Competition Law and Machine-Negotiated Agreements and Antitrust Concerns

1. Introduction

Machine-negotiated agreements arise when software, artificial intelligence (“AI”) agents, pricing algorithms, procurement systems, autonomous contracting systems, or other computational tools negotiate commercial terms with other machines or with systems operated by competing firms.

Examples include:

• AI agents negotiating prices automatically;
• algorithms negotiating discounts, rebates, delivery charges or commissions;
• procurement bots interacting with suppliers' systems;
• autonomous platforms adjusting contractual terms in response to competitors;
• AI agents exchanging offers and counteroffers;
• algorithms using competitors' information to determine commercial terms; and
• competing firms delegating negotiations to a common AI or software provider.

The central competition-law problem is that the fact that a machine rather than a human conducts the negotiation does not necessarily remove the underlying conduct from antitrust law. Existing competition law generally focuses on the economic relationship and the conduct of the undertaking, rather than on whether the final decision was made manually or computationally.

The U.S. Department of Justice has expressly taken the position that competitors cannot evade antitrust rules by using algorithms to accomplish conduct that would be unlawful if carried out by humans.

A particularly important distinction is between:

1. legitimate autonomous negotiation — each firm independently instructs its system to maximize its own interests; and
2. algorithmic coordination — competing firms use machines, shared data, common software or an intermediary to reduce or eliminate independent competitive decision-making.

2. Meaning of Machine-Negotiated Agreements

A machine-negotiated agreement may involve several layers.

A. Human-to-machine instruction

A company establishes parameters such as:

"Never pay more than ₹100 per unit."

The AI then negotiates with suppliers.

This is generally not problematic merely because the negotiation is automated.

B. Machine-to-machine negotiation

Two autonomous systems exchange:

• price;
• quantity;
• delivery terms;
• credit terms;
• rebates;
• exclusivity;
• service levels; or
• other contractual provisions.

Again, automation alone does not constitute an antitrust violation.

C. Competitor-to-competitor algorithmic coordination

The situation becomes substantially more problematic where competing undertakings configure their systems to coordinate:

• prices;
• output;
• discounts;
• capacity;
• customers;
• market allocation; or
• other competitively sensitive variables.

D. Common-agent negotiation

A particularly important scenario is where several competitors employ the same AI provider or algorithm and feed that system non-public competitive information.

The machine can then become a technological intermediary through which competitors coordinate.

The DOJ's RealPage litigation illustrates this concern: competing landlords allegedly supplied competitively sensitive information to a common pricing system that generated rental recommendations.

3. Fundamental Antitrust Principle

The principal question is not:

"Did the machine make the agreement?"

It is:

"Did competing undertakings cease making independent competitive decisions?"

This distinction is critical.

Suppose five competing sellers independently instruct five unrelated algorithms to negotiate the lowest possible wholesale price. The resulting contracts may be entirely legitimate.

Conversely, if five competitors agree to let a common algorithm determine their prices using their confidential data, the algorithm may become the mechanism through which an unlawful horizontal agreement operates.

The DOJ and FTC stated in the hotel-algorithm litigation that competitors cannot lawfully cooperate on prices through an algorithm simply because they do not communicate with one another directly.

4. Applicable Competition-Law Framework

A. United States

The principal provisions include:

Section 1, Sherman Act

Section 1 addresses agreements, combinations and conspiracies restraining interstate commerce.

Machine negotiation becomes relevant where the algorithm facilitates an agreement between competitors.

Section 2, Sherman Act

Section 2 becomes relevant where algorithmic infrastructure is used to:

• monopolize;
• maintain monopoly power;
• exclude competitors;
• control an essential technological input; or
• reinforce an existing dominant position.

FTC Act

Section 5 may also become relevant to unfair methods of competition, depending upon the conduct.

5. European Union

Article 101 TFEU prohibits agreements, decisions and concerted practices having the object or effect of restricting competition.

Machine-mediated conduct can therefore raise questions concerning:

• agreement;
• concerted practice;
• exchange of competitively sensitive information;
• price coordination;
• market allocation;
• output restriction; and
• facilitating practices.

Article 102 TFEU may become relevant where a dominant technology provider uses an AI system to exclude rivals or impose discriminatory terms.

The Eturas judgment is particularly important because the Court of Justice considered automated restrictions implemented through a common computerized booking system.

6. India

Under the Competition Act, 2002, machine-negotiated agreements can potentially fall within:

Section 3

Section 3(1) prohibits agreements that cause or are likely to cause an appreciable adverse effect on competition.

Section 3(3) is especially relevant to horizontal arrangements involving:

• price fixing;
• limitation of production or supply;
• market sharing;
• bid rigging; and
• similar coordination.

Section 4

Section 4 may apply where a dominant AI platform, algorithm provider or digital intermediary abuses its dominant position.

Possible concerns include:

• discriminatory access to an AI negotiation system;
• self-preferencing;
• exclusionary interoperability conditions;
• tying;
• refusal to provide access;
• discriminatory algorithmic terms; and
• exploitation of competitively sensitive data.

The fact that the agreement is generated electronically should not, by itself, change the substantive competition-law analysis.

7. Six Important Case Laws

Because fully autonomous machine-to-machine negotiation is a relatively new phenomenon, there is not yet a large body of reported judgments dealing specifically with AI agents independently negotiating contracts. The following cases are therefore important because they establish the legal principles that apply when algorithms, computerized systems or intermediaries are used to implement or facilitate competitive coordination.

Case 1: United States v. David Topkins (2015)

Facts

David Topkins and other sellers operated on Amazon Marketplace.

They allegedly agreed to fix prices for certain posters and used algorithm-based pricing software to implement their agreement.

The software collected competitor pricing information and applied rules established by the sellers.

Topkins pleaded guilty to the price-fixing offence.

Legal significance

This is one of the clearest examples demonstrating that algorithmic implementation does not immunize a cartel.

The human agreement remained the source of the antitrust violation; the algorithm merely made the coordination more efficient and largely self-executing.

Principle

A machine can be the instrument through which an unlawful horizontal agreement is implemented.

This principle is directly relevant to machine-negotiated agreements.

Case 2: United States v. Daniel William Aston & Trod Ltd. (2015–2016)

Facts

Trod Ltd. and its director were prosecuted in connection with an alleged agreement concerning prices of posters sold through Amazon Marketplace.

The defendants used algorithm-based pricing software as part of the scheme.

Trod subsequently pleaded guilty to fixing prices of posters sold online.

Legal significance

The case demonstrates that automated pricing systems cannot be used to disguise a conventional cartel.

The algorithm did not become a legally independent actor. The conduct was attributed to the businesses that configured and used the technology.

Principle

Where businesses deliberately configure software to implement a horizontal price-fixing agreement, the automated nature of execution does not break the chain of antitrust responsibility.

Case 3: Eturas UAB and Others v. Lithuanian Competition Council, Case C-74/14

Facts

Travel agencies used a common computerized booking system.

The system administrator sent a message concerning restrictions on discounts available through the online booking system, and the system automatically restricted discounts.

The case reached the Court of Justice of the European Union.

The Court considered whether the use of the computerized system and the circumstances surrounding the system administrator's communication could constitute evidence of a concerted practice under Article 101 TFEU.

Legal significance

This case is particularly valuable for machine-negotiated agreements because it demonstrates that technological architecture can become relevant evidence of coordination.

At the same time, the Court emphasized evidentiary safeguards: merely receiving a system message was not automatically sufficient to establish participation in the concerted practice.

Principle

Automated restrictions can constitute part of a competition-law infringement, but authorities must still establish the required elements of agreement or concerted practice.

This is particularly important for autonomous AI systems because parallel machine behaviour should not automatically be equated with an unlawful agreement.

Case 4: United States v. Airline Tariff Publishing Co. (ATP), 836 F. Supp. 9 (D.D.C. 1993)

Facts

Several airlines used the Airline Tariff Publishing Company computerized system.

The system allowed airlines to communicate fare information and make changes to proposed fares.

The DOJ challenged practices that facilitated coordination of airline prices.

The resulting proceedings produced restrictions on the airlines' use of the system for certain coordinating practices.

Legal significance

ATP is an important predecessor to modern algorithmic-collusion cases.

The technology itself was not necessarily unlawful. The concern was that its architecture provided competitors with a mechanism for communicating and coordinating prices.

Principle

A common computerized communication infrastructure can become a vehicle for collusion when competitors use it to reduce uncertainty about their future competitive conduct.

That principle becomes even more significant when modern AI systems negotiate or react automatically.

Case 5: Cornish-Adebiyi v. Caesars Entertainment

Facts

The case concerns allegations involving hotel-room pricing algorithms.

The plaintiffs alleged that competing hotels used a common algorithmic pricing mechanism to coordinate room prices.

The DOJ and FTC filed a statement of interest explaining that competitors cannot use an algorithm to accomplish conduct that would be unlawful if carried out by humans.

Legal significance

The government agencies emphasized two particularly important points.

First, plaintiffs do not necessarily need to identify direct competitor-to-competitor communications where the allegations establish concerted conduct through an algorithm provider.

Second, competitors cannot avoid antitrust liability simply because they retain some discretion over the final price.

Principle

Delegating the negotiation or pricing function to a common algorithm does not necessarily eliminate the existence of an agreement.

This is highly relevant to AI agents that negotiate independently within parameters supplied by competing businesses.

Case 6: United States and States v. RealPage, Inc.

Facts

The DOJ brought an antitrust action concerning RealPage's revenue-management software.

According to the government's allegations, competing landlords supplied non-public information concerning:

• rents;
• lease terms;
• vacancies; and
• other commercially sensitive information.

RealPage's software then used this information to generate pricing recommendations.

The government alleged violations of Sections 1 and 2 of the Sherman Act.

The case subsequently produced proposed settlements and additional proceedings involving landlords. In November 2025, the DOJ announced a proposed settlement requiring restrictions on the use of competitors' non-public information in the software and changes to certain algorithmic features.

Legal significance

RealPage is particularly significant for machine-negotiated arrangements because the technology functions as a common computational intermediary between competitors.

The alleged concern was not simply that prices happened to be similar. It concerned the combination of:

1. competitor data sharing;
2. common algorithmic processing;
3. pricing recommendations;
4. mechanisms encouraging adherence; and
5. reduced independent pricing decisions.

Principle

An AI or algorithmic intermediary can become an important component of an unlawful coordination theory when competitors supply it with competitively sensitive information and rely on it to determine or influence their commercial decisions.

8. Comparative Importance of the Six Cases

9. Machine-Negotiated Agreements: Major Antitrust Concerns

A. Price coordination

The most obvious concern is autonomous price negotiation.

Suppose competing sellers instruct their AI systems:

"Negotiate with competitors but never offer below the industry price generated by the common system."

If the systems interact and converge on a common price, the legal question becomes whether there is an underlying agreement or concerted practice.

B. Information exchange

Machine negotiation creates enormous quantities of data.

An AI agent may receive:

• competitor prices;
• inventory;
• costs;
• future capacity;
• production schedules;
• discounts;
• customer information;
• margins; and
• strategic plans.

If competing firms permit a common AI system to process such non-public information, competition authorities may examine whether the arrangement reduces strategic uncertainty between competitors.

C. Common algorithm problem

Suppose:

Competitor A → Common AI → Competitor B

The common AI receives information from both companies.

It may therefore become a digital hub connecting competing firms.

This creates a potential hub-and-spoke theory.

The legal issue becomes whether the vertical relationships between each company and the AI provider, taken together with the surrounding circumstances, constitute a horizontal coordination arrangement.

10. Hub-and-Spoke Machine Negotiation

A particularly important future scenario is:

A → AI Provider ← B

A and B are competitors.

Both provide confidential information to the same AI.

The AI negotiates prices or contract terms on behalf of both.

The danger increases if:

• the AI uses A's confidential information when negotiating with B;
• the AI uses B's information when negotiating with A;
• both firms know that this occurs;
• both firms agree to follow the AI's recommendations;
• the algorithm deliberately reduces competitive uncertainty; or
• firms collectively determine the algorithm's objectives.

This architecture can transform a seemingly vertical software arrangement into a potential horizontal coordination mechanism.

11. Autonomous AI and the "Agreement" Problem

The most difficult future legal question is:

Can independently acting AI systems create an antitrust agreement without a human explicitly agreeing to coordinate?

There are three different situations.

Situation 1 — Explicit human agreement

A and B agree:

"Our AI agents will maintain prices at ₹1,000."

This is the easiest case legally.

The machines simply execute the agreement.

Situation 2 — Human agreement to use a common coordinating mechanism

A and B do not directly agree on the precise price.

Instead:

"We will both use the same AI system and permit it to determine our prices using our respective market information."

This creates a much more difficult but highly significant antitrust question.

The legal analysis may focus on whether the agreement to use the system itself constitutes coordination.

The hotel-algorithm litigation and RealPage proceedings illustrate why this issue is receiving increasing attention.

Situation 3 — Fully autonomous machine coordination

A and B independently deploy AI agents.

No human communicates with the competitor.

Each AI observes the market and independently learns that coordinated pricing maximizes profit.

The machines eventually produce supra-competitive prices.

This is the hardest unresolved category.

Current antitrust law generally requires an agreement or concerted practice for traditional horizontal liability under provisions such as Sherman Act §1 and Article 101 TFEU.

Therefore:

Parallel algorithmic behaviour alone should not automatically be treated as proof of an agreement.

The Eturas decision illustrates the importance of evidence concerning actual participation and knowledge rather than simply relying on automated parallel behaviour.

12. Tacit Algorithmic Collusion

Algorithms can potentially react much faster than humans.

An AI can:

1. observe competitor price;
2. adjust its own price;
3. observe the competitor's response;
4. revise its strategy;
5. repeat the process thousands of times.

This may create a stable equilibrium without an express communication channel.

The resulting question is whether competition law should treat such autonomous coordination as:

• lawful conscious parallelism;
• unilateral conduct;
• concerted practice;
• an agreement; or
• potentially an abuse of dominance under other circumstances.

There is currently substantial legal debate over how existing agreement requirements should apply to completely autonomous algorithmic coordination.

13. Delegation Does Not Necessarily Transfer Legal Responsibility

A company cannot necessarily defend itself by arguing:

"Our AI made the decision."

Competition law generally looks at the undertaking's conduct, organizational structure and relationship with the technology.

If management intentionally configures an AI agent to:

• fix prices;
• divide customers;
• suppress discounts;
• exchange confidential information; or
• coordinate with competitors,

the fact that the final decision was computational may provide no substantive defence.

Topkins and Trod demonstrate this proposition particularly clearly.

14. AI Negotiation and Vertical Agreements

Not every machine-negotiated agreement is horizontal.

For example:

Manufacturer → AI Procurement Agent → Supplier

could involve:

• quantity discounts;
• minimum purchase commitments;
• territorial restrictions;
• exclusivity;
• resale-price provisions;
• rebates; or
• tying.

These may raise vertical-restraint issues rather than cartel concerns.

The analysis must therefore identify the parties' positions in the supply chain.

15. Algorithmic Resale Price Maintenance

Suppose a manufacturer programs its distributor's AI:

"Never sell below ₹500."

The distributor's machine automatically rejects transactions below ₹500.

This may create resale-price-maintenance concerns depending upon the applicable jurisdiction and circumstances.

The fact that the distributor's machine automatically enforces the restriction does not necessarily change the underlying economic substance.

16. Market Allocation by AI

Machine negotiation could also allocate markets.

For example:

• AI 1 handles Northern India;
• AI 2 handles Southern India;
• neither system solicits customers allocated to the other.

If competing firms intentionally configure their systems this way, it could raise classic market-allocation concerns.

The technology would simply provide a sophisticated implementation mechanism.

17. Bid Rigging and Procurement Bots

Another major concern involves automated procurement.

Imagine five suppliers using AI bidding agents.

If they independently determine their bids, automated bidding may increase competition.

But if the systems are configured to:

• rotate winning suppliers;
• maintain predetermined margins;
• avoid bidding against one another;
• allocate government contracts; or
• exchange confidential bid information,

the technology could facilitate bid rigging.

The legal analysis should therefore focus on the competitive independence of the bidding process, rather than the technological sophistication of the bidding system.

18. Evidence and Digital Forensics

Machine-negotiated agreements create a major evidentiary advantage for competition authorities because AI systems generate extensive digital records.

Potential evidence includes:

• source code;
• system prompts;
• model instructions;
• API calls;
• negotiation logs;
• timestamps;
• system messages;
• pricing histories;
• training datasets;
• model versions;
• configuration files;
• internal communications;
• audit logs;
• data-access records; and
• records showing whether humans overrode AI decisions.

Consequently, future antitrust investigations may increasingly examine the machine's decision architecture rather than merely traditional emails and meetings.

19. "Human-in-the-Loop" Does Not Automatically Cure the Problem

A company might argue:

"A human approves every AI recommendation."

That fact may be relevant, but it is not necessarily determinative.

If:

• the AI obtains competitors' confidential information;
• generates coordinated prices;
• strongly pressures users to follow recommendations; and
• the human routinely accepts them,

the existence of nominal human approval may not eliminate the underlying competition concerns.

The DOJ's position in the hotel-algorithm litigation specifically addressed the argument that firms retain pricing discretion.

20. Competition Compliance for Machine-Negotiated Agreements

Businesses using autonomous negotiation systems should consider:

1. Independent-data architecture

Competitors should not unnecessarily expose competitively sensitive information to a common AI.

2. Separate models

Where appropriate, competitors should maintain technically and organizationally separate models.

3. Data minimization

The system should receive only information necessary for legitimate negotiations.

4. Auditability

Companies should preserve:

• prompts;
• model versions;
• inputs;
• outputs;
• overrides;
• negotiation histories.

5. Competition-law constraints

AI instructions should expressly prohibit:

• price fixing;
• market allocation;
• customer allocation;
• bid rotation;
• competitor information exchange; and
• other prohibited coordination.

6. Human oversight

Human review should be meaningful rather than merely ceremonial.

7. Common-provider safeguards

Where several competitors use the same AI provider, contractual and technical safeguards should prevent inappropriate use of one competitor's confidential information for another competitor.

21. A Useful Legal Test

A competition authority examining a machine-negotiated arrangement could conceptually ask:

Step 1 — Who are the parties?

Are the undertakings:

• competitors;
• suppliers and distributors;
• unrelated firms; or
• parts of the same undertaking?

Step 2 — What is the machine doing?

Is it:

• merely negotiating;
• recommending;
• enforcing;
• communicating;
• exchanging information; or
• determining the final commercial term?

Step 3 — What data does it receive?

Is the information:

• public;
• historical;
• aggregated; or
• current, individualized and competitively sensitive?

Step 4 — Is there human coordination?

Look for:

• agreements;
• common instructions;
• communications;
• shared objectives;
• common software arrangements.

Step 5 — Does the system reduce competitive independence?

This is the central economic question.

Step 6 — What is the market effect?

Consider:

• price;
• output;
• innovation;
• quality;
• choice;
• entry;
• switching costs.

22. Key Distinction: Automation vs Coordination

23. Key Doctrinal Lessons From the Case Law

The six cases collectively support several important propositions.

First

Technology is not a defence to cartel conduct.

Topkins and Trod demonstrate this particularly clearly.

Second

A computerized system can facilitate a concerted practice.

Eturas demonstrates the relevance of common computerized infrastructure to Article 101 analysis.

Third

A common technological intermediary may facilitate coordination between competitors.

RealPage illustrates this issue in modern algorithmic pricing.

Fourth

Direct competitor-to-competitor communication is not necessarily indispensable to an algorithmic-coordination theory.

The DOJ and FTC made this point in Cornish-Adebiyi.

Fifth

Parallel algorithmic outcomes alone should not automatically establish an agreement.

Eturas illustrates the importance of proving the necessary elements of participation and concerted conduct.

Sixth

The central issue remains preservation of independent competitive decision-making.

The more competitors delegate competitively sensitive decisions to a common technological mechanism, the greater the competition-law risk.

24. Emerging Legal Problem: AI Agents as "Electronic Negotiators"

Future AI agents may have authority to:

• identify counterparties;
• negotiate prices;
• make offers;
• accept counteroffers;
• negotiate warranties;
• negotiate exclusivity;
• exchange data;
• sign contracts; and
• continuously renegotiate terms.

At that stage, the traditional concept of an "agreement" becomes more complicated.

A machine might conclude a contract in seconds without any human intervention.

Nevertheless, competition law is unlikely to focus exclusively on the metaphysical question of whether the machine itself is capable of forming an agreement. The more practical question is likely to be whether the undertakings designed, authorized, deployed, or knowingly participated in the system that produced the coordination.

25. Conclusion

Machine-negotiated agreements represent an important new frontier of competition law.

The fundamental principle is relatively straightforward:

Automation does not transform an unlawful competitive agreement into a lawful one.

The difficult question arises where autonomous systems coordinate without an express human agreement.

The existing cases—particularly Topkins, Trod, Eturas, Airline Tariff Publishing, Cornish-Adebiyi and RealPage—show the development of the law from traditional computerized coordination toward modern AI-mediated competition concerns.

The emerging legal framework is therefore likely to distinguish between:

Independent machine negotiation → generally compatible with competition

and

Machine-mediated coordination between competitors → potentially prohibited anticompetitive conduct.