Competition Law and Personal AI Ecosystem Dominance

1. Introduction

Personal AI refers to AI systems that operate as individualized assistants for a particular user and may continuously interact with the user's:

communications;

calendar;

files;

browser;

search services;

shopping accounts;

financial applications;

health and fitness applications;

smart-home devices;

operating system;

workplace software;

cloud storage;

other AI models; and

autonomous AI agents.

A personal AI ecosystem therefore goes beyond a conventional chatbot. The AI may become the principal interface through which a person searches, communicates, purchases products, manages information and instructs other digital services.

This creates a distinctive competition-law problem:

The relevant source of market power may no longer be the individual AI model, but the ecosystem surrounding the user's personal AI.

For example, a company could combine:

Operating System → App Store → Search → Browser → Cloud → AI Model → Personal Assistant → Payments → Consumer Data → AI Agents

and thereby create substantial barriers for rival personal-AI providers.

The issue is particularly contemporary. In July 2026, the European Commission adopted measures requiring Google to provide competing AI services with effective interoperability with key Android functionality, including features relevant to AI assistants. The Commission specifically identified the importance of Android as a distribution channel for AI assistants. (Digital Markets Act (DMA))

2. Meaning of Personal AI Ecosystem Dominance

Traditional AI competition might ask:

Does Company A have a dominant position in the market for foundation models?

Personal-AI competition requires a broader question:

Does Company A control the ecosystem through which users access, operate and integrate AI services?

An ecosystem can become dominant through control over several complementary assets.

Typical ecosystem

Device

↓

Operating system

↓

App store

↓

Browser/search

↓

Personal AI assistant

↓

Personal data

↓

Third-party applications

↓

AI agents

↓

Payments and transactions

The competitive concern becomes particularly strong where one undertaking controls multiple layers and can use power in one layer to reinforce another.

3. Relevant Competition-Law Theories

A. Abuse of Dominance

A dominant personal-AI ecosystem may potentially engage in:

tying;

bundling;

self-preferencing;

discriminatory access;

refusal to interoperate;

exclusionary contracts;

foreclosure of rival AI assistants;

leveraging;

data exploitation;

predatory pricing;

margin squeeze; and

discriminatory ranking.

Under Article 102 TFEU, Section 2 of the Sherman Act, Section 4 of the Indian Competition Act, and analogous provisions in other jurisdictions, the precise legal test depends upon the jurisdiction.

4. Tying Personal AI to an Operating System

Suppose an operating-system provider makes its AI assistant the default assistant and prevents competing assistants from accessing important operating-system functions.

The company may effectively tie:

OS dominance → AI-assistant advantage.

The competitive significance increases if the operating system provides its own AI assistant with privileged access to:

notifications;

contacts;

messages;

camera;

microphone;

files;

location;

applications;

device controls; and

cross-application actions.

This is precisely why interoperability has become an important contemporary issue. The European Commission's 2026 Android proceedings concern equal access for competing AI services to Android features used by Google's own AI services. (Digital Markets Act (DMA))

5. Six Major Case Laws and Their Application

Because personal AI ecosystem dominance is an emerging field, there are not yet six major reported judicial decisions specifically deciding dominance by a "personal AI ecosystem." The appropriate legal method is therefore to apply established antitrust precedents concerning operating systems, tying, digital platforms, search, interoperability, essential inputs and ecosystem foreclosure.

Case 1: Google Android — Google Android, Commission Decision AT.40099

The European Commission found Google dominant in several relevant markets, including licensing of smart mobile operating systems, Android app stores and general search services.

The Commission considered Google's contractual arrangements with manufacturers and concluded that certain practices involving pre-installation and defaults could reinforce Google's position in search. (European Commission)

Application to Personal AI

This is perhaps the most directly relevant precedent.

A personal AI provider controlling an operating system could potentially:

make its AI assistant the default;

restrict rival assistants;

require pre-installation;

limit third-party access to device functions;

give its AI privileged APIs; or

use app-store rules to disadvantage rival AI services.

The important concept is leveraging power from an upstream ecosystem layer into an adjacent AI market.

Case 2: Microsoft v Commission — Windows Media Player

The Microsoft Windows Media Player case concerned Microsoft's tying of Windows Media Player to the Windows operating system.

The General Court largely upheld the Commission's approach concerning the tying practice and Microsoft's conduct in relation to interoperability information.

Application to Personal AI

The analogy is strong.

Replace:

Windows + Media Player

with:

Operating System + Personal AI Assistant.

If an operating-system provider makes its AI assistant inseparable from the operating system while restricting rival assistants' ability to provide equivalent functionality, the competition-law questions can include:

whether the products are distinct;

whether there is coercion;

whether rivals are foreclosed;

whether interoperability is impaired;

whether the conduct produces consumer harm; and

whether objective justifications exist.

The case demonstrates why technical integration can have competition consequences even where the integrated product appears convenient to consumers.

Case 3: United States v Microsoft Corp. — D.C. Circuit

The Microsoft litigation concerned Microsoft's conduct in the operating-system and browser markets.

The court's analysis addressed Microsoft's use of Windows distribution power to protect its position against competing browser technologies.

Application to Personal AI

The lesson is the importance of distribution control.

A personal AI provider does not necessarily need to possess the best underlying AI model if it controls the principal distribution channel.

For example:

70% model quality + universal distribution

could potentially be more commercially powerful than:

95% model quality + limited distribution.

A dominant OS provider could theoretically disadvantage rivals through:

default settings;

installation restrictions;

API restrictions;

warnings;

interoperability limitations;

technical degradation; or

preferential system integration.

The Microsoft case therefore provides an important framework for analysing AI distribution dominance.

Case 4: Intel v Commission, Case C-413/14 P

In Intel, the Court of Justice clarified the treatment of exclusivity rebates by dominant undertakings and emphasised that, where the undertaking submits evidence that its conduct is not capable of restricting competition, the authority must consider the circumstances relevant to assessing exclusionary effects.

Application to Personal AI

Personal-AI ecosystems may use incentives to lock in:

device manufacturers;

application developers;

cloud providers;

enterprise customers;

AI developers; and

users.

For example, a dominant ecosystem might offer developers:

lower platform fees;

preferential AI access;

cloud credits;

revenue-sharing benefits; or

preferential ranking

in exchange for making its AI the exclusive or primary assistant.

The Intel framework is relevant because the competition analysis cannot simply stop at identifying an exclusivity arrangement; its actual or potential exclusionary effects and relevant circumstances matter.

Case 5: Bronner v Mediaprint, Case C-7/97

The CJEU established a stringent framework for refusal-to-deal claims involving access to an allegedly indispensable facility.

The case concerned access to a newspaper distribution system.

Application to Personal AI

Personal AI creates potential "digital essential input" questions.

Consider a dominant operating-system provider that controls access to:

device APIs;

voice activation;

system notifications;

application actions;

personal-device context;

authentication;

secure enclave functionality; or

other essential technical capabilities.

A rival AI provider might argue that it cannot compete effectively without access to those capabilities.

The Bronner doctrine reminds us that not every refusal to provide access constitutes abuse.

The relevant questions include:

Is the input indispensable?

Is effective competition impossible without access?

Is duplication practically or economically impossible?

Is access objectively justified?

Would compulsory access interfere with incentives to innovate?

These questions will become increasingly important as personal AI becomes deeply integrated into operating systems.

Case 6: European Commission v Google — Google Shopping, Case C-48/22 P

In Google Shopping, the CJEU upheld the central finding that Google's conduct concerning comparison-shopping services could constitute an abuse of dominance where its general-search position was used to advantage Google's own comparison-shopping service.

Application to Personal AI

The principle is relevant to AI self-preferencing.

Imagine a personal AI assistant that recommends:

hotels;

restaurants;

financial products;

shopping products;

travel services;

software;

news;

apps; and

other AI agents.

If the AI operator also owns businesses providing those services, the AI could potentially become a distribution gatekeeper.

For example:

User asks AI: "Find me the cheapest laptop."

The AI could potentially rank:

its affiliated marketplace;

its affiliated retailer;

its affiliated payment service;

above competing services.

Competition law would then examine whether the AI interface is being used to foreclose competing providers.

Case 7: Google Search (AdSense), Case AT.40411

The Commission's Google AdSense case concerned restrictions imposed through agreements involving Google's online-search advertising intermediation.

The case is relevant to the broader concept of using dominance in one digital layer to restrict competition in another.

Application to Personal AI

Personal AI may become an intermediary between consumers and businesses.

Instead of:

Consumer → Search Engine → Website

the future model may become:

Consumer → Personal AI → Merchant/Service Provider.

The AI therefore controls the consumer's access point.

That could create competition concerns if the AI systematically:

excludes rivals;

imposes discriminatory conditions;

limits alternative providers;

favours affiliated services; or

uses contractual restrictions to prevent multihoming.

6. The Most Important Competition Problem: AI as a Gatekeeper

The traditional digital economy had several major gateways:

search engine;

operating system;

browser;

app store;

social network;

marketplace.

Personal AI could combine many of them.

A user's AI could become the single interface for:

Search + Communication + Shopping + Finance + Travel + Work + Applications + Information + Agents.

This produces what may be called interface concentration.

The competitive concern is that the AI may become the intermediary through which consumers access competing businesses.

7. Personal AI and Self-Preferencing

Self-preferencing can become particularly powerful in AI because users often do not see the complete ranking process.

Suppose an AI receives:

"Find me an accounting application."

The AI could recommend:

its own application;

an affiliated application; or

an independent competitor.

Unlike traditional search results, AI-generated recommendations may not present dozens of alternatives.

The AI may instead provide one or a few recommendations.

Therefore, preferential treatment could have an amplified competitive effect.

8. AI Agent Marketplaces

Personal AI may eventually operate as an agent marketplace.

A user might say:

"Book my flight."

The AI could independently:

search airlines;

compare prices;

select flights;

access the user's payment method;

make the reservation; and

update the calendar.

This means the AI is not simply supplying information.

It is executing transactions.

The AI therefore becomes a gateway between users and numerous downstream markets.

This creates potential competition concerns concerning:

agent ranking;

access fees;

commissions;

preferred agents;

exclusivity;

transaction routing;

payment integration;

self-preferencing; and

discriminatory API access.

9. Data Advantages

Personal AI ecosystems can create an unusual competitive advantage through personal-context data.

An AI could potentially learn:

what a person purchases;

how the person writes;

whom the person communicates with;

what the person searches for;

what applications the person uses;

what documents the person reads;

what services the person prefers; and

how the person makes decisions.

The resulting dataset can improve the AI's:

personalization;

prediction;

recommendation;

automation; and

agent performance.

This can create a feedback loop:

More users → more data → better AI → better personalization → more users.

This is a classic data/network-effect feedback mechanism.

10. Data Portability and Switching Costs

Suppose a user has spent five years building a personal AI profile.

It contains:

preferences;

memories;

contacts;

instructions;

workflows;

application permissions;

purchasing history;

personal knowledge;

agent configurations.

Switching to another AI may therefore be costly.

This creates personal-AI switching costs.

Competition authorities may need to distinguish:

Legitimate switching costs

Costs arising naturally from building a personalised service.

Artificial switching costs

Costs deliberately created by the provider through:

incompatible formats;

restrictive APIs;

refusal to export data;

contractual restrictions;

technical barriers; or

degradation of competing services.

11. Interoperability

Interoperability may become one of the most important competition-law remedies.

A user might want:

Personal AI A + Email B + Calendar C + Shopping D + Banking E.

A dominant AI ecosystem could instead attempt to create:

Personal AI A + its own email + its own calendar + its own marketplace + its own payments.

This is the difference between:

open ecosystem competition

and

closed ecosystem competition.

The European Commission's 2026 Android interoperability measures are highly relevant to this emerging problem. They require effective interoperability between third-party AI services and specified Android functionalities controlled by Google. (Digital Markets Act (DMA))

12. The DMA and Personal AI

The Digital Markets Act is particularly significant because it can address certain gatekeeper practices without requiring the same traditional Article 102 dominance analysis.

In January 2026, the Commission began proceedings concerning Google's obligations to provide third-party developers effective interoperability with Android features used by Google's AI services. (Digital Markets Act (DMA))

In July 2026, the Commission adopted binding specification measures relating to that interoperability. The Commission stated that competing AI assistants should be able to access relevant Android functionality on an effective basis. (Digital Markets Act (DMA))

This illustrates a major regulatory shift:

Competition law traditionally asks whether conduct by a dominant undertaking is abusive; digital regulation increasingly establishes ex ante interoperability obligations for designated gatekeepers.

13. Cloud-AI Integration

Personal AI also depends heavily upon cloud infrastructure.

A vertically integrated company might control:

Cloud → AI model → AI application → Personal assistant.

This creates potential foreclosure risks.

For example, a cloud provider could potentially:

offer preferential computing resources to its own AI;

impose switching costs;

restrict portability;

bundle cloud services with AI;

provide preferential access to chips;

restrict competing AI developers; or

obtain commercially sensitive information.

The FTC's study of Microsoft-OpenAI, Amazon-Anthropic and Google-Anthropic partnerships identified potential competition issues including access to computing resources, switching costs, exclusivity/control provisions and access to sensitive technical or business information. (Federal Trade Commission)

Importantly, those findings concern potential competition implications, not a final determination that each partnership violates antitrust law.

14. Vertical Integration

Personal AI ecosystems are particularly susceptible to vertical integration.

A single undertaking could control:

The more layers controlled by one company, the greater the potential for ecosystem leveraging.

However, vertical integration is not inherently unlawful. It may generate substantial efficiencies.

15. Network Effects

Personal AI markets may display several network effects.

Direct network effects

More users can attract more developers.

Indirect network effects

More developers create more integrations, making the AI more useful.

Data network effects

More usage produces more data that may improve the service.

Ecosystem network effects

More connected applications increase the value of the personal AI.

Agent network effects

More AI agents make the personal AI more useful, attracting more users and therefore more agents.

These feedback loops can create rapid concentration.

16. Killer Acquisitions

A dominant technology company may acquire:

promising AI startups;

personal-assistant applications;

agent platforms;

AI memory companies;

interoperability providers;

AI-search startups;

personal-data management platforms.

Even where the target has little current revenue, it could represent a significant future competitive threat.

Therefore, merger authorities may need to consider:

innovation competition;

potential competition;

access to AI talent;

data assets;

proprietary technology;

user communities;

interoperability;

future AI-agent markets.

17. Exclusive Dealing

A dominant personal AI provider could potentially negotiate exclusive arrangements with:

smartphone manufacturers;

browser companies;

application developers;

retailers;

banks;

airlines;

cloud providers;

telecommunications companies.

For example:

A manufacturer agrees to make AI X the exclusive assistant for three years.

If AI X already possesses substantial market power, such an arrangement could potentially make market entry significantly more difficult for rival assistants.

The Intel case provides an important framework for examining exclusionary effects of exclusivity-related conduct.

18. Algorithmic Discrimination

AI recommendation systems can potentially discriminate among competing suppliers.

For example:

Company A — affiliated: receives 90% exposure.

Company B — independent: receives 10% exposure.

The user may never know why.

Possible competition concerns include:

biased ranking;

preferential recommendations;

discriminatory API access;

exclusion from agent marketplaces;

preferential transaction routing.

The analytical difficulty is that the discriminatory mechanism may be embedded within a machine-learning model rather than a conventional contractual rule.

19. Competition Between AI Models

Personal AI competition should also consider the underlying model market.

A dominant personal-AI ecosystem might use:

proprietary models;

third-party models;

open-source models;

specialised models.

If the ecosystem gives its proprietary model privileged access to:

user context;

hardware acceleration;

system APIs;

data;

application actions;

then technically superior rival models may still be unable to compete effectively.

This creates an important distinction:

Model competition is not necessarily the same as ecosystem competition.

20. Open-Source AI and Competition

Open-source AI can potentially reduce concentration by providing alternative models.

But a dominant personal-AI ecosystem may nevertheless control the interface through which users interact with those models.

Thus:

Open model ≠ open ecosystem.

A company may permit users to run third-party models while still controlling:

distribution;

APIs;

user data;

device integration;

agent permissions;

ranking;

payment;

transaction execution.

21. Indian Competition-Law Application

Under the Competition Act, 2002, the principal provisions are:

Section 3

Relevant to agreements producing or likely to cause an appreciable adverse effect on competition.

Possible AI examples:

exclusive AI distribution agreements;

cartelisation among AI providers;

coordinated pricing;

market allocation;

agreements restricting interoperability.

Section 4

Potentially relevant where a dominant enterprise abuses its position.

Potential conduct includes:

discriminatory access;

tying;

bundling;

refusal to deal;

self-preferencing;

unfair conditions;

exclusionary contracts.

Sections 5 and 6

Relevant to combinations involving:

AI startups;

cloud providers;

device manufacturers;

search businesses;

AI-agent platforms;

data businesses.

The Indian analysis would need careful market definition because "personal AI" may encompass several potentially distinct markets.

22. Possible Relevant Markets

Authorities might consider separate markets for:

AI foundation models;

consumer AI assistants;

personal AI assistants;

AI-powered search;

AI agent marketplaces;

mobile operating systems;

cloud computing;

AI application distribution;

AI-enabled digital advertising;

AI transaction-intermediation services.

Alternatively, some cases may require an ecosystem-based theory of harm involving several related markets.

The precise market definition will depend on substitution, user behaviour, geographic scope, technology and competitive constraints.

23. Consumer Welfare

Personal AI creates a complicated consumer-welfare analysis.

Possible benefits include:

lower search costs;

personalised recommendations;

automation;

greater convenience;

reduced transaction costs;

improved accessibility;

better productivity;

easier access to digital services.

Potential harms include:

reduced choice;

lock-in;

discriminatory recommendations;

higher prices;

lower innovation;

exclusion of rival developers;

reduced interoperability;

exploitation of data.

Competition authorities therefore need to evaluate both innovation efficiencies and foreclosure risks.

24. Remedies

If competition concerns are established, possible remedies could include:

Interoperability

Require competing AI services to access relevant operating-system capabilities.

Data portability

Allow users to transfer personal AI data and configurations.

API access

Require fair and non-discriminatory access to important technical interfaces.

Non-discrimination

Prevent preferential treatment of affiliated AI services.

Choice screens

Allow users to select among AI assistants.

Default flexibility

Permit users to change the default AI assistant.

Contractual restrictions

Limit exclusivity arrangements where they foreclose competition.

Structural remedies

In exceptional circumstances, separation of businesses may be considered under applicable law.

25. Six Core Competition-Law Questions

For any personal-AI ecosystem, the following questions provide a useful analytical framework:

1. Who controls the user interface?

Does the user interact directly with competing services or through one dominant AI?

2. Who controls the data?

Can users transfer their personal AI context elsewhere?

3. Who controls interoperability?

Can rival AI systems interact with the same applications and devices?

4. Who controls distribution?

Can rival assistants reach users on equivalent terms?

5. Who controls recommendations?

Does the AI objectively rank competitors or favour affiliated services?

6. Who controls transactions?

Does the AI become the gatekeeper for purchases, bookings, payments and other commercial activity?

26. Consolidated Case-Law Table

27. Overall Legal Analysis

The distinctive competition-law problem of personal AI is that market power can migrate from the model to the ecosystem.

A company does not necessarily need the most powerful AI model to dominate personal AI.

It may instead control:

the device + operating system + data + distribution + applications + search + cloud + payment system + AI interface.

That combination can create substantial barriers to entry even where individual AI models remain technically contestable.

The emerging European regulatory approach illustrates this point particularly clearly. The Commission's 2026 Android interoperability measures specifically address the ability of competing AI assistants to obtain access to important Android functionality, while its search-data measures contemplate access for eligible search providers including AI chatbots offering search functionality. (Digital Markets Act (DMA))

28. Conclusion

Personal AI ecosystem dominance represents a new form of digital-platform power in which the AI assistant can become the primary gateway between an individual user and the wider digital economy.

The principal antitrust concerns are:

operating-system leveraging;

AI default-setting;

tying and bundling;

self-preferencing;

API and interoperability restrictions;

personal-data accumulation;

AI-agent marketplace control;

exclusive distribution;

cloud/model vertical integration;

algorithmic discrimination;

consumer lock-in;

killer acquisitions;

foreclosure of competing AI models; and

control over AI-mediated transactions.

The most useful established authorities are therefore Google Android, Microsoft, United States v Microsoft, Intel, Bronner, Google Shopping and Google AdSense. They do not establish that personal AI ecosystems are inherently anticompetitive; rather, they supply established doctrines—tying, leveraging, exclusivity, refusal to deal, interoperability, self-preferencing and distribution foreclosure—that can be applied to the emerging personal-AI environment.

The central competition-law principle is:

The decisive question may no longer be who has the best AI model, but who controls the ecosystem through which users access, personalise and act through AI.