Global Transition to Autonomous Competitive Systems

Introduction

The global transition to autonomous competitive systems refers to the movement from conventional markets, in which human firms independently make most pricing, production, distribution, procurement, investment and strategic decisions, toward markets increasingly operated or influenced by AI agents, algorithms, autonomous software, smart contracts, automated marketplaces, algorithmic pricing systems and machine-controlled supply chains.

In such systems, competition may no longer be determined exclusively by human managerial decisions. Autonomous systems can:

• set or modify prices dynamically;
• allocate scarce resources;
• select suppliers and customers;
• negotiate contracts;
• optimise production and logistics;
• make investment or procurement decisions;
• coordinate through shared algorithms or data;
• determine search rankings and market access;
• learn from competitors' conduct;
• execute contractual or commercial decisions automatically.

This creates a fundamental competition-law question:

How should competition law respond when economically significant competitive decisions are made, adapted or executed autonomously by machines?

The existing doctrines of agreement, concerted practice, unilateral abuse, market power, causation, intent, attribution and liability were largely developed around human conduct. Autonomous competitive systems therefore create a potential gap between economic effects and traditional legal responsibility.

1. Meaning of Autonomous Competitive Systems

An autonomous competitive system is a market arrangement in which software or AI has substantial authority to make or execute economically consequential decisions without contemporaneous human approval.

A useful model is:

Data → AI/Algorithm → Prediction → Decision → Automated Execution → Market Response → New Data → Learning

For example, an autonomous procurement agent could:

1. monitor thousands of suppliers;
2. predict future prices;
3. negotiate automatically;
4. switch suppliers;
5. alter order volumes;
6. respond to competitors;
7. update its strategy continuously.

The important distinction is between automation and autonomy.

Ordinary automation

A human specifies:

"Buy 1,000 units when the price falls below ₹100."

The system executes a predetermined instruction.

Autonomous competition

An AI system is instructed:

"Optimise procurement costs."

It independently determines:

• what to purchase;
• from whom;
• when;
• at what quantity;
• what price to offer;
• whether to switch suppliers;
• and potentially how to respond to competing buyers.

The second situation creates substantially greater competition-law complexity.

2. Why Autonomous Systems Matter to Competition Law

Traditional competition assumes that competitors are relatively independent decision-makers.

Autonomous systems can undermine this assumption without necessarily creating a conventional agreement.

For example:

Competitor A's AI
↓
observes market prices
↓
Competitor B's AI
↓
observes the resulting prices
↓
both algorithms adapt
↓
prices converge
↓
consumers face higher prices

There may be no telephone call, meeting or explicit agreement.

This creates the possibility of algorithmic coordination without conventional human communication.

3. Principal Competition Risks

A. Algorithmic Collusion

Autonomous systems may learn that avoiding aggressive price competition maximises long-term profits.

Two independent algorithms could therefore repeatedly converge upon supra-competitive prices.

The central legal problem is whether:

autonomous learning itself can constitute a legally relevant form of coordination.

Competition authorities may need to distinguish:

• legitimate parallel conduct;
• conscious adaptation;
• algorithmic interdependence;
• tacit coordination;
• explicit algorithmic communication;
• human-directed collusion.

4. The Attribution Problem

A fundamental issue is:

Who is responsible for an autonomous system's decision?

Potentially responsible actors include:

1. the company deploying the AI;
2. the AI developer;
3. the data provider;
4. the system integrator;
5. the human manager supervising it;
6. the marketplace hosting it;
7. multiple firms using the same algorithm.

Competition law generally imposes liability on undertakings, not machines.

Therefore, an autonomous AI cannot ordinarily become the legal undertaking itself merely because it made the decision.

The legal challenge is determining when the machine's conduct can be attributed to its human or corporate operator.

5. Shared Algorithms and Common Providers

A particularly important risk arises where competitors use the same algorithmic service.

Suppose:

Retailer A → Algorithm Provider X
Retailer B → Algorithm Provider X
Retailer C → Algorithm Provider X

If Provider X recommends identical prices based upon common market data, the system could potentially become a coordination infrastructure.

The competitive concern increases where the provider:

• receives confidential information from competitors;
• recommends prices to multiple competitors;
• incorporates competitors' prices into its optimisation;
• prevents customers from deviating from recommendations;
• monitors compliance;
• automatically punishes deviations.

The algorithm provider may effectively become an intermediary for competitive coordination.

6. Autonomous Agents and Market Power

Autonomous systems may also produce dominance independent of traditional market-share calculations.

An AI platform may control:

• data;
• compute;
• models;
• APIs;
• distribution;
• cloud infrastructure;
• identity;
• payments;
• search;
• recommendation systems.

This can create a stacked form of market power.

Autonomous competitive stack

Compute
↓
Foundation model
↓
Agent
↓
Data
↓
Marketplace/API
↓
Users
↓
Transactions

Control at several layers can make entry extremely difficult.

7. Network Effects and Autonomous Learning

Autonomous systems may become stronger as they process more transactions.

More transactions
↓
more data
↓
better predictions
↓
better performance
↓
more users
↓
more transactions

This creates a self-reinforcing competitive feedback loop.

Unlike traditional network effects, the system may combine:

• direct network effects;
• data advantages;
• machine learning;
• switching costs;
• interoperability barriers;
• economies of scale;
• computational advantages.

This can produce extremely rapid concentration.

8. Autonomous Pricing Systems

Pricing is one of the most significant areas.

AI pricing systems can continuously modify prices according to:

• demand;
• inventory;
• competitor prices;
• customer characteristics;
• time;
• location;
• willingness to pay;
• weather;
• supply conditions.

Competition authorities therefore face a difficult distinction between:

Legitimate dynamic pricing

Prices respond independently to supply and demand.

Potentially anticompetitive algorithmic pricing

An algorithm deliberately uses competitor information or coordination mechanisms to sustain supra-competitive prices.

9. Personalised Autonomous Pricing

Autonomous systems can also move beyond market-wide pricing toward individualised prices.

For example:

Customer A → ₹1,000
Customer B → ₹1,400
Customer C → ₹1,800

based upon predictions of willingness to pay.

This raises competition concerns involving:

• price discrimination;
• exploitation;
• consumer lock-in;
• data advantages;
• exclusion;
• discriminatory access;
• reduced price transparency.

Competition law may increasingly have to consider algorithmic extraction of consumer surplus, rather than simply average prices.

10. Autonomous Procurement

Autonomous procurement agents can independently select suppliers.

This can produce efficiency gains through:

• lower transaction costs;
• rapid supplier comparison;
• automated negotiation;
• fraud detection;
• inventory optimisation.

But it can also create risks where a dominant buyer's AI:

• excludes smaller suppliers;
• systematically favours affiliated suppliers;
• uses discriminatory scoring;
• imposes standardised contractual conditions;
• coordinates purchasing behaviour among multiple firms.

11. Autonomous Marketplaces

An autonomous marketplace could potentially determine:

• which sellers receive visibility;
• search ranking;
• commission rates;
• advertising placement;
• consumer recommendations;
• eligibility;
• pricing suggestions;
• access to data.

The marketplace therefore becomes more than an intermediary.

It can become an autonomous market governor.

This creates potential issues involving:

• self-preferencing;
• discriminatory ranking;
• exclusion;
• tying;
• leveraging;
• foreclosure;
• preferential access;
• manipulation of market conditions.

12. Autonomous Supply Chains

AI-controlled supply chains can coordinate:

• production;
• inventory;
• transportation;
• warehousing;
• supplier allocation;
• demand forecasting.

If major competitors rely on interoperable autonomous systems, their supply-chain decisions could become increasingly similar.

The resulting concern is not merely price coordination but structural convergence of competitive behaviour.

13. Autonomous Agents and Tacit Coordination

Traditional tacit coordination often requires firms to recognise that mutual accommodation is profitable.

Autonomous agents may discover such strategies through reinforcement learning.

For example:

Agent A raises price
↓
Agent B observes higher margin
↓
B raises price
↓
A learns that retaliation is unnecessary
↓
stable high-price equilibrium develops

No human necessarily instructed the AI to collude.

This produces a conceptual distinction between:

human intention to coordinate

and

machine-generated strategic coordination.

That distinction could become one of the central competition-law issues of the AI economy.

14. Autonomous Systems and Article 101 TFEU

Under EU competition law, Article 101 addresses agreements, decisions by associations of undertakings and concerted practices.

Autonomous systems therefore raise several questions.

Possible Article 101 scenarios

Scenario 1 — Human agreement

Two companies agree to use an algorithm to maintain prices.

→ Conventional Article 101 analysis remains comparatively straightforward.

Scenario 2 — Algorithm provider

A common provider deliberately designs its system to coordinate competitors.

→ Potential concerted-practice or intermediary theories become relevant.

Scenario 3 — Independent machine learning

Two algorithms independently learn similar strategies.

→ Establishing a legally cognisable concerted practice becomes considerably more difficult.

15. Autonomous Systems and Article 102 TFEU

Article 102 may become particularly important where an autonomous system is controlled by a dominant undertaking.

Potential abuses include:

• self-preferencing;
• discriminatory access;
• refusal to interoperate;
• tying;
• exclusionary ranking;
• predatory algorithmic pricing;
• discriminatory AI recommendations;
• leveraging dominance from one AI layer into another.

The relevant question increasingly becomes:

Does the autonomous system merely implement the undertaking's commercial policy, or does it independently determine that policy?

16. UK Competition Law

Under the UK Competition Act 1998, autonomous systems can potentially raise issues under:

• Chapter I prohibition;
• Chapter II prohibition;
• market investigation powers;
• digital-markets regulation;
• merger control.

The UK approach is particularly significant because digital competition regulation increasingly focuses on firms possessing strategic market status and their conduct in digital activities.

Autonomous AI could therefore become both:

1. the object of regulatory scrutiny, and
2. the instrument through which market power is exercised.

17. United States Antitrust Law

US antitrust law similarly encounters the problem of distinguishing:

• conscious parallelism;
• algorithmic coordination;
• explicit agreement;
• information exchange;
• monopolisation.

Section 1 of the Sherman Act generally requires concerted action, while Section 2 addresses unilateral monopolisation and attempted monopolisation.

Autonomous systems may therefore generate a particularly difficult question:

Can algorithmic coordination satisfy the agreement requirement where no human expressly agreed to coordinate?

Where a company intentionally deploys an algorithm designed to facilitate coordination, traditional doctrines become easier to apply.

Purely independent machine learning remains substantially more difficult.

18. Six Important Case Laws

1. United States v. Apple Inc. — 791 F.3d 290 (2d Cir. 2015)

The Apple e-books litigation is important because it demonstrates that a company cannot avoid competition-law responsibility merely because coordination is implemented through sophisticated contractual or technological structures.

Relevance

The case illustrates the importance of examining:

• the architecture of coordination;
• intermediary mechanisms;
• commercial strategy;
• communications;
• the overall economic arrangement.

Autonomous-system significance

If firms deliberately design an AI system to implement an anticompetitive strategy, the technological intermediary should not necessarily obscure the underlying competitive conduct.

2. United States v. Topkins — 2016

Topkins involved online sellers using algorithms to coordinate prices for posters and related products.

The case is one of the clearest practical examples of algorithm-assisted price fixing.

Significance

The important principle is that:

using an algorithm to implement price coordination does not transform cartel conduct into lawful independent competition.

Autonomous-system relevance

It provides a foundation for distinguishing:

algorithm as tool of human collusion

from

algorithm as independent strategic decision-maker.

The first category is considerably easier to address under conventional antitrust law.

3. Eturas UAB v. Lietuvos Respublikos konkurencijos taryba — C-74/14

The European Court of Justice examined an electronic booking platform through which a system administrator communicated a restriction on discounts to participating travel agencies.

The case is highly relevant to technology-mediated coordination.

Significance

The case demonstrates that an electronic platform can become part of a competition-law problem where undertakings receive and act upon information capable of coordinating their commercial behaviour.

Autonomous-system relevance

An AI platform that communicates strategic pricing recommendations to competing firms could create similar questions concerning:

• knowledge;
• participation;
• acceptance;
• implementation;
• evidentiary inference.

4. T-Mobile Netherlands BV v. Netherlands Competition Authority — C-8/08

The ECJ considered the concept of concerted practice and emphasised that coordination can infringe competition law even where the parties have not reached a traditional contractual agreement.

Importance

The case is significant for understanding the distinction between:

• agreement;
• concerted practice;
• independent market behaviour.

Autonomous relevance

It becomes particularly important when AI systems repeatedly exchange or process strategically sensitive information.

The challenge is determining whether machine-mediated behaviour can provide evidence of a human undertaking's participation in coordination.

5. Wood Pulp — Ahlström Osakeyhtiö and Others v Commission — Joined Cases 89/85, 104/85, 114/85, 116/85, 117/85 and 125–129/85

The Wood Pulp litigation is an important authority concerning parallel conduct and the limits of inferring collusion from market behaviour.

Significance

The case demonstrates that parallel pricing alone does not automatically establish an unlawful concerted practice.

Other explanatory factors and evidence of coordination matter.

Autonomous relevance

This becomes crucial where independent AI systems produce similar prices.

Identical algorithmic outcomes should not automatically be treated as proof of collusion.

The regulator must investigate whether the convergence results from:

• common cost conditions;
• common data;
• market structure;
• independent optimisation;
• common algorithmic architecture;
• or deliberate coordination.

6. United States v. Airline Tariff Publishing Co. — 1994

The Airline Tariff Publishing litigation involved sophisticated electronic dissemination of fare information.

Significance

The case illustrates how highly transparent, computer-mediated markets can facilitate coordination where firms use public or disseminated information strategically.

Autonomous relevance

Modern AI systems can make such markets considerably more sophisticated because algorithms can:

• observe competitors continuously;
• process millions of observations;
• react instantly;
• predict competitor responses;
• implement price changes automatically.

The competitive risks therefore may be significantly greater than in conventional information-exchange environments.

19. Additional Relevant Case Law

7. United States v. Microsoft Corp. — 253 F.3d 34 (D.C. Cir. 2001)

Microsoft demonstrates how control over an important technological platform can be used to protect or extend market power.

Autonomous-system relevance

AI ecosystems may similarly leverage dominance from:

operating system → cloud → model → API → agent → marketplace

The case is therefore useful for analysing technological foreclosure and leveraging.

8. Google Shopping — Commission Decision AT.39740

The Google Shopping proceedings concern self-preferencing by a dominant digital platform.

Autonomous relevance

Autonomous recommendation and ranking systems can reproduce or intensify self-preferencing.

An AI marketplace could automatically favour its own products through:

• ranking;
• recommendation;
• visibility;
• pricing;
• search results.

The key competition issue becomes whether algorithmic neutrality is genuine or merely apparent.

20. Evidentiary Problems

Autonomous competition creates a major evidence problem.

Traditional investigations often rely upon:

• emails;
• meetings;
• WhatsApp messages;
• contracts;
• executive instructions.

Autonomous systems may instead produce:

• model weights;
• logs;
• API calls;
• system prompts;
• training datasets;
• reward functions;
• telemetry;
• model versions;
• automated decision records.

Competition authorities may therefore require algorithmic audit trails.

21. Explainability and Competition Enforcement

A competition authority may ask:

Why did the AI increase the price?

The company may answer:

"The model learned that this strategy maximised long-term returns."

That response creates a regulatory difficulty.

If the company cannot reconstruct the model's decision, establishing:

• intent;
• knowledge;
• causation;
• discriminatory purpose;
• coordination;
• responsibility

may become difficult.

Therefore, autonomous competition could create pressure for explainability obligations in competition-sensitive systems.

22. Human Oversight

A major regulatory solution is meaningful human oversight.

However, nominal oversight may be insufficient.

For example:

"A human could technically override the AI."

That does not necessarily mean the human actually exercises meaningful control.

Regulators may increasingly distinguish:

Formal human oversight

Human intervention is legally possible.

Effective human oversight

Human decision-makers can actually understand, monitor and override the system.

23. Competition by Autonomous Agents

A more radical possibility is an economy where firms deploy autonomous agents to compete directly.

Imagine:

Company A Agent ↔ Company B Agent

The agents negotiate:

• prices;
• contracts;
• delivery;
• quantities;
• warranties;
• dispute resolution.

Humans merely define broad objectives.

This could create an economy in which machine-to-machine negotiation becomes a principal competitive mechanism.

The law may therefore need to determine whether an agent's negotiated conduct is:

• an offer;
• acceptance;
• agreement;
• representation;
• corporate act;
• automated execution.

24. Autonomous Smart Contracts

Smart contracts add another layer.

An AI agent could:

1. negotiate terms;
2. execute a smart contract;
3. monitor performance;
4. automatically impose penalties;
5. terminate the relationship;
6. select another counterparty.

Competition concerns may arise where smart contracts:

• restrict switching;
• automatically exclude competitors;
• enforce resale restrictions;
• prevent interoperability;
• facilitate coordinated pricing.

The absence of a human intervention point makes ex post correction more difficult.

25. Autonomous Competition and Merger Control

Autonomous systems can also affect merger analysis.

A merger may combine:

• datasets;
• AI models;
• cloud infrastructure;
• autonomous agents;
• distribution channels;
• compute resources.

The competitive harm may not immediately appear through traditional market-share analysis.

Instead, the merger could increase:

autonomous learning capacity and strategic decision-making power.

This suggests that future merger assessment may need to consider:

• access to compute;
• quality and quantity of data;
• model capabilities;
• agent distribution;
• interoperability;
• switching costs;
• ecosystem control.

26. Algorithmic Entry Barriers

Autonomous systems can increase barriers to entry through:

Data advantage

Incumbents possess more behavioural data.

Learning advantage

Their models improve faster.

Compute advantage

Large firms can train and operate larger models.

Distribution advantage

Existing platforms control user access.

Integration advantage

AI agents can be integrated across multiple services.

Feedback advantage

More users produce more data, improving the system.

This can produce a dynamic entry barrier rather than a traditional static barrier.

27. Autonomous Systems and Essential Facilities

Where a small number of firms control indispensable AI infrastructure, competition concerns may arise around access to:

• cloud compute;
• GPUs;
• foundation models;
• APIs;
• datasets;
• identity infrastructure;
• autonomous-agent marketplaces.

The traditional essential-facilities doctrine may therefore acquire new relevance in AI markets.

28. Interoperability

Interoperability becomes particularly important.

A dominant AI ecosystem might prevent autonomous agents from communicating effectively with rival systems.

This can create:

closed ecosystem → switching costs → reduced interoperability → weaker rivals → greater dominance

Competition authorities may therefore consider interoperability remedies where technical exclusion materially restricts competition.

29. Autonomous Systems and Consumer Welfare

The consumer harm may not always appear as a simple price increase.

Potential harms include:

• reduced choice;
• degraded quality;
• surveillance;
• privacy loss;
• reduced innovation;
• discriminatory treatment;
• exclusion of suppliers;
• reduced interoperability;
• manipulation of consumer attention.

Therefore, competition analysis may increasingly have to examine multi-dimensional competitive welfare.

30. Global Regulatory Fragmentation

Autonomous competitive systems operate globally, but competition regimes remain nationally organised.

One AI system may simultaneously affect:

• EU markets;
• US markets;
• UK markets;
• Indian markets;
• Chinese markets;
• ASEAN markets.

Different jurisdictions may reach different conclusions concerning:

• market definition;
• algorithmic collusion;
• dominance;
• data access;
• interoperability;
• remedies;
• AI transparency.

This creates a major problem of regulatory fragmentation.

31. Extraterritorial Enforcement

An autonomous algorithm located in one jurisdiction can automatically affect consumers elsewhere.

For example:

AI infrastructure — United States
↓
platform — EU
↓
agent provider — UK
↓
supplier — India
↓
consumer — Asia

Determining jurisdiction and applicable competition law becomes increasingly complex.

32. Appropriate Regulatory Framework

A future regulatory framework should probably combine several mechanisms.

1. Algorithmic transparency

Require firms to document competition-sensitive algorithms.

2. Auditability

Maintain sufficient logs to reconstruct significant decisions.

3. Competition-by-design

Competition considerations should be integrated into system architecture.

4. Human accountability

Corporate responsibility should remain even where decisions are automated.

5. Interoperability

Prevent dominant systems from unnecessarily blocking rival agents.

6. Data-access remedies

Where data is an essential competitive input, appropriate access remedies may be considered.

7. Algorithmic monitoring

Competition authorities should possess technical capabilities to monitor autonomous systems.

8. Rapid intervention

Traditional litigation can take years, while AI markets can change within months.

33. Core Legal Test for Autonomous Competitive Conduct

A useful analytical framework is:

Step 1 — Identify the undertaking

Who controls or economically benefits from the system?

Step 2 — Identify the autonomous function

What does the system independently decide?

Step 3 — Identify the market

What product, geographic and technological market is affected?

Step 4 — Establish market power

Does the undertaking possess substantial market power?

Step 5 — Examine coordination

Does the system communicate or process competitors' strategic information?

Step 6 — Examine intent/design

Was the system deliberately designed to coordinate or exclude?

Step 7 — Examine effects

Did the system produce:

• higher prices;
• foreclosure;
• reduced output;
• reduced innovation;
• discriminatory access;
• consumer harm?

Step 8 — Attribution

Can the autonomous decision legally be attributed to the undertaking?

Step 9 — Remedy

Should the regulator impose:

• behavioural remedies;
• interoperability;
• data access;
• algorithmic monitoring;
• structural remedies;
• penalties?

34. Future Concept: Autonomous Market Power

Traditional market power is generally associated with the ability to behave independently of competitive constraints.

Autonomous systems could create a new form:

Autonomous market power = the ability of an undertaking's computational system to continuously predict, influence and respond to market participants faster and more effectively than rivals.

This power could derive from:

data + compute + algorithms + network effects + distribution + autonomous execution.

That combination may become one of the most important characteristics of future digital markets.

35. Overall Assessment

The transition to autonomous competitive systems does not necessarily require an entirely new competition law.

Many traditional doctrines remain applicable.

The more immediate challenge is adapting existing concepts to machine-mediated conduct.

The critical distinctions will be:

Conclusion

The global transition to autonomous competitive systems represents a structural transformation of competition rather than merely another application of AI.

The central issue is not simply whether algorithms can make decisions. It is whether economically consequential decisions can increasingly be made, coordinated and executed without direct human intervention.

Competition law must therefore confront three fundamental questions:

1. Attribution — Who is legally responsible for autonomous conduct?
2. Coordination — When does machine interaction become unlawful coordination?
3. Market power — How should competition law assess control over autonomous decision-making infrastructure?

The most important lesson from cases such as Topkins, Eturas, T-Mobile Netherlands, Wood Pulp, Airline Tariff Publishing, Apple and Microsoft is that technology does not automatically remove competition-law responsibility. At the same time, independent machine learning creates genuinely novel questions where traditional concepts of agreement, intention and conscious coordination may not map neatly onto machine behaviour.