Competition Law and Future Institutional Design for AGI Societies

Introduction

Artificial General Intelligence (AGI) refers to highly capable AI systems capable of performing a broad range of cognitive tasks across domains rather than being confined to a single application. If AGI becomes commercially deployable, competition law may face a structural transformation. Market power could arise not merely from ownership of factories, patents, or conventional data assets, but from control over compute, foundation models, proprietary data, model interfaces, AI agents, cloud infrastructure, autonomous decision systems, and interconnected digital ecosystems.

The central institutional question will therefore be:

How should competition authorities be designed when economically significant firms may control intelligent systems that themselves can learn, coordinate, negotiate, allocate resources, and potentially operate across multiple markets?

Existing competition institutions—such as the Competition Commission of India (CCI), European Commission, U.S. Department of Justice (DOJ), Federal Trade Commission (FTC), and UK's Competition and Markets Authority (CMA)—provide useful foundations. However, AGI societies may require institutions with substantially greater technical expertise, continuous monitoring capacity, interoperability powers, algorithmic-audit capabilities, merger foresight, and coordination mechanisms.

I. Meaning of Institutional Design in AGI Competition

Institutional design concerns the structure through which competition law is administered, enforced, monitored, and adapted.

For AGI markets, institutional design must address five interconnected layers:

1. Infrastructure layer – chips, data centres, electricity and cloud computing.
2. Model layer – foundation models and AGI systems.
3. Interface layer – APIs, operating systems, app stores and agent platforms.
4. Application layer – healthcare, finance, transport, education, commerce and other sectors.
5. Agentic layer – autonomous AI systems capable of making commercial decisions.

Traditional antitrust intervention frequently begins after harmful conduct becomes observable. AGI may require institutions capable of detecting risks before market structures become irreversible.

II. Why AGI Creates New Competition Problems

1. Concentration of Compute

Training frontier AI systems requires enormous computational resources.

If only a limited number of firms control:

• advanced processors;
• hyperscale cloud infrastructure;
• specialised data centres;
• energy capacity;
• networking infrastructure; and
• model-training clusters,

competition at the model layer may become dependent upon infrastructure access.

This creates a potential vertical competition problem.

2. Control of Foundation Models

A dominant AGI provider could become an indispensable upstream supplier for thousands of downstream businesses.

For example:

Cloud → Foundation Model → API → AI Agent → Application → Consumer

Control at one level may therefore affect competition throughout the entire chain.

3. Data Advantages

AGI systems may generate enormous quantities of behavioural and commercial information.

A platform controlling both:

• the AI system; and
• the data generated through its use

could potentially obtain a self-reinforcing advantage.

4. AI Agents as Competitors

An autonomous agent could:

• compare prices;
• negotiate contracts;
• purchase products;
• switch suppliers;
• allocate advertising;
• optimise inventory; and
• negotiate with other agents.

This creates a new competition question:

Who is responsible when autonomous systems engage in potentially anticompetitive conduct?

5. Algorithmic Coordination

Traditional cartels generally involve human or corporate decision-makers.

AGI societies could involve autonomous systems interacting continuously.

Potential problems include:

• algorithmic price coordination;
• autonomous market allocation;
• personalised exclusion;
• discriminatory access;
• coordinated bidding;
• automated refusal to deal.

Competition institutions will therefore need to distinguish legitimate independent optimisation from unlawful coordination.

III. Future Institutional Architecture

A future AGI competition authority could be organised around specialised divisions.

A. AI Competition Division

This division would investigate:

• AI platform dominance;
• model access;
• API restrictions;
• algorithmic discrimination;
• AI-agent coordination;
• self-preferencing; and
• exclusionary model practices.

B. Compute and Infrastructure Division

It could supervise competition involving:

• GPUs;
• AI accelerators;
• cloud services;
• data centres;
• electricity access;
• specialised networking; and
• AI compute marketplaces.

C. Algorithmic Audit Division

Competition authorities would need technical teams capable of examining:

• source code where legally accessible;
• model behaviour;
• training processes;
• pricing algorithms;
• reinforcement systems;
• recommendation systems;
• agent-to-agent communications; and
• logs.

The institutional objective would not necessarily be to regulate AI generally, but to determine whether technological architecture is being used to restrict competition.

D. Continuous Market Monitoring Division

Traditional competition investigations are often event-driven.

AGI markets may require continuous monitoring of:

• market concentration;
• switching rates;
• API access;
• model dependency;
• compute allocation;
• acquisition activity;
• interoperability;
• prices; and
• algorithmic behaviour.

IV. Ex-Ante and Ex-Post Competition Regulation

A major institutional change could be the combination of ex-ante and ex-post enforcement.

Ex-post model

The authority intervenes after conduct occurs.

Examples:

• abuse of dominance;
• cartel;
• exclusionary agreement;
• anticompetitive merger.

Ex-ante model

The authority establishes obligations before competitive harm occurs.

Possible AGI obligations include:

• interoperability;
• data portability;
• API access;
• non-discrimination;
• transparency;
• restrictions on self-preferencing;
• merger notification;
• access to essential infrastructure.

The Digital Markets Act provides an important institutional example of this broader ex-ante approach.

V. Future AGI Merger Control

AGI could substantially alter merger analysis.

A conventional turnover-based threshold may fail to capture acquisitions of:

• promising AI start-ups;
• research teams;
• specialised datasets;
• model developers;
• AI-agent companies;
• safety technologies; or
• emerging competitors with limited current revenue.

Consequently, future institutional design may employ:

1. Transaction-value thresholds

A very high acquisition price may indicate competitive significance even when the target has little turnover.

2. Innovation-based assessment

Authorities could examine whether an acquisition eliminates a potential future competitor.

3. Capability concentration

Authorities could assess whether one firm is acquiring strategically important:

• models;
• compute;
• data;
• talent;
• agents;
• infrastructure.

4. Portfolio effects

Several individually small acquisitions could collectively create substantial ecosystem control.

VI. AGI and the Essential-Facilities Concept

Traditional essential-facilities doctrine may become relevant to:

• frontier compute;
• specialised AI chips;
• critical model APIs;
• data infrastructure;
• cloud interfaces;
• agent marketplaces.

A competition authority may face questions such as:

When does control over a particular AI infrastructure become sufficiently indispensable to justify access obligations?

The institutional difficulty is determining whether mandated access would encourage competition or reduce incentives for innovation and investment.

VII. Interoperability as a Competition Remedy

Interoperability could become one of the most important remedies in AGI markets.

Authorities could require dominant systems to permit:

• model portability;
• API interoperability;
• agent portability;
• data portability;
• identity portability;
• secure switching;
• communication between competing AI systems.

This would reduce ecosystem lock-in.

However, interoperability rules would have to incorporate:

• cybersecurity;
• privacy;
• intellectual property;
• safety;
• confidentiality; and
• model integrity.

VIII. Institutional Independence and Technical Expertise

AGI competition authorities would require institutional independence comparable to traditional competition agencies but with substantially enhanced technical capability.

A future authority could include:

The authority would therefore become a multidisciplinary economic-technical institution.

IX. Six Important Case Laws and Their AGI Relevance

The following cases do not concern AGI directly in most instances. Their importance lies in the competition principles they establish that can inform future AGI regulation.

1. United States v. Microsoft Corp. (2001)

The U.S. courts examined Microsoft's conduct concerning the Windows operating-system ecosystem and its relationships with competing technologies.

Importance

The case demonstrated how dominance in one technological layer can be used to protect or extend power into adjacent markets.

AGI relevance

An AGI platform could potentially use dominance in:

Operating System → AI Assistant → Agent Marketplace → Applications

to disadvantage competitors.

Future competition authorities may therefore need to examine ecosystem leveraging, not merely individual product markets.

2. Google Search (Shopping) – European Commission, 2017

The European Commission found that Google had given prominent placement to its comparison-shopping service while disadvantaging competing comparison-shopping services.

Competition principle

The case is associated with concerns regarding self-preferencing by a dominant platform.

AGI relevance

A dominant AI platform could potentially favour:

• its own applications;
• its own agents;
• its own shopping services;
• its own advertising products; or
• its own data services

when users rely upon the platform's AI-generated recommendations.

Future institutional design may therefore require mechanisms to detect AI-mediated self-preferencing.

3. Google Android – European Commission, 2018

The European Commission examined Google's practices involving Android, including restrictions relating to applications and search services.

Competition principle

The case illustrates how contractual arrangements surrounding a dominant platform can affect adjacent markets.

AGI relevance

Comparable concerns could arise if an AGI platform requires developers to use:

• its payment system;
• its model;
• its cloud service;
• its identity system; or
• its agent marketplace.

This supports specialised scrutiny of AI ecosystem tying and bundling.

X. 4. United Brands v Commission (1978)

The European Court of Justice developed important principles concerning abuse of dominance and the special responsibility of dominant firms.

Competition principle

A dominant undertaking may have a special responsibility not to allow its conduct to impair genuine undistorted competition.

AGI relevance

A firm controlling a highly significant AGI infrastructure could potentially possess enormous influence over downstream markets.

Future institutions may therefore need to apply special scrutiny to:

• discriminatory API access;
• refusal to interoperate;
• exclusionary contractual arrangements;
• discriminatory model performance; and
• preferential treatment of affiliated services.

XI. 5. Bronner v Mediaprint (1998)

The Court of Justice examined the conditions relevant to refusal-to-deal and essential-facilities-type arguments.

Competition principle

The case illustrates the high threshold traditionally associated with compelling a dominant undertaking to provide access to infrastructure.

AGI relevance

Suppose a particular AI infrastructure becomes indispensable to competing firms.

The institutional question becomes:

Should the dominant AI infrastructure provider be required to provide access?

Authorities would need to balance:

• indispensability;
• elimination of competition;
• feasibility of duplication;
• innovation incentives;
• security;
• investment incentives.

XII. 6. Intel v Commission (CJEU, 2017)

The Intel litigation concerned rebates and the assessment of exclusionary conduct by a dominant undertaking.

Competition principle

The judgment contributed significantly to the modern analysis of potentially exclusionary rebates and economic effects.

AGI relevance

An AGI platform might offer:

• discounted model access;
• cloud credits;
• preferential API pricing;
• bundled compute;
• exclusive developer incentives.

Competition authorities may need sophisticated economic analysis to determine whether such arrangements merely compete aggressively or instead exclude rivals.

XIII. 7. Amazon Marketplace – European Commission

The European Commission investigated Amazon's use of marketplace seller data and concerns relating to competition between Amazon's retail business and independent sellers.

AGI relevance

An AGI marketplace could simultaneously:

1. operate the platform;
2. observe competing businesses;
3. collect transaction data;
4. provide AI services to those businesses; and
5. compete against them.

This creates a potential dual-role conflict.

Future authorities may need rules preventing dominant AI platforms from using competitively sensitive information obtained through their intermediary role to disadvantage dependent businesses.

XIV. AGI and Algorithmic Collusion

One of the most difficult institutional problems will be autonomous coordination.

Suppose several AI agents continuously observe market prices and independently modify their conduct.

There may be no explicit human agreement.

The authority would need to determine:

• whether there was communication;
• whether algorithms were designed to coordinate;
• whether firms intentionally deployed coordination mechanisms;
• whether the outcome was merely parallel conduct;
• whether human responsibility can be established.

This could require new evidentiary standards.

XV. AI Audit Trails as Competition Evidence

Future competition authorities may require powerful AI systems to preserve:

• decision logs;
• model versions;
• pricing instructions;
• agent communications;
• API calls;
• training modifications;
• deployment records;
• human intervention records.

This could become the AGI equivalent of conventional documentary evidence.

An AI Competition Evidence Code could establish:

1. retention periods;
2. audit procedures;
3. authentication requirements;
4. chain-of-custody rules;
5. confidentiality protections;
6. regulator access procedures.

XVI. Cross-Border Institutional Cooperation

AGI markets will likely be inherently international.

A single model may be:

• trained in one country;
• hosted in another;
• supplied globally;
• accessed through multiple platforms; and
• integrated into applications worldwide.

Competition authorities would therefore require mechanisms for:

• information sharing;
• coordinated investigations;
• merger review;
• dawn raids;
• algorithmic evidence;
• remedies;
• cross-border enforcement.

International cooperation between the European Commission, FTC, DOJ, CMA, CCI and other authorities could become increasingly important.

XVII. Competition Authority as a Continuous Market Institution

A future AGI competition authority may evolve from a largely reactive regulator into a continuous market-governance institution.

Its functions could include:

Monitoring

Continuous observation of critical markets.

Prediction

Identification of structural risks before they become irreversible.

Investigation

Traditional enforcement against unlawful conduct.

Technical auditing

Examination of AI systems.

Structural remedies

Interoperability, separation, divestiture or access remedies where legally justified.

Merger supervision

Monitoring acquisitions of emerging competitors and strategic assets.

Regulatory experimentation

Controlled testing of new competition remedies.

XVIII. AGI Regulatory Sandboxes

Competition authorities could establish AI competition sandboxes.

Companies could test:

• interoperability;
• agent marketplaces;
• data-sharing arrangements;
• pricing algorithms;
• switching mechanisms.

The authority could evaluate whether proposed designs create competition risks before large-scale deployment.

This could reduce the need for enforcement after market structures have already become entrenched.

XIX. Institutional Separation of AI Safety and Competition

AI safety and competition regulation overlap but should not necessarily become one regulatory function.

For example:

AI Safety Authority

focuses on:

• model safety;
• catastrophic risks;
• cybersecurity;
• misuse;
• reliability.

Competition Authority

focuses on:

• market power;
• exclusion;
• collusion;
• mergers;
• access;
• interoperability.

Coordination between the two would be essential, but their statutory objectives could remain distinct.

XX. Future Remedies

Possible AGI competition remedies include:

Behavioural remedies

• non-discrimination;
• transparent access;
• restrictions on tying;
• fair API terms.

Structural remedies

• divestiture;
• separation of infrastructure and applications;
• ownership restrictions.

Interoperability remedies

• open interfaces;
• portability;
• switching mechanisms.

Data remedies

• data portability;
• controlled data access;
• restrictions on use of competitively sensitive information.

Merger remedies

• prohibition;
• divestiture;
• licensing;
• access commitments.

XXI. Central Institutional Challenge: Innovation Versus Competition

AGI competition policy must account for an important tension.

Over-regulation may:

• discourage investment;
• increase compliance costs;
• reduce innovation;
• disadvantage smaller firms;
• slow technological development.

Under-regulation may:

• entrench dominant firms;
• eliminate emerging competitors;
• increase switching costs;
• concentrate critical infrastructure;
• reduce innovation incentives.

Institutional design must therefore distinguish between large scale resulting from successful innovation and market power maintained through exclusionary conduct.

XXII. Proposed Future Model

A conceptual institutional structure could be:

                    AGI COMPETITION AUTHORITY                              │          ┌───────────────────┼───────────────────┐          │                   │                   │   Market Intelligence   AI/Algorithmic       Merger &       Division              Audit             Acquisition          │                   │                   │          ├───────────────┬───┴────┬──────────────┤          │               │        │              │       Compute         Foundation  Agents       Platforms    Infrastructure      Models                   │          │               │        │              │          └───────────────┴────────┴──────────────┘                              │                    Competition Remedies                              │          ┌──────────────┬────┴─────┬──────────────┐          │              │          │              │   Interoperability   Portability  Access      Structural                                             Remedies

XXIII. Key Legal Principles for Future AGI Competition Law

Future institutional design can be built around the following principles:

1. Technological neutrality – competition law should regulate competitive effects rather than particular technologies.
2. Continuous monitoring – critical AI markets may require ongoing supervision.
3. Ex-ante intervention – some systemic risks may require preventive obligations.
4. Evidence-based enforcement – AI behaviour must be supported by auditable evidence.
5. Interoperability – switching and multi-homing should remain practically possible.
6. Data neutrality – dominant intermediaries should not unfairly exploit competitively sensitive data.
7. Merger foresight – potential competition should be considered.
8. Technical expertise – authorities require permanent AI capabilities.
9. Institutional independence – enforcement should remain insulated from commercial and political pressure.
10. International cooperation – global AI markets require cross-border enforcement mechanisms.

Conclusion

The future of competition law in AGI societies will depend not only upon what substantive rules prohibit, but upon whether competition institutions possess the technical, economic and investigative capabilities necessary to understand intelligent markets.

The traditional competition authority was designed primarily to investigate firms. The AGI-era authority may increasingly have to investigate firms + algorithms + infrastructure + data + autonomous agents + ecosystems.

The cases of Microsoft, Google Shopping, Google Android, United Brands, Bronner, Intel and Amazon demonstrate that many foundational competition-law doctrines already contain principles relevant to AGI: leveraging of dominance, self-preferencing, tying, refusal to deal, exclusionary conduct, access to infrastructure and exploitation of platform information.