1. Introduction

Model-to-model interoperability refers to the ability of different artificial intelligence (AI) models, agents, applications, and computing infrastructures to communicate, exchange information, transfer tasks, and operate together without unnecessary technical, contractual, or commercial barriers.

For example, a business may use one AI model for document analysis, another for translation, a third for fraud detection, and a fourth for decision support. Genuine interoperability enables these models to work together through compatible application programming interfaces (APIs), common data formats, agreed communication protocols, and portable workflows.

Model-to-model interoperability constraints arise when a model provider or ecosystem controller prevents, limits, degrades, or makes prohibitively expensive the interaction between its models and competing models.

Such constraints may include:

Restricting access to proprietary APIs, model endpoints, or agent communication protocols.

Preventing users from transferring prompts, context, memory, or workflow state to rival models.

Imposing contractual restrictions on connecting competing models.

Designing systems that make switching between models costly or technically difficult.

Withholding essential compatibility information or imposing discriminatory access conditions.

Using cloud infrastructure, identity systems, proprietary tools, or model registries to privilege affiliated models.

These restrictions are not automatically unlawful. Some may be justified by cybersecurity, privacy, intellectual property, reliability, or legitimate safety requirements. The central legal question is whether a restriction is objectively justified or instead protects market power by foreclosing competing AI providers.

2. Legal framework

A. European Union competition law

Article 101 TFEU prohibits agreements and concerted practices that restrict competition, where the legal requirements are satisfied. Contractual provisions preventing independent AI models from interoperating may be relevant if they restrict competition by object or effect.

Article 102 TFEU prohibits the abuse of a dominant position. Depending on the facts, potentially abusive practices include:

Refusing access to an indispensable interface or facility.

Discriminating between affiliated and independent model providers.

Tying access to a dominant platform to the use of its own AI model.

Degrading interoperability with rival models.

Using technical restrictions to exclude competitors from adjacent markets.

A refusal to provide access is not automatically unlawful. The applicable legal test depends on the market, the nature of the resource, the effects of the refusal, and the relevant case law.

B. Digital Markets Act

The EU Digital Markets Act (DMA) imposes specific obligations on designated gatekeepers. Its provisions concerning interoperability, access, data portability, and restrictions on business users can be relevant to AI-enabled ecosystems.

Article 6(7), for example, addresses interoperability with certain hardware and software features controlled by designated gatekeepers. Article 6(9) concerns end-user data portability, while Article 6(10) addresses access to certain data by business users. Their application depends on the designated service, the obligation's precise scope, and applicable implementation requirements.

The DMA does not establish a universal right to access every proprietary AI model or its weights. Its relevance must be assessed against the gatekeeper designation and the particular service concerned.

C. German competition law

Section 19 of the German Act Against Restraints of Competition (GWB) prohibits specified abuses of dominance. Section 19a gives the Bundeskartellamt powers concerning undertakings of paramount significance for competition across markets and specified abusive conduct.

These provisions may be relevant where a large technology company controls cloud infrastructure, a dominant platform, a model distribution channel, or technical interfaces that competing AI providers need to reach customers.

Section 19a does not apply automatically to every large AI company; the statutory requirements and applicable determination must be satisfied.

D. United Kingdom competition law

Under Chapter II of the Competition Act 1998, conduct that amounts to an abuse of a dominant position may be prohibited. Relevant theories include exclusionary refusal to deal, discriminatory access, tying, and conduct that raises rivals' costs.

The Digital Markets, Competition and Consumers Act 2024 also creates a framework for strategic market status (SMS) firms and conduct requirements in designated digital activities. Its application depends on the firm's designation and the scope of the relevant requirements.

E. United States antitrust law

Sections 1 and 2 of the Sherman Act address, respectively, agreements restraining trade and monopolization or attempted monopolization. Interoperability restrictions may be examined where they form part of exclusionary conduct, unlawful agreements, or a strategy to preserve monopoly power.

US law generally does not impose a free-standing duty on every company to make its products interoperable with competitors. The legality of a refusal depends on the circumstances and the applicable doctrine.

3. Major legal issues in model-to-model interoperability

3.1 Refusal to provide access to model interfaces

A provider may prevent rival models from accessing an API, tool interface, orchestration layer, or communication protocol that is necessary to compete effectively.

The legal analysis asks:

Does the provider hold a dominant position in a properly defined market?

Is the interface genuinely important for effective competition?

Does the refusal eliminate or substantially restrict effective competition?

Are there objective justifications, such as security, privacy, or technical limitations?

Would a proportionate access remedy preserve competition without compromising legitimate interests?

The European essential-facilities and intellectual-property case law establishes demanding conditions for compulsory access. The mere fact that interoperability would benefit a competitor is insufficient.

3.2 Technical incompatibility and deliberate degradation

An AI provider might permit rival models to connect nominally but give them inferior functionality. Examples include reduced context transfer, delayed access to tools, incomplete memory migration, restricted function calls, or less reliable routing.

A technical difference is not necessarily anticompetitive. Regulators should examine whether the difference is objectively explained by security, architecture, or performance considerations, or whether it selectively disadvantages rival models.

Relevant evidence includes comparative latency, error rates, feature availability, access conditions, and internal technical documentation.

3.3 Data, context, and memory portability

AI systems can derive competitive advantages from accumulated user context, feedback, conversation history, workflow configurations, and integrations with enterprise data.

Switching becomes more difficult if users cannot export those materials in usable formats. The issue is especially significant when the provider controls both the original model and the environment in which the user's context is stored.

Data portability, however, is not equivalent to transferring proprietary model weights, confidential information, or every internal representation. Legal duties depend on the applicable statutory framework, contractual rights, privacy rules, trade secrets, and the type of data involved.

3.4 Contractual and ecosystem restrictions

Restrictions may prohibit customers from using third-party models, connecting competing agents, or deploying alternative models on the same infrastructure.

These arrangements require a competition assessment that considers market power, duration, scope, exclusivity, foreclosure, efficiencies, and less restrictive alternatives.

A security certification or compatibility requirement may be legitimate. A requirement that effectively forces customers to purchase an affiliated model, without sufficient justification, may raise tying or exclusionary-conduct concerns.

3.5 Standards and protocol control

If one provider controls a widely adopted AI communication standard, it may influence which models can communicate efficiently and which remain outside the ecosystem.

Risks include discriminatory certification, refusal to license necessary specifications, strategically delayed compatibility updates, and changes that disadvantage independent implementations.

Standardisation can also increase competition by reducing integration costs. The appropriate response is therefore not to prohibit proprietary standards generally, but to assess exclusion, governance, access conditions, and the availability of realistic alternatives.

4. At least 10 important case laws

The following cases provide the principal legal foundations for analysing AI interoperability constraints. They are established competition-law precedents, not judgments specifically deciding the legality of modern model-to-model interoperability. Their relevance to AI depends on the facts and the applicable legal test.

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

Jurisdiction: United States

Legal principle: Exclusionary conduct involving operating-system dominance and restrictions on rival technologies.

The US proceedings concerned Microsoft's conduct in protecting its Windows operating-system monopoly, including measures affecting browser competition and the distribution of competing technologies.

The case illustrates how control over a foundational software platform can be used to disadvantage complementary products and rival distribution channels.

Application to AI: A dominant AI operating environment could potentially restrict third-party models through privileged system integration, default settings, API access, or technical limitations that protect an affiliated model.

Limitation: The judgment does not establish that every platform owner must make its technology interoperable with every competitor. The analysis concerns the particular exclusionary conduct and market circumstances.

Case 2: European Commission v. Microsoft (2004)

Jurisdiction: European Union

Legal principle: Access to interoperability information and the use of platform dominance to protect adjacent markets.

The European Commission found that Microsoft had abused its dominant position, including through its refusal to supply interoperability information needed by competing work-group server operating systems. The decision imposed remedies concerning disclosure and other conduct.

Application to AI: The case is especially relevant where a dominant AI platform controls interface specifications, orchestration protocols, or technical information that rival providers need to interoperate effectively.

A comparable case would require proof that the conduct meets the relevant abuse-of-dominance test; not every proprietary API or undocumented interface triggers a duty to disclose.

Case 3: Microsoft Corp. v. Commission (Case T-201/04, 2007)

Jurisdiction: European Union

Legal principle: Conditions governing compulsory interoperability disclosure by a dominant undertaking.

The General Court largely upheld the Commission's 2004 decision, including its findings concerning interoperability information and the remedies imposed.

The judgment recognised the importance of preserving competition in adjacent markets while addressing the legitimate interests of the intellectual-property holder.

Application to AI: If a provider controls an interface necessary for effective competition between enterprise AI models, the decision offers a framework for examining whether access obligations could be justified and how they might be designed.

The decision does not establish an unconditional right to model weights, training data, or proprietary reasoning mechanisms.

Case 4: Commercial Solvents Corp. v. Commission (1974)

Jurisdiction: European Union

Legal principle: Refusal to supply an input used by a customer competing in a downstream market.

The Court of Justice upheld findings of abuse where a dominant supplier withdrew supplies of a necessary raw material from a customer operating in a downstream market, in circumstances that threatened to eliminate that customer's competition.

Application to AI: An analogous issue may arise where a dominant infrastructure provider withdraws access to a crucial model-serving capability, compute interface, or integration service from an independent AI provider that competes downstream.

The analogy is strongest where the input is genuinely important and the refusal threatens competition rather than merely disadvantaging one rival.

Case 5: Oscar Bronner GmbH & Co. KG v. Mediaprint (Case C-7/97, 1998)

Jurisdiction: European Union

Legal principle: Strict conditions for compelling access to infrastructure controlled by a dominant undertaking.

The Court set a demanding standard for requiring a dominant undertaking to grant access to a facility it had developed for its own business. In the circumstances, the refusal to grant access to Mediaprint's newspaper home-delivery system was not abusive under the essential-facilities doctrine.

Application to AI: A claimant seeking access to a proprietary model interface must do more than show that access would make its product cheaper or more convenient. Questions of indispensability, elimination of competition, and the feasibility of alternatives are central where the Bronner framework applies.

This is a critical counterweight to arguments for universal AI interoperability.

Case 6: IMS Health GmbH & Co. OHG v. NDC Health GmbH & Co. KG (Case C-418/01, 2004)

Jurisdiction: European Union

Legal principle: Exceptional compulsory licensing of intellectual property.

The Court identified conditions under which refusal to license intellectual property by a dominant undertaking may constitute abuse, including circumstances involving a new product for which there is potential consumer demand, the absence of objective justification, and the elimination of competition in a secondary market.

Application to AI: The case may be relevant if a provider uses intellectual-property rights over an interface, format, or protocol to prevent the development of competing AI services.

The precise conditions depend on the circumstances; interoperability is not itself sufficient to compel licensing.

Case 7: Magill (Joined Cases C-241/91 P and C-242/91 P, 1995)

Jurisdiction: European Union

Legal principle: Exceptional circumstances in which refusal to license intellectual property can amount to abuse.

The case concerned television broadcasters refusing to license programme listings that a third party needed to produce a comprehensive weekly television guide. The Court upheld the finding of abuse in the circumstances.

Application to AI: If one company controls information or an interface that prevents the creation of a genuinely distinct competing AI service, Magill may provide a relevant analogy.

It does not mean that all proprietary model outputs, datasets, or interfaces must be made available to competing firms.

Case 8: Huawei Technologies Co. Ltd v. ZTE Corp. (Case C-170/13, 2015)

Jurisdiction: European Union

Legal principle: Competition-law limits on seeking injunctions based on standard-essential patents subject to FRAND commitments.

The Court set out a framework of reciprocal steps for negotiations involving a standard-essential patent holder and an implementer, including notice of infringement and a willingness to negotiate on fair, reasonable and non-discriminatory terms.

Application to AI: Where interoperability depends on standard-essential technologies, patent licensing and FRAND commitments may affect access conditions. The case can inform analysis of standards-based AI ecosystems, although it does not itself impose FRAND obligations on all AI interfaces or protocols.

Case 9: Slovak Telekom a.s. v. Commission (Case C-165/19 P, 2021)

Jurisdiction: European Union

Legal principle: Abuse of dominance through restrictions on access to infrastructure, and the relationship between sector-specific access obligations and Article 102 TFEU.

The litigation concerned conduct by a dominant telecommunications operator relating to access to its local-loop infrastructure. The Court clarified aspects of the relationship between general competition law and sector-specific regulation.

Application to AI: The case is relevant when assessing whether access to AI infrastructure is already governed by a specific regulatory framework and whether additional competition-law duties may arise. Its reasoning must be applied in light of the differences between telecommunications infrastructure and AI services.

Case 10: Google Android (European Commission decision, 2018; General Court, Case T-604/18, 2022)

Jurisdiction: European Union

Legal principle: Restrictions involving mobile-platform distribution, tying, and contractual arrangements affecting rival services.

The Commission's Android decision concerned Google's conduct relating to Android device manufacturers and mobile application distribution. The General Court upheld the central findings while annulling the finding concerning one element of the exclusivity-payment arrangements and adjusting the fine.

Application to AI: A similar theory may arise where an AI ecosystem controller conditions access to a dominant operating system, app marketplace, or distribution channel on using its own model, or imposes restrictions that foreclose rival AI providers.

The case does not make bundling or integration inherently unlawful. Market power, contractual structure, foreclosure, and justification remain important.

5. Additional case law and its relevance

Case 11: Google Shopping (Case T-612/17, 2021)

Jurisdiction: European Union

Legal principle: Self-preferencing by a dominant platform may constitute abuse where it departs from competition on the merits and produces exclusionary effects.

The General Court upheld the Commission's finding that Google had favoured its own comparison-shopping service in its general search results, disadvantaging competing services.

Application to AI: A model marketplace could favour its owner's model in rankings, routing, tool recommendations, or default selection while placing competing models at a disadvantage. An enforcement case would need to establish the relevant market power, conduct, and exclusionary effects.

Case 12: United States v. Google LLC (US District Court for the District of Columbia, 2024, search-distribution judgment)

Jurisdiction: United States

Legal principle: Exclusive distribution agreements can help preserve monopoly power where they unlawfully restrict access to important distribution channels.

The court found that Google's distribution agreements concerning search violated Section 2 of the Sherman Act.

Application to AI: Exclusive arrangements between model providers, cloud platforms, device manufacturers, and enterprise software distributors may warrant scrutiny if they foreclose meaningful opportunities for rival models to reach customers.

The decision concerns search distribution, not a general prohibition on AI exclusivity. The precise agreement and its competitive effects matter.

6. Comparative legal principles

These decisions should be used as doctrinal analogies, not treated as direct holdings on AI-to-AI interoperability.

7. How regulators should assess interoperability restrictions

A sound investigation should distinguish legitimate engineering choices from exclusionary strategies.

Competition assessment framework

Define the relevant market. Examine foundation models, inference services, orchestration tools, cloud infrastructure, model marketplaces, or enterprise AI applications, as appropriate.

Identify control and dependency. Determine whether the undertaking controls an important interface, protocol, data resource, or distribution channel and whether credible alternatives exist.

Characterise the restriction. Distinguish a refusal to supply from degraded access, discriminatory terms, technical incompatibility, tying, or contractual exclusivity.

Assess competitive effects. Examine switching costs, foreclosure, entry barriers, rival expansion, customer choice, innovation, and effects on prices and quality.

Test objective justifications. Evaluate security, privacy, reliability, intellectual-property protection, and genuine technical limitations.

Choose a proportionate remedy. Consider interface access, common protocols, portability, non-discrimination, independent testing, or narrowly defined licensing obligations.

Evidence that may be relevant

Regulators and courts may examine:

API documentation and changes to interface specifications.

Comparative latency, reliability, feature access, and model-routing outcomes.

Internal communications about rival models and interoperability decisions.

Contractual restrictions, exclusivity clauses, and switching costs.

Technical evidence concerning security risks and alternative implementations.

Customer testimony, integration costs, and evidence of lost market opportunities.

Whether affiliated models receive preferential treatment under equivalent conditions.

The analysis should not infer anticompetitive intent solely from the existence of proprietary technology or a technical incompatibility.

8. Remedies and regulatory responses

A. Technical interoperability

Authorities may consider requiring access to specified interfaces or communication protocols where the relevant legal test is met. Remedies should identify the functions required for interoperability rather than demand indiscriminate disclosure of source code, model weights, or confidential training data.

B. Data and context portability

Where applicable law provides the right, portability can reduce switching costs by allowing users to transfer eligible data and configurations in usable formats. For AI services, implementation may require structured export of conversation histories, user-managed memory, workflow configurations, and tool settings.

Portability obligations must respect privacy, security, intellectual property, and third-party rights.

C. Non-discrimination

A dominant ecosystem may be required, where legally justified, to apply equivalent technical and commercial conditions to affiliated and independent models. Relevant measures could include consistent API access, transparent eligibility rules, and independent compliance audits.

D. Open standards and certification

Industry standards can improve compatibility and reduce integration costs. Governance should guard against standards capture, discriminatory certification, and unnecessary technical requirements that exclude smaller providers.

E. Structural and behavioural remedies

In serious cases, regulators may consider behavioural remedies or, where authorised and proportionate, structural measures. A remedy should target the source of the competitive harm and avoid undermining security, innovation, or legitimate product differentiation.

9. Hypothetical example

Suppose a dominant enterprise AI platform offers document analysis, coding assistance, and workflow automation. Independent models can technically connect to the platform, but the platform restricts their access to long-context processing and selected tools. Its own model receives full functionality, and enterprise customers face substantial costs when moving their workflows elsewhere.

A competition authority would examine whether:

The platform has dominance in a relevant market.

The restrictions materially disadvantage rival models.

The restrictions are necessary for security or reflect genuine technical differences.

Customers have realistic alternatives.

The conduct forecloses competition or protects the platform's position in an adjacent market.

Existing legal obligations, including any applicable DMA or national competition rules, govern the conduct.

If the evidence establishes unlawful exclusionary conduct, possible remedies could include equivalent access to specified interfaces, transparent technical criteria, and monitoring of compliance. If the restrictions are objectively necessary and proportionate, competition law may not require their removal.

10. Critical analysis

Three competing considerations shape the law.

First, interoperability promotes competition. It can reduce switching costs, enable multi-model deployment, support smaller innovators, and prevent a single provider from controlling an entire AI workflow.

Second, forced interoperability can create risks. Poorly designed access requirements may expose confidential information, increase cybersecurity vulnerabilities, or impose disproportionate engineering burdens. Interoperability can also create new dependencies if one common protocol becomes another gatekeeper's control point.

Third, legal intervention must remain evidence-based. Competition law generally does not guarantee every competitor access to every privately developed technology. Compulsory access is most defensible where the applicable legal standard is met and the remedy is proportionate to the demonstrated harm.

A particularly important distinction is between technical interoperability and competitive neutrality. Two models can communicate successfully while still competing under unequal conditions because of pricing, ranking, access to customers, compute allocation, or contractual restrictions. Conversely, two models may remain technically separate without creating an antitrust problem if customers have effective alternatives.

11. Conclusion

Model-to-model interoperability constraints are an emerging competition-law concern because control over interfaces, context transfer, orchestration, distribution, and infrastructure can allow an AI ecosystem to extend its power across adjacent markets.

The most relevant established authorities include Microsoft, Bronner, IMS Health, Magill, Commercial Solvents, Huawei v. ZTE, Slovak Telekom, Google Android, and Google Shopping. Collectively, they help distinguish legitimate proprietary control from potentially unlawful exclusion, while emphasising that the applicable legal tests are fact-sensitive.