Competition Law and Competition Implications of Autonomous Market-Making Systems

1. Introduction

Autonomous market-making systems (AMMS) are algorithmic or AI-driven systems that automatically place, cancel, modify, and price buy and sell orders in financial or other electronic markets. Unlike traditional market makers, an autonomous system may continuously determine:

• bid and ask prices;
• order size;
• inventory exposure;
• spread levels;
• order placement and cancellation;
• responses to competitors' quotes;
• liquidity allocation; and
• trading strategies,

with limited or no human intervention.

Market making can increase liquidity, reduce spreads and improve price discovery. However, autonomous systems also create distinctive competition-law risks because several competing systems may continuously observe one another and adapt their conduct at machine speed.

The central legal question is therefore:

When does autonomous market making remain legitimate competition, and when does algorithmic interaction become an instrument or mechanism of anti-competitive coordination or market manipulation?

The existing case law does not establish a general rule that autonomous algorithms are themselves unlawful. Instead, existing cases provide principles concerning algorithmic coordination, information exchange, price fixing, market manipulation, exclusionary conduct and automated trading.

2. Meaning and Characteristics of Autonomous Market-Making Systems

An autonomous market maker generally performs five functions:

A. Continuous quotation

The system simultaneously offers to buy and sell an asset.

B. Dynamic spread determination

The algorithm adjusts the bid-ask spread according to:

• volatility;
• inventory;
• order-book depth;
• market demand;
• competitor quotations;
• transaction costs;
• liquidity conditions.

C. Automated order management

The system can place, modify and cancel thousands of orders without individual human approval.

D. Machine-to-machine interaction

Competing algorithms may react to each other's:

• prices;
• quantities;
• cancellations;
• liquidity;
• trading speed.

E. Autonomous learning

Machine-learning systems can modify trading strategies based on observed market outcomes.

This last characteristic creates the greatest competition-law difficulty because an algorithm may potentially discover a commercially profitable strategy without its operator expressly instructing it to coordinate with competitors.

3. Competition-Law Framework

A. Anti-competitive agreements

The traditional cartel provisions of competition law prohibit agreements or concerted practices involving:

• price fixing;
• market allocation;
• output restriction;
• bid manipulation;
• exchange of competitively sensitive information.

In India, the principal provision is Section 3 of the Competition Act, 2002.

In the EU, the principal framework is Article 101 TFEU.

In the United States, Section 1 of the Sherman Act is particularly relevant.

An autonomous algorithm does not automatically create an "agreement." The difficult question is whether the conduct of the firms operating the systems demonstrates some form of coordination, communication, commitment or facilitating mechanism.

4. Algorithmic Tacit Coordination

One of the most important concerns is tacit algorithmic coordination.

Suppose five autonomous market makers independently use reinforcement-learning systems. Each system discovers that aggressive undercutting reduces profitability. Eventually, each algorithm learns to maintain a wider spread.

There may be:

• no meeting;
• no email;
• no explicit agreement;
• no communication between traders.

Nevertheless, the resulting market may display persistently supra-competitive spreads.

This creates a distinction between:

Explicit algorithmic collusion

Human actors agree to coordinate and use software to implement the agreement.

and

Autonomous algorithmic coordination

Independent algorithms learn or converge on strategies producing coordinated outcomes without an express human agreement.

The first is much easier to address under traditional cartel law. The second presents a difficult question concerning the requirement of an agreement or concerted practice. Contemporary competition scholarship specifically identifies autonomous pricing agents as creating this problem.

5. Information Exchange

Autonomous market-making systems can process enormous amounts of market information.

Competition concerns arise if the system receives or exchanges:

• confidential order-flow information;
• future pricing intentions;
• inventory positions;
• planned liquidity withdrawals;
• proprietary trading strategies;
• non-public customer information.

Even where the ultimate price is generated by software, the underlying information architecture may facilitate coordination.

The important distinction is between public market signals and competitively sensitive private information.

A market maker responding to publicly visible order-book information normally represents ordinary competition. Sharing a competitor's confidential information to enable coordinated pricing presents a substantially different issue.

6. Market Manipulation and Competition Law

Autonomous market making also intersects with financial-market regulation.

A system may theoretically be programmed to create artificial impressions of:

• supply;
• demand;
• liquidity;
• market depth;
• price direction.

Examples include:

• spoofing;
• layering;
• wash trading;
• quote stuffing;
• artificial liquidity withdrawal;
• coordinated order cancellation.

These activities may violate securities or commodities legislation independently of competition law.

Thus, AMMS compliance requires both competition-law compliance and market-integrity compliance.

7. Dominance and Autonomous Market-Making Platforms

An autonomous market-making infrastructure could itself become a dominant platform.

Competition issues may arise where one operator controls:

• the principal trading infrastructure;
• access to liquidity;
• essential market data;
• matching technology;
• low-latency infrastructure;
• market-making APIs;
• transaction-routing infrastructure.

A dominant platform might theoretically engage in:

• discriminatory access;
• exclusionary rebates;
• self-preferencing;
• refusal of access;
• discriminatory latency;
• preferential order routing;
• tying market data to execution services.

The relevant competition-law analysis would depend upon market definition, dominance, foreclosure effects and legitimate business justification.

8. Six Important Case Laws

1. Samir Agrawal v. Competition Commission of India — Supreme Court of India (2021)

This is one of the most important Indian cases for understanding algorithmic pricing.

The case concerned allegations that Ola and Uber's algorithmic pricing systems facilitated price coordination among drivers.

The argument was that individual drivers could not independently negotiate prices because the platform algorithm determined the fare.

The Supreme Court ultimately rejected the competition-law challenge, emphasizing the absence of the necessary agreement or concerted arrangement between the relevant parties.

Principle

The case demonstrates that:

The existence of an algorithmically determined price does not, by itself, establish a cartel or anti-competitive agreement.

For autonomous market-making systems, this means that regulators would still need to establish the legally relevant connection between algorithmic conduct and prohibited coordination.

2. United States v. Topkins — U.S. Department of Justice

In United States v. Topkins, an online seller participated in a price-fixing agreement concerning posters sold through an online marketplace.

The participant used computer code and an automated pricing system to implement the agreed pricing strategy.

This is particularly relevant to autonomous systems because it demonstrates that:

Software does not shield an underlying human cartel from antitrust liability.

Where competitors agree on prices first and subsequently use algorithms to execute the agreement, the algorithm is simply the technological mechanism through which the cartel operates.

Relevance to AMMS

If competing market makers agree to maintain:

• minimum spreads;
• particular bid prices;
• common liquidity levels; or
• coordinated withdrawal strategies,

and then program autonomous systems to implement those arrangements, traditional cartel law remains highly relevant.

3. United States v. RealPage, Inc.

The RealPage litigation represents a major modern development concerning algorithmic coordination.

The U.S. Department of Justice alleged that competing landlords supplied non-public, competitively sensitive information to RealPage and that its algorithm generated pricing recommendations using information derived from competitors.

The DOJ alleged violations of Sections 1 and 2 of the Sherman Act. In 2025, the DOJ announced a proposed settlement addressing information sharing and alignment of pricing.

Principle

The significance for autonomous market makers is that an algorithm need not merely be a neutral computational tool.

An algorithm can potentially become the institutional mechanism through which competitors' decisions are coordinated.

The critical questions include:

1. What data enters the system?
2. Who supplies the data?
3. Is the data competitively sensitive?
4. Are competitors aware of how the data is used?
5. Are firms effectively delegating pricing decisions to a common mechanism?
6. Does the system reduce independent competitive decision-making?

4. CFTC v. Navinder Singh Sarao

The Sarao proceedings concerned manipulation and spoofing in the E-mini S&P 500 futures market.

According to the CFTC, Sarao used an automated "layering" program that placed large orders at multiple price levels and repeatedly modified them, while attempting to avoid execution. The alleged purpose was to affect the market and benefit from resulting price movements.

Principle

The case demonstrates that:

Automation does not eliminate legal responsibility for manipulative market conduct.

For AMMS, automated order placement and cancellation therefore require controls capable of distinguishing legitimate liquidity provision from manipulative strategies.

5. In re JPMorgan Chase & Co. — CFTC (2020)

The CFTC's 2020 JPMorgan enforcement action involved extensive spoofing and manipulation in precious-metals and U.S. Treasury futures.

The CFTC stated that hundreds of thousands of spoof orders were involved and imposed approximately $920.2 million in monetary relief.

Principle

The case demonstrates the enormous enforcement exposure associated with systematic automated or high-frequency trading misconduct.

For autonomous market-making systems, compliance cannot depend exclusively upon individual trader supervision because:

• order volumes are extremely high;
• decisions occur at machine speed;
• problematic patterns may emerge only statistically;
• individual orders may appear innocuous when examined separately.

Therefore, system-level surveillance becomes essential.

6. In re Tower Research Capital LLC — CFTC (2019)

The CFTC brought enforcement proceedings against Tower Research Capital concerning spoofing in equity-index futures.

The CFTC stated that the conduct involved thousands of spoofing incidents and imposed approximately $67.4 million in monetary relief.

Principle

The case illustrates the importance of:

• algorithmic controls;
• trader supervision;
• order surveillance;
• cancellation monitoring;
• detection of recurring manipulation patterns.

For autonomous market makers, these controls must operate continuously rather than merely after suspicious conduct is identified.

9. Comparative Significance of the Six Cases

10. Key Competition Concerns

A. Autonomous price coordination

Multiple AMMS may independently converge upon similar pricing strategies.

The legal difficulty is distinguishing:

competitive adaptation
from
unlawful coordination.

B. Spread coordination

Market makers may independently discover that maintaining wider spreads produces greater returns.

If algorithms systematically maintain those spreads, regulators may investigate whether there is:

• communication;
• common software;
• common data;
• explicit instructions;
• coordinated restrictions;
• conscious parallelism supported by facilitating conduct.

C. Information pooling

A common algorithmic infrastructure can create a hub-and-spoke risk.

For example:

Market Maker A →
Market Maker B → Common Algorithm ← Market Maker C

If the central system receives commercially sensitive information from each competitor and uses it to determine prices for all participants, the competition concerns become substantially greater.

D. Algorithmic exclusion

A dominant trading platform could potentially use its algorithm to:

• prioritize its own market-making arm;
• disadvantage independent market makers;
• manipulate access fees;
• provide discriminatory latency;
• restrict API functionality.

Such conduct could potentially raise abuse-of-dominance concerns.

11. Self-Learning Algorithms and Tacit Collusion

The most difficult future scenario involves reinforcement-learning market makers.

Consider:

• 10 independent market makers;
• each uses reinforcement learning;
• each observes public order-book data;
• each adjusts its spread automatically;
• no human communicates with another;
• the systems gradually discover that aggressive competition is unprofitable.

Suppose the algorithms eventually settle into:

Bid = market reference − 50 basis points
Ask = market reference + 50 basis points.

The resulting spread could be materially higher than under competitive conditions.

The competition-law difficulty is that there may be no traditional agreement.

This is why autonomous pricing research identifies machine-learning systems as potentially capable of sustaining coordinated outcomes without conventional human communication.

12. Difference Between Legitimate Adaptation and Anti-Competitive Coordination

The mere presence of similar algorithmic behaviour should not automatically be treated as proof of an infringement.

13. Market Definition Issues

Competition analysis may require careful definition of the relevant market.

Potential markets include:

Product market

• equity market making;
• derivatives market making;
• cryptocurrency market making;
• fixed-income market making;
• ETF market making;
• prediction-market liquidity provision;
• decentralized-exchange liquidity provision.

Geographic market

The relevant market may be:

• national;
• regional;
• global;
• exchange-specific;
• asset-specific.

Digital markets make geographic definition particularly difficult because an autonomous market maker can operate across jurisdictions almost instantaneously.

14. Network Effects and Data Advantages

Autonomous market-making systems can generate powerful feedback loops:

More trading → more data → better prediction → better execution → more trading

This may produce substantial competitive advantages.

A leading AMMS may accumulate:

• order-book data;
• execution data;
• latency information;
• volatility histories;
• customer-flow information;
• liquidity information.

A competition authority may therefore examine whether data advantages constitute a barrier to entry.

15. Barriers to Entry

Potential barriers include:

1. high computing costs;
2. access to low-latency infrastructure;
3. proprietary trading data;
4. exchange connectivity;
5. technical expertise;
6. capital requirements;
7. network effects;
8. regulatory licensing;
9. access to liquidity;
10. sophisticated machine-learning infrastructure.

Where these barriers become substantial, autonomous market-making markets may become concentrated.

16. Competition Implications of Common AI Infrastructure

A particularly important risk is the emergence of a shared autonomous market-making service.

Suppose competing firms outsource their market-making decisions to the same AI provider.

The provider receives:

• Firm A's trading information;
• Firm B's trading information;
• Firm C's trading information.

It then recommends or automatically executes trades for each firm.

This could potentially transform an apparently competitive market into a technologically mediated coordination structure.

The RealPage controversy illustrates why the architecture of information sharing and algorithmic decision-making matters for competition analysis.

17. Competition Law and Financial Regulation: Overlapping Regimes

AMMS may simultaneously fall within several regulatory regimes.

Competition law

Addresses:

• cartels;
• coordination;
• abuse of dominance;
• exclusionary conduct;
• anti-competitive information exchange.

Securities regulation

Addresses:

• market manipulation;
• insider trading;
• misleading orders;
• market integrity.

Derivatives regulation

May address:

• spoofing;
• manipulation;
• disruptive trading;
• position-related conduct.

Data regulation

May address:

• confidential information;
• personal data;
• cross-platform data sharing.

Consequently, one autonomous trading architecture can produce multiple forms of regulatory exposure.

18. Evidentiary Problems

Autonomous systems create difficult evidentiary questions.

Competition authorities may need to establish:

• what the algorithm was instructed to do;
• what data it received;
• how it learned;
• what parameters changed;
• whether competitors communicated;
• whether humans intervened;
• whether the resulting coordination was foreseeable;
• whether the conduct produced competitive harm.

Traditional documentary evidence may be insufficient.

Investigations may therefore require:

• source-code examination;
• model logs;
• version histories;
• training datasets;
• API records;
• order-level data;
• system prompts;
• model weights where legally accessible;
• communications between firms and software providers.

19. Compliance Requirements for Autonomous Market Makers

A robust competition-compliance programme should include:

1. Algorithmic governance

Every autonomous system should have a clearly documented purpose and permitted operating parameters.

2. Data controls

Competitively sensitive information should not be unnecessarily shared between competing firms.

3. Independent pricing controls

Each competitor should retain genuine independent decision-making.

4. Audit trails

The operator should maintain records of:

• algorithm versions;
• parameter changes;
• training data;
• decisions;
• interventions.

5. Collusion detection

Systems should monitor abnormal convergence in:

• spreads;
• quotes;
• order sizes;
• cancellations.

6. Manipulation controls

Algorithms should be tested for:

• spoofing;
• layering;
• wash trading;
• quote stuffing.

7. Human override

A suitable emergency mechanism should allow problematic autonomous behaviour to be suspended.

8. Periodic competition testing

Algorithms should be tested not merely for profitability but also for potential competitive effects.

20. Potential Remedies

Where competition concerns arise, authorities could consider:

Behavioural remedies

• prohibit sensitive information sharing;
• require independent algorithmic decision-making;
• impose monitoring obligations;
• require algorithmic audits;
• prohibit discriminatory access.

Structural remedies

In serious cases involving dominant infrastructure, possible structural measures could include:

• separation of market-making and exchange functions;
• divestiture;
• interoperability requirements;
• access obligations.

Technical remedies

Authorities may also require:

• audit logs;
• kill switches;
• independent monitoring;
• algorithm certification;
• explainability mechanisms.

The appropriate remedy would depend upon the particular infringement and market structure.

21. Autonomous Market Making in Decentralised Finance

AMMS can be particularly significant in decentralised finance (DeFi).

Automated market makers such as decentralized liquidity protocols replace traditional order-book market making with mathematical pricing mechanisms.

Competition questions may concern:

• concentration of liquidity;
• protocol governance;
• token-holder coordination;
• liquidity-provider incentives;
• interoperability;
• front-running;
• MEV;
• access to transaction ordering;
• common algorithmic pricing formulas.

A protocol may technically operate without a conventional corporation, making attribution and jurisdiction particularly complicated.

22. The Indian Competition-Law Perspective

For India, the principal provisions likely to be relevant include:

Section 3

Anti-competitive agreements, including arrangements involving:

• price fixing;
• limiting supply;
• market allocation;
• bid manipulation.

Section 4

Abuse of dominant position, including potentially:

• discriminatory conditions;
• denial of market access;
• leveraging;
• exclusionary practices.

Sections 5 and 6

These become relevant where autonomous market-making platforms are involved in qualifying combinations or acquisitions.

Section 19

The Competition Commission of India may investigate agreements and abuse of dominance.

The Samir Agrawal litigation is particularly important because it shows that algorithmic price determination alone does not necessarily satisfy the agreement requirement under Section 3.

23. Emerging Legal Test for Autonomous Market-Making Systems

A useful analytical framework is:

Step 1 — Identify the algorithm

What exactly does the AMMS do?

Step 2 — Identify the market

What product, trading venue and geographic market are affected?

Step 3 — Identify the parties

Who owns, operates, trains or controls the system?

Step 4 — Identify the data

What information does the algorithm receive?

Step 5 — Examine interaction

Does the system interact with competing systems?

Step 6 — Examine coordination

Is there evidence of:

• communication;
• common instructions;
• shared software;
• shared data;
• coordinated objectives?

Step 7 — Examine effects

Has competition been reduced through:

• higher spreads;
• reduced liquidity;
• exclusion;
• increased barriers to entry?

Step 8 — Examine manipulation

Are orders being used to create artificial market signals?

Step 9 — Establish causation

Did the algorithm cause or materially facilitate the competitive harm?

Step 10 — Determine remedy

The remedy should address the actual source of the competition problem rather than merely the fact that AI was used.

24. Key Legal Distinction

The most important conceptual distinction is:

Autonomy is not itself anti-competitive.

An autonomous system may generate substantial efficiencies through:

• faster price discovery;
• lower transaction costs;
• increased liquidity;
• reduced spreads;
• better inventory management.

The competition problem arises where autonomy is combined with conduct such as:

autonomy + agreement → cartel risk

autonomy + sensitive information sharing → coordination risk

autonomy + dominance → exclusionary-conduct risk

autonomy + manipulative orders → market-manipulation risk

autonomy + common infrastructure → hub-and-spoke risk

autonomy + self-learning convergence → unresolved tacit-collusion problem

25. Conclusion

Autonomous market-making systems represent a significant development in competition law because they shift portions of competitive decision-making from humans to machines.

The existing case law provides an important foundation. Samir Agrawal demonstrates that algorithmic pricing by itself does not establish an anti-competitive agreement. Topkins demonstrates that algorithms can nevertheless be used to implement conventional price-fixing arrangements. RealPage illustrates the importance of shared competitively sensitive information and common algorithmic pricing infrastructure. The Sarao, JPMorgan and Tower Research proceedings demonstrate the separate but closely connected dangers of automated market manipulation and spoofing.

The most difficult future issue is fully autonomous tacit coordination: competing AI systems independently learning that reduced competitive intensity produces greater profits. Current competition law is considerably clearer where humans communicate or deliberately employ algorithms to implement an agreement than where independent algorithms reach coordinated outcomes without human communication.

Accordingly, the central regulatory objective should be to preserve the efficiency benefits of autonomous market making while ensuring independent competitive decision-making, controlled information flows, transparent algorithmic governance, and effective detection of manipulative or exclusionary behaviour.