Competition Law and Future Competition Regulation of Automated Agreements

Introduction

Automated agreements are contracts whose formation, performance, adjustment, or enforcement is substantially carried out by software, algorithms, smart contracts, artificial-intelligence systems, or machine-to-machine communications. Examples include algorithmically generated prices, automated supply contracts, smart-contract transactions, dynamic procurement agreements, platform terms accepted automatically, and agreements in which an algorithm continuously adjusts commercial conditions.

Competition law traditionally assumes that human firms make decisions and enter agreements. Automation complicates this assumption. Two competing businesses may deploy pricing systems that independently reach the same price; alternatively, firms may intentionally design algorithms to implement an unlawful agreement. A future competition regime therefore has to distinguish between:

1. genuine independent algorithmic conduct;
2. human decisions implemented through automation;
3. facilitated coordination through a common algorithm or intermediary;
4. algorithmic conduct that creates or strengthens tacit coordination; and
5. automated agreements that themselves constitute contracts or concerted practices.

The central regulatory question is not whether an algorithm "intended" to restrict competition, but whether firms used automation in a manner that produces, implements, facilitates, or sustains conduct prohibited by competition law.

I. Meaning and Characteristics of Automated Agreements

An automated agreement may contain software that determines:

• price;
• quantity;
• discounts;
• delivery conditions;
• allocation of customers;
• inventory;
• bidding strategies;
• access conditions;
• contractual penalties;
• renewal;
• termination;
• matching of buyers and sellers; or
• responses to competitors.

There are several important forms.

1. Algorithmically generated agreements

A software system may automatically generate contractual offers based on predetermined parameters.

2. Dynamic agreements

The contractual price or condition changes automatically according to market data.

For example:

Price = reference market price + automated adjustment + demand coefficient.

3. Smart contracts

A blockchain-based program may automatically execute contractual obligations when predetermined conditions are satisfied.

4. Platform-mediated agreements

A platform may require competing sellers to use common software for pricing, ranking, allocation or other commercial decisions.

5. Machine-to-machine agreements

Two autonomous systems may negotiate or exchange commercial terms without direct human intervention at the moment of agreement.

II. Competition-Law Problem

Traditional competition law generally prohibits agreements, decisions or concerted practices that restrict competition.

The automation of the process does not necessarily change the legal character of the underlying conduct.

If competitors agree:

"We will charge ₹100."

and subsequently use software to implement that agreement, the use of software does not make the arrangement lawful.

The harder situation is:

Competitor A and Competitor B independently deploy algorithms that continuously observe one another and rapidly adjust prices.

There may be no explicit communication between the firms. This raises the distinction between collusion and parallel conduct.

III. Automated Agreements and Cartel Regulation

A. Traditional cartel implemented by software

This is the simplest case.

Suppose five manufacturers agree to maintain a minimum price and program their pricing systems accordingly.

The relevant competition violation is the underlying agreement. The algorithm is merely the implementation mechanism.

Legal principle

Automation cannot be used as a shield against liability for an agreement that would otherwise violate competition law.

IV. Algorithmic Tacit Coordination

A more difficult problem arises when algorithms learn from market behaviour.

An algorithm may:

1. observe competitors;
2. predict their future prices;
3. modify its own price;
4. observe the resulting market;
5. repeat the process.

Over time, competing systems may converge on supra-competitive prices.

The difficulty is establishing whether this amounts to:

• independent competitive behaviour;
• conscious parallelism;
• facilitated coordination;
• a concerted practice; or
• an agreement.

Competition regulators therefore need increasingly sophisticated evidence rules.

V. The Common-Algorithm Problem

Suppose competing retailers independently use the same pricing provider.

The provider's algorithm recommends:

"Increase price whenever another participating retailer increases its price."

This creates a potentially important competition concern.

Although the retailers may not directly communicate, the common intermediary can become a mechanism through which commercially sensitive information is transmitted or coordinated.

The legal analysis should examine:

• what information the algorithm receives;
• what information it transmits;
• whether competitors know that the same system is being used;
• whether the system intentionally facilitates coordination;
• whether the firms could reasonably understand its operation;
• whether the provider encourages coordinated behaviour; and
• whether the resulting conduct is attributable to the firms.

VI. Smart Contracts and Competition Law

Smart contracts create a special problem because execution may be automatic.

Suppose competitors use blockchain-based contracts that automatically:

• maintain agreed prices;
• impose penalties for discounting;
• restrict supplies;
• allocate customers; or
• prevent deviations from a cartel arrangement.

Once deployed, the code may execute without additional human action.

Nevertheless, automatic execution does not necessarily eliminate competition-law responsibility.

The relevant questions become:

1. Who created the code?
2. Who commissioned it?
3. Who controlled its parameters?
4. Who knew its commercial purpose?
5. Could the parties modify or terminate it?
6. Did the parties coordinate before deployment?
7. Did the code facilitate implementation of an anticompetitive arrangement?

VII. Important Case Laws

1. United States v. Topkins (2015)

This is one of the most important early cases concerning algorithmic price coordination.

Online sellers used pricing algorithms in connection with an agreement to coordinate prices for posters and framed art sold through an online marketplace.

Significance

The case demonstrated that:

• online markets are subject to ordinary cartel principles;
• software can be used to implement price coordination;
• algorithmic implementation does not remove cartel liability; and
• digital evidence can establish the underlying agreement.

Principle

The use of pricing software does not transform an unlawful price-fixing agreement into independent competitive conduct.

2. United States v. Airline Tariff Publishing Co.

This case involved airline fare information and the use of sophisticated computerised systems to communicate and coordinate pricing information.

Significance

The matter illustrated the importance of electronic information systems in detecting and facilitating coordination.

It is particularly relevant to modern automated agreements because pricing systems can make it easier for competitors to:

• monitor prices;
• signal intentions;
• detect deviations; and
• rapidly respond to competitors.

Principle

Electronic communication mechanisms can form part of the evidentiary structure of a competition-law violation.

3. Eturas UAB v Lietuvos Respublikos Konkurencijos Taryba (CJEU, C-74/14)

This is one of the most important European cases involving an electronic platform.

A common electronic booking system imposed a restriction on discounts offered by travel agencies.

The issue was whether participating undertakings could be held responsible for anti-competitive conduct communicated through the platform.

Significance

The Court examined:

• electronic communication;
• common software;
• knowledge of restrictions;
• participation in coordinated conduct; and
• evidentiary presumptions.

Principle

A technological platform can constitute the mechanism through which competitors receive and implement a competition-restricting measure. Liability depends on the relevant undertaking's knowledge and participation rather than simply on the existence of software.

This case is particularly important for platform-mediated automated agreements.

4. AC-Treuhand AG v European Commission

The CJEU developed important principles concerning liability for facilitating cartel conduct.

An undertaking can face competition-law consequences even when it is not itself a conventional seller of the cartelised product, if its conduct contributes to the operation of the cartel.

Significance for automated agreements

Modern algorithm providers, data intermediaries and software operators may occupy a similar facilitating role.

Potentially relevant actors therefore include:

• algorithm developers;
• pricing-software providers;
• data intermediaries;
• digital platforms; and
• industry information providers.

Principle

Competition law can extend beyond the immediate contracting parties where an undertaking knowingly contributes to anticompetitive coordination.

5. Wood Pulp / A. Ahlström Osakeyhtiö and Others v Commission

The Wood Pulp litigation remains significant for understanding coordinated conduct and the evidentiary limits surrounding parallel market behaviour.

Significance

Competition authorities cannot automatically equate similar commercial behaviour with an unlawful agreement.

This becomes particularly important with AI and machine-learning systems.

Two algorithms may independently respond to the same market variables and produce similar outcomes.

Principle

Parallel behaviour alone does not automatically establish an unlawful agreement or concerted practice.

This principle remains critical for future algorithmic enforcement.

6. T-Mobile Netherlands BV v Raad van bestuur van de Nederlandse Mededingingsautoriteit (CJEU, C-8/08)

The CJEU examined the concept of concerted practice and the significance of information exchange between competitors.

Significance

The case is relevant because automated systems can exchange or process information much more rapidly than human actors.

A modern regulator may therefore need to examine whether algorithmic information flows:

• reduce strategic uncertainty;
• facilitate coordination;
• communicate future commercial intentions; or
• enable competitors to monitor deviations.

Principle

Information exchange can constitute a competition concern where it reduces strategic uncertainty sufficiently to facilitate coordination.

7. Dole Food and Dole Fresh Fruit Europe v European Commission (CJEU, C-286/13 P)

The case concerned information exchange and coordination in the banana market.

Significance

It reinforces the importance of examining communications and information exchanges between competitors rather than merely looking for an express written cartel agreement.

For automated markets, equivalent evidence may be contained in:

• API communications;
• system logs;
• configuration files;
• software instructions;
• algorithmic messages;
• database records; and
• machine-generated communications.

Principle

Competition-law analysis can focus on the substance and competitive effect of information exchanges rather than their technological form.

VIII. Evidence in Automated-Agreement Cases

Future competition investigations will increasingly involve technical evidence.

Important evidence may include:

1. Source code

Regulators may examine whether code contains instructions facilitating:

• price coordination;
• market allocation;
• customer exclusion;
• bid rotation; or
• information sharing.

2. Model documentation

Authorities may examine:

• training objectives;
• optimisation criteria;
• reward functions;
• constraints;
• prohibited actions; and
• deployment instructions.

3. System logs

Logs can establish:

• when algorithms interacted;
• what information they received;
• what decisions they made; and
• whether changes followed competitor behaviour.

4. API records

APIs may reveal automated information exchange between competing systems.

5. Version histories

Software repositories can establish when an anticompetitive feature was introduced.

6. Human instructions

Emails, internal documents and management instructions remain crucial.

IX. The Role of Human Responsibility

One of the most important future principles should be:

Automation should not automatically transfer responsibility from humans to machines.

A company should not be able to argue:

"The algorithm decided."

The relevant inquiry should instead include:

• Who designed the system?
• Who selected the objective?
• Who selected the constraints?
• Who approved deployment?
• Who knew how the system operated?
• Who monitored its effects?
• Who could modify or deactivate it?

X. Algorithmic Pricing and Automated Agreements

Automated pricing presents several distinct situations.

The critical distinction is between automation of competition and automation of coordination.

XI. Automated Agreements and Dominant Digital Platforms

A dominant platform may impose automated contractual rules on businesses using its ecosystem.

Examples include:

• automatic delisting;
• automated ranking;
• algorithmic commission adjustment;
• automatic access restrictions;
• automated parity requirements;
• algorithmic refusal of interoperability;
• automatic allocation of advertising inventory.

The issue may not be cartelisation. Instead, the relevant theories may include:

• abuse of dominance;
• discriminatory access;
• tying;
• exclusionary conduct;
• self-preferencing;
• margin squeeze; or
• exploitative contractual conditions.

XII. Automated Agreements and Vertical Restraints

Automation can also create vertical competition concerns.

For example, a manufacturer might require distributors to use software that automatically:

• prevents discounts;
• monitors resale prices;
• terminates non-compliant distributors;
• restricts territories;
• limits online sales.

This could transform traditional contractual restrictions into continuous algorithmic enforcement.

The competition analysis should therefore consider both:

1. the contractual provision; and
2. the technological mechanism enforcing it.

XIII. Future Regulation of Automated Agreements

Future competition regulation is likely to develop around several principles.

1. Algorithmic transparency

Authorities may require firms to preserve sufficient information concerning:

• algorithm objectives;
• material parameters;
• data inputs;
• decision rules;
• major model changes; and
• governance procedures.

This does not necessarily mean that all source code must become public.

2. Algorithmic audit

Competition authorities may increasingly conduct technical audits.

An audit could examine:

Inputs → Model → Decision → Market response → Feedback → Subsequent decision

This allows investigators to determine whether coordination is embedded in the system.

3. Competition-by-design

Businesses could be required to incorporate competition safeguards during system development.

For example:

• prohibit competitor-sensitive data;
• prevent automatic coordination;
• implement compliance constraints;
• maintain audit trails;
• establish human oversight;
• test for collusive outcomes.

4. Duty to monitor

A firm deploying a sophisticated pricing algorithm may increasingly be expected to monitor reasonably foreseeable competition-law risks.

This creates a distinction between:

"We did not instruct the algorithm to collude"

and

"We knew the system was producing coordinated outcomes and continued deploying it."

The latter situation can raise substantially different legal questions.

XIV. Regulation of Algorithm Providers

A major future issue concerns the responsibility of third-party algorithm providers.

Suppose the same company supplies pricing software to 1,000 competitors.

The software provider could potentially become a coordination intermediary if it:

• collects competitors' sensitive pricing information;
• processes it into recommendations;
• communicates strategic information;
• designs systems to respond to competitors;
• encourages supra-competitive pricing; or
• knowingly facilitates coordination.

Future legislation may therefore impose specific obligations on high-risk algorithm intermediaries.

XV. Safe-Harbour and Compliance Framework

A future regulatory framework could establish a compliance safe harbour for systems satisfying requirements such as:

1. independent data sources;
2. no exchange of confidential competitor information;
3. documented competition-risk assessment;
4. algorithmic auditability;
5. human compliance oversight;
6. automatic detection of unusual coordination;
7. emergency suspension capability; and
8. preservation of relevant logs.

Such a framework could encourage innovation while maintaining competition safeguards.

XVI. AI Agents and Machine-to-Machine Negotiation

The next generation of automated agreements may involve AI agents negotiating with other AI agents.

For example:

AI Buyer → AI Negotiator → AI Supplier Agent → Automated Contract → Blockchain Execution

The agents could independently negotiate:

• price;
• quantity;
• delivery;
• financing;
• insurance;
• warranties;
• renewal;
• dispute mechanisms.

This creates a major question:

Who is legally responsible for an agreement created by autonomous agents?

Future regulation may need to establish that the legal responsibility remains connected to the businesses that:

• deploy the agents;
• determine their objectives;
• provide their authority;
• supply their data; and
• control their commercial activity.

XVII. Competition Law and Autonomous Learning

Machine-learning systems can create a special problem because their behaviour may not have been explicitly programmed.

A system could discover that:

"When competitors raise prices, maintaining a high price produces higher returns."

If the system repeatedly learns this strategy, the resulting market may become less competitive.

The regulator must then distinguish:

A. Independent adaptation

The algorithm independently responds to market conditions.

B. Coordinated learning

The algorithm learns from competitors in a way that facilitates sustained coordination.

C. Human-enabled coordination

The firm knowingly designs or permits the system to coordinate.

This distinction will become increasingly important.

XVIII. Proposed Future Regulatory Model

A comprehensive framework could contain five levels.

Level 1 — Registration and documentation

High-risk automated commercial systems should have adequate internal documentation.

Level 2 — Competition impact assessment

Firms should assess whether the system can:

• coordinate prices;
• exchange sensitive information;
• discriminate against rivals;
• exclude competitors; or
• facilitate market allocation.

Level 3 — Continuous monitoring

Companies should monitor the actual market effects of deployed systems.

Level 4 — Regulatory audit

Competition authorities should have appropriate powers to inspect:

• logs;
• APIs;
• model documentation;
• contractual configurations;
• relevant communications; and
• algorithmic outputs.

Level 5 — Corrective measures

Potential remedies could include:

• modification of algorithms;
• deletion of problematic parameters;
• information-access restrictions;
• interoperability requirements;
• behavioural commitments;
• structural remedies where legally appropriate; and
• penalties for established violations.

XIX. Challenges for Competition Authorities

1. Attribution

Who is responsible for machine-generated conduct?

2. Explainability

Some machine-learning systems cannot easily explain individual decisions.

3. Evidence preservation

Algorithms may change continuously.

4. Cross-border enforcement

Cloud systems can operate across several jurisdictions simultaneously.

5. Technical expertise

Competition authorities need economists, lawyers, data scientists and software specialists.

6. False positives

Similar algorithmic behaviour does not necessarily prove collusion.

7. Innovation concerns

Over-regulation could discourage legitimate automation.

Therefore, regulation should distinguish technological complexity from unlawful coordination.

XX. Relationship Between Competition Law and Contract Law

Automated agreements sit at the intersection of:

• contract law;
• competition law;
• consumer protection;
• data protection;
• AI regulation;
• cybersecurity;
• electronic transactions;
• blockchain regulation; and
• platform regulation.

An agreement may be technologically valid as a contract while simultaneously raising competition-law issues.

Thus:

Contractual enforceability and competition-law legality are separate questions.

XXI. Future Competition-Law Doctrine

A future doctrine of automated agreements could be built around six principles:

Principle 1 — Technological neutrality

The same competition rule should generally apply whether conduct is performed manually or through software.

Principle 2 — Substance over code

Authorities should examine the economic and competitive substance rather than merely the technical architecture.

Principle 3 — Attribution through control

Responsibility should consider who designed, deployed, controlled and monitored the automated system.

Principle 4 — Evidence through digital records

Logs, code, APIs and model documentation should become important forms of competition evidence.

Principle 5 — Risk-based regulation

Not every algorithm requires intensive regulatory intervention. Greater scrutiny should apply to systems capable of materially affecting competition.

Principle 6 — Human accountability

Automation should not create an artificial legal gap in which unlawful conduct becomes immune merely because a machine executed it.

XXII. Key Case-Law Principles — Consolidated

These cases collectively demonstrate that the technological mechanism does not by itself determine competition-law liability. The decisive questions concern agreement, communication, coordination, knowledge, facilitation, market effects and attribution.

Conclusion

Automated agreements represent a transition from human-mediated contracting to machine-mediated commerce. Competition law must therefore move beyond the traditional image of competitors sitting together and expressly agreeing on commercial terms.

The future regulatory challenge will involve systems in which:

humans design → algorithms negotiate → machines exchange information → AI systems adapt → smart contracts execute → markets respond automatically.

The appropriate competition-law framework should preserve the fundamental prohibition on anticompetitive coordination while recognising that algorithmic similarity is not automatically collusion.

The most important future concepts are therefore algorithmic accountability, auditability, attribution, digital evidence, competition-by-design, intermediary responsibility, monitoring obligations and technologically neutral enforcement. The objective is not to prohibit automated agreements as such, but to ensure that automation does not become a mechanism for concealing, facilitating or continuously enforcing restrictions of competition.