• economic coordination and legally actionable collusion are not necessarily identical.

The underlying dispute specifically concerned allegations that algorithmic pricing affected competition between drivers.

Relevance to machine-designed protocols

If an AI independently establishes prices or allocation rules, the authority must still examine the legal relationship between:

platform → algorithm → participating businesses → market outcome.

Case 2: Trod Ltd. and GB Eye Ltd. — CMA

CMA, 2016

Two competing online sellers agreed not to undercut each other's prices on Amazon Marketplace.

They used automated repricing software to implement the arrangement.

The CMA found a competition-law infringement and fined Trod £163,371; GB Eye obtained immunity after reporting the cartel and cooperating with the investigation.

Importance

This is a leading practical example of:

human cartel → machine implementation → automated market conduct.

The software itself did not create the cartel.

Humans created the anti-competitive agreement, while software implemented it.

Principle

Automation is not a defence to cartel conduct.

Case 3: United States v. Topkins

United States District Court for the Northern District of California, 2015

This involved online sellers who used pricing algorithms in connection with an agreement concerning prices for posters.

The case is significant because it demonstrated that conventional antitrust principles could apply when algorithms are used to implement coordinated pricing.

Importance

It establishes the basic proposition that:

A cartel does not become lawful merely because the agreed pricing mechanism is implemented through software.

Relevance

Machine-designed protocols may therefore require scrutiny of both:

1. the technological architecture; and
2. the commercial instructions underlying that architecture.

Case 4: Eturas UAB v. Lietuvos Respublikos konkurencijos taryba

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

This case concerned an online travel-booking system in which a technical message was distributed through the system concerning discounts.

The CJEU examined when undertakings participating in an electronic platform could be considered involved in concerted conduct.

Importance

The case is particularly relevant because the communication mechanism was electronic and platform-mediated.

The Court emphasised the importance of knowledge and participation in determining whether undertakings could be held responsible for concerted practices.

Relevance to machine-designed protocols

It provides an important conceptual foundation for asking:

• Did participants know about the protocol?
• Did they receive the relevant information?
• Did they knowingly continue participating?
• Can their subsequent conduct demonstrate acceptance?

Thus, technical architecture does not automatically eliminate the legal significance of communication and participation.

Case 5: Google Shopping

European Commission / General Court, Google and Alphabet v Commission, Case T-612/17

The case concerned Google's treatment of competing comparison-shopping services and the manner in which Google's search system displayed results from its own specialised comparison-shopping service.

The General Court upheld the essential finding of abuse of dominance concerning the preferential treatment of Google's own service.

Importance for machine-designed protocols

This case is highly relevant to AI-generated ranking and allocation protocols.

A machine-generated ranking system can determine:

• visibility;
• access to consumers;
• traffic;
• commercial opportunities.

Where a dominant undertaking designs or controls the ranking architecture, competition law may scrutinise whether the system systematically advantages its own service.

Case 6: Cornish-Adebiyi v. Caesars Entertainment

U.S. litigation concerning algorithmic hotel pricing

In 2024, the FTC and U.S. Department of Justice filed a statement of interest concerning allegations involving hotel-room pricing algorithms.

The agencies explained that businesses cannot use an algorithm to engage in conduct that would be unlawful if performed by a person. They also highlighted the possibility that a common algorithm provider could facilitate coordination among competitors.

Importance

The case is particularly significant for the future of machine-designed protocols because it addresses the question:

Can competitors accomplish indirectly through an algorithm what they could not lawfully accomplish directly?

The enforcement position indicates that technological mediation does not itself immunise otherwise unlawful coordination.

12. Comparative Case-Law Table

13. Machine-Designed Protocols and Section 3 of the Indian Competition Act

Section 3 analysis can be organised around several questions.

Step 1 — Identify the undertakings

Who controls or participates in the protocol?

Step 2 — Identify the protocol

What exactly does the machine-designed system do?

Step 3 — Identify the data inputs

Does it use:

• competitors' prices?
• confidential information?
• customer information?
• capacity information?
• future commercial plans?

Step 4 — Identify the decision rule

Does it:

• independently optimise;
• match competitors;
• punish deviations;
• allocate markets;
• restrict output;
• discriminate against rivals?

Step 5 — Determine human involvement

Were the parameters:

• expressly programmed?
• approved?
• knowingly accepted?
• continuously supervised?

Step 6 — Examine competitive effects

Does the protocol:

• raise prices?
• reduce output?
• reduce innovation?
• exclude competitors?
• restrict consumer choice?

14. Section 4 and Dominant Machine Protocols

For a dominant undertaking, a machine-designed protocol could potentially constitute abuse where it is used to:

A. Deny access

A competing service is technically prevented from connecting.

B. Degrade interoperability

The competitor technically remains connected but receives inferior functionality.

C. Self-preference

The dominant platform's algorithm systematically favours its affiliated products.

D. Discriminate

Equivalent commercial partners receive materially different conditions.

E. Foreclose rivals

The protocol increases rivals' costs or prevents effective market entry.

F. Exploit users

The protocol may impose excessive or discriminatory conditions on users.

15. Machine-Designed Standards and Interoperability

A particularly important emerging issue concerns technical standards designed by AI.

Suppose several businesses depend upon an AI-designed communication protocol.

If one undertaking controls the protocol and changes it in a manner that makes rival products incompatible, the modification could have significant competitive effects.

Relevant questions include:

• Who owns the protocol?
• Who controls modifications?
• Are changes transparent?
• Are competitors consulted?
• Is access objectively available?
• Are technical changes necessary?
• Is there a less restrictive alternative?

This makes protocol governance an important component of competition compliance.

16. Competition Risks from Autonomous Learning

A machine-designed protocol can potentially learn strategies that humans did not explicitly program.

Potential risks include:

1. Coordinated pricing

Algorithms converge on elevated prices.

2. Market segmentation

Algorithms divide customers or geographic markets.

3. Retaliation

An algorithm automatically responds to aggressive competition.

4. Capacity coordination

Systems reduce output in response to competitors.

5. Entry deterrence

Protocols automatically react to new entrants.

6. Exclusion

A dominant platform's system makes competing services less visible or accessible.

7. Data exploitation

Large datasets improve the protocol while making market entry progressively more difficult.

17. The "Black Box" Problem

A machine-designed protocol may be difficult even for its operator to explain.

This creates a major evidentiary issue.

A competition authority may ask:

Why did the algorithm choose this outcome?

The undertaking may respond:

The model generated the result autonomously.

This does not necessarily resolve liability.

Authorities may instead investigate:

• training data;
• objectives;
• constraints;
• model architecture;
• optimisation functions;
• reward mechanisms;
• input variables;
• output history;
• audit logs;
• model updates;
• human supervision.

Therefore, algorithmic explainability can become a competition-compliance issue.

18. Evidence and Accountability

Businesses using machine-designed protocols should maintain:

• version histories;
• model documentation;
• audit logs;
• input-data records;
• decision records;
• parameter changes;
• approval records;
• compliance testing;
• monitoring mechanisms;
• records of human intervention.

This is especially important where an algorithm changes its behaviour over time.

The CMA has emphasised that businesses need to understand how pricing systems work and the risks associated with algorithmic coordination.

19. Legitimate Pro-Competitive Uses

Machine-designed protocols are not inherently anti-competitive.

They can generate significant benefits, including:

• lower transaction costs;
• better resource allocation;
• faster matching;
• reduced waste;
• improved logistics;
• better capacity utilisation;
• improved interoperability;
• lower prices;
• improved product quality;
• innovation.

The CMA has noted that algorithmic pricing can produce benefits such as faster adjustment to demand and supply, lower business costs, and stronger competitive opportunities for smaller businesses.

Therefore, competition law should distinguish technological innovation from anti-competitive exploitation of technology.

20. Compliance Framework for Businesses

A company deploying machine-designed protocols should establish an AI Competition Compliance Programme.

Before deployment

• conduct competition-risk assessment;
• identify competitors' data;
• identify possible coordination mechanisms;
• determine whether common algorithmic providers are used.

During deployment

• monitor outputs;
• test for discriminatory outcomes;
• restrict access to competitor-sensitive information;
• establish human oversight;
• monitor unexpected coordination.

After deployment

• maintain audit logs;
• investigate anomalous market behaviour;
• periodically test the model;
• document model modifications;
• preserve evidence explaining major decisions.

21. Emerging Legal Test

A useful analytical framework for machine-designed protocols is:

Objective → Data → Architecture → Autonomy → Conduct → Effect

Objective:
What was the machine instructed to achieve?

Data:
What information does it receive?

Architecture:
Who designed and controls the protocol?

Autonomy:
How independently can it change its strategy?

Conduct:
What does it actually do?

Effect:
What happens to competition?

This approach avoids treating "AI" or "machine design" as a legal category by itself.

22. Key Legal Issues for Future Litigation

Future cases are likely to focus on questions such as:

1. Can autonomous AI coordination constitute an agreement?
2. When does algorithmic information exchange become unlawful?
3. Can a common AI provider become a hub facilitating competitor coordination?
4. Can a dominant firm be responsible for exclusionary outcomes generated by an autonomous system?
5. How should courts distinguish tacit coordination from unlawful concerted practice?
6. Who bears responsibility when the algorithm modifies its own protocol?
7. Can algorithmic neutrality be objectively demonstrated?
8. What level of explainability should competition authorities require?
9. Can machine-generated standards constitute exclusionary conduct?
10. How should competition law address AI protocols that become industry infrastructure?

23. Conclusion

Machine-designed protocols represent a significant evolution of competition problems from human-designed rules to algorithmically generated market structures.

The central legal principle is that the technological origin of a decision does not determine its competition-law legality.

The key questions are:

Who designed the system?
Who controls it?
What information does it use?
What objectives does it pursue?
How autonomous is it?
What market conduct does it generate?
What effect does that conduct have on competition?

The existing cases—particularly Samir Agrawal, Trod/GB Eye, Topkins, Eturas, Google Shopping, and Cornish-Adebiyi—show different stages of the transition from traditional human coordination to increasingly automated competitive decision-making.

The most important distinction is therefore between machine-assisted competition and machine-enabled anti-competitive coordination or exclusion. Competition law can address the latter without treating technological innovation itself as unlawful.