Global Research Governance, AI Systems and Innovation Steering

1. Introduction

Global research governance of AI systems and innovation steering refers to the legal, institutional, economic and competition-law mechanisms through which governments, universities, research institutions, technology companies, funding agencies and international organisations influence the direction, accessibility, ownership and deployment of AI research and innovation.

The issue is no longer limited to regulating finished AI products. Increasingly, governance intervenes much earlier:

• allocation of research funding;
• access to advanced computing infrastructure;
• control over datasets and research repositories;
• university–industry research partnerships;
• intellectual-property ownership;
• publication and disclosure rules;
• AI safety and evaluation requirements;
• export controls on advanced chips;
• government procurement;
• licensing of foundation models;
• access to public research infrastructure;
• concentration of researchers and compute resources; and
• strategic direction of national AI innovation.

The central legal question is therefore:

When does legitimate coordination of AI research become excessive control over the direction of technological innovation, or create competition, market-access, equality and intellectual-property concerns?

This question cuts across competition law, administrative law, public procurement, intellectual property, constitutional law, research freedom, data governance, national-security regulation and international economic law.

2. Meaning of Research Governance in AI

AI research governance can be understood as a system having five layers.

A. Input governance

Controls who receives:

• public research grants;
• datasets;
• GPUs and supercomputing resources;
• research infrastructure;
• cloud credits;
• government contracts.

B. Process governance

Controls how research is conducted through:

• ethics review;
• safety assessment;
• institutional review boards;
• cybersecurity requirements;
• data-protection obligations;
• model evaluations;
• documentation requirements.

C. Output governance

Controls:

• publication;
• patents;
• model releases;
• licensing;
• commercialisation;
• open-source distribution;
• transfer of technology.

D. Market governance

Controls whether research can translate into competitive markets.

Examples include:

• access to compute;
• interoperability;
• licensing;
• merger control;
• exclusive research partnerships;
• platform access;
• cloud access.

E. Strategic governance

Governments increasingly identify priority technologies such as:

• foundation models;
• AI chips;
• quantum-AI;
• autonomous systems;
• biotechnology AI;
• defence AI;
• robotics;
• scientific discovery systems.

This is innovation steering.

3. Innovation Steering

Innovation steering occurs when public or private institutions deliberately influence which technologies are developed, by whom, and for what purposes.

It can be legitimate.

For example, governments may encourage:

• medical AI;
• climate modelling;
• energy optimisation;
• public-sector AI;
• cybersecurity;
• scientific discovery.

But steering can also produce competitive distortions.

Example

Suppose a government provides enormous compute subsidies only to three incumbent technology companies.

Those companies then obtain:

1. cheaper compute;
2. better models;
3. more researchers;
4. more data;
5. more investment;
6. stronger commercial applications.

The subsidy therefore becomes more than a research policy.

It may become a mechanism for creating durable market power.

4. Competition-Law Dimension

Competition law becomes relevant where research governance affects competitive conditions.

Important theories include:

4.1 Research-input foreclosure

A dominant firm controls an essential research input, such as:

• specialised datasets;
• AI accelerators;
• cloud infrastructure;
• model evaluation infrastructure.

Competitors cannot innovate effectively without access.

4.2 Research collaboration

Universities and firms may collaborate on AI research.

Collaboration becomes problematic where it facilitates:

• exchange of competitively sensitive information;
• allocation of research areas;
• exclusion of rival firms;
• coordination of future commercial conduct.

4.3 Acquisition of innovation

A large AI company may acquire:

• a promising startup;
• a university spinout;
• an AI research team;
• a specialised model developer.

Even when current turnover is small, the acquisition can eliminate a future innovation competitor.

4.4 Labour-market effects

AI research governance can also influence:

• recruitment;
• non-compete arrangements;
• researcher mobility;
• salary competition;
• access to specialised talent.

Concentration of AI researchers may become a source of market power.

5. Six Major Case Laws

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

Principle

The Microsoft litigation remains important for understanding how technological ecosystems can be used to protect an incumbent's position and restrict future competitive innovation.

Microsoft's conduct concerning operating systems, browsers and distribution demonstrated how control over one technological layer can influence adjacent markets.

Relevance to AI research governance

The analogy is significant for AI ecosystems.

A company controlling:

• cloud infrastructure;
• operating systems;
• developer tools;
• model APIs;
• AI assistants

may have the ability to steer innovation toward its own ecosystem.

Legal lesson

Competition authorities should not examine AI innovation solely at the level of today's product prices.

They should consider:

whether control of an infrastructure layer allows an incumbent to determine the direction and accessibility of future innovation.

6. Case 2: United States v. Google LLC — Search / Distribution Litigation

The Google litigation concerning search distribution is relevant to AI because it illustrates the importance of default access and distribution channels.

Google's position in search was examined in relation to agreements affecting default placement and access to users.

AI relevance

AI systems increasingly depend upon:

• default assistants;
• browser integration;
• mobile-device placement;
• search interfaces;
• cloud platforms;
• app stores.

If an AI provider controls a critical distribution channel, it may steer users and developers toward its own AI system.

Research-governance implication

Innovation steering can occur indirectly.

A firm does not necessarily need to prohibit rival research.

It may instead control:

compute → developer access → distribution → user data → model improvement.

That creates a reinforcing innovation loop.

7. Case 3: European Commission v. Google Shopping (Google and Alphabet)

The Google Shopping litigation established important principles concerning exclusionary conduct by a dominant digital platform.

The European courts examined Google's preferential treatment of its own comparison-shopping service within its search results.

AI relevance

The principle can extend conceptually to AI ecosystems.

Suppose a dominant platform operates:

• an AI model;
• an AI marketplace;
• an AI research repository;
• an AI developer platform.

If it systematically favours its own AI services, it may disadvantage independent innovators.

Research implication

Innovation steering may therefore occur through algorithmic visibility.

The question becomes:

Who determines which research, models, tools or applications receive computational or algorithmic visibility?

This creates a connection between algorithmic governance and innovation competition.

8. Case 4: IMS Health GmbH & Co. OHG v. NDC Health GmbH

The IMS Health case is particularly important for the relationship between intellectual property and access to indispensable information infrastructure.

The European Court of Justice developed stringent conditions for compulsory access to protected infrastructure or information.

AI relevance

AI innovation increasingly depends upon proprietary:

• datasets;
• model weights;
• APIs;
• scientific databases;
• training resources;
• evaluation systems.

A dominant owner may argue that access would interfere with intellectual-property rights.

Competition law must balance:

innovation incentives

against

exclusionary control over indispensable innovation inputs.

Legal lesson

Compulsory access should not automatically be imposed merely because an input is useful.

The exceptional circumstances doctrine requires careful analysis.

9. Case 5: Bronner v. Mediaprint

The Bronner judgment is foundational for the European essential-facilities doctrine.

The Court adopted a restrictive approach toward compelling a dominant firm to provide access to infrastructure.

AI application

Consider a dominant AI-cloud provider controlling scarce computational infrastructure.

A competitor might argue:

"Without access to this compute infrastructure, I cannot compete."

But that alone does not automatically establish an Article 102 violation.

Authorities must examine factors such as:

• indispensability;
• duplication possibilities;
• technical feasibility;
• economic feasibility;
• elimination of competition;
• justification for refusal.

Innovation-governance significance

The case prevents competition law from becoming a general obligation requiring every successful AI infrastructure provider to share everything with competitors.

At the same time, genuine bottlenecks can justify intervention in exceptional circumstances.

10. Case 6: Microsoft Corp. v. Commission of the European Communities (2007)

The European Microsoft case is highly relevant to technological interoperability.

The Commission found problems concerning Microsoft's refusal to provide interoperability information to competitors.

AI relevance

Interoperability is becoming central to AI research.

Examples include:

• model interoperability;
• API interoperability;
• agent-to-agent communication;
• data portability;
• cloud portability;
• model evaluation portability;
• identity systems.

A dominant AI ecosystem could potentially make rival innovation difficult by preventing meaningful interoperability.

Legal lesson

Innovation governance should therefore ask:

Does technical architecture itself become a competitive exclusion mechanism?

11. Case 7: FTC v. Meta Platforms — Innovation Competition

The Meta litigation concerning acquisitions of Instagram and WhatsApp illustrates the modern concern that acquisitions can eliminate nascent or potential competitors.

The importance for AI is considerable.

AI startups may initially have:

• little revenue;
• few users;
• substantial research potential;
• unique researchers;
• valuable datasets;
• promising architectures.

Traditional turnover-based merger screening can therefore underestimate their competitive importance.

Innovation-steering problem

An incumbent may acquire a startup not because of its current market share but because:

its future research trajectory threatens the incumbent's technological position.

This creates a central AI merger-control issue: competition for innovation rather than competition over existing sales.

12. Case 8: Dow/Dupont

The Dow/DuPont merger litigation and European Commission analysis are important for the concept of innovation competition.

Competition authorities considered whether the transaction could reduce incentives or capacity to innovate.

AI relevance

The same framework can apply to AI:

Two firms may have limited overlap in current products while competing intensely in:

• next-generation models;
• autonomous agents;
• scientific AI;
• AI chips;
• model safety;
• robotics.

Therefore:

Current product-market overlap is not always sufficient to measure competitive harm in AI.

13. Research Funding as a Competition Issue

Government funding can influence market structure.

Potential benefits

Public funding can:

• correct underinvestment;
• support fundamental research;
• create public goods;
• accelerate scientific discovery;
• reduce technological dependence.

Potential risks

Poorly designed funding can:

• favour incumbents;
• exclude smaller firms;
• create discriminatory access;
• subsidise downstream commercial activities;
• produce dependency on particular vendors.

A particularly important issue is compute subsidies.

If only large companies receive access to subsidised AI infrastructure, public research policy may unintentionally strengthen concentration.

14. Public Research Infrastructure

AI research increasingly requires infrastructure that resembles a utility.

Examples:

• supercomputers;
• national AI compute centres;
• public datasets;
• scientific repositories;
• testing laboratories;
• model evaluation facilities.

Governance should therefore address:

Access

Who can use the infrastructure?

Pricing

Is access free, subsidised or commercially priced?

Priority

Who gets computational priority?

Transparency

How are applications evaluated?

Independence

Can infrastructure operators favour particular companies?

Interoperability

Can researchers transfer workloads elsewhere?

15. University–Big Tech Partnerships

University research partnerships are another major governance issue.

A technology company may provide:

• GPUs;
• cloud resources;
• grants;
• researchers;
• datasets.

In return, it may seek:

• intellectual-property rights;
• exclusive licensing;
• early access;
• confidentiality;
• commercialisation rights.

This creates an important tension.

Public-interest research

versus

private appropriation of publicly supported knowledge.

If public money substantially supports the research, policymakers may require:

• non-exclusive licensing;
• publication rights;
• open research requirements;
• access for smaller innovators;
• transparency of funding;
• conflict-of-interest disclosure.

16. AI Research and Intellectual Property

AI innovation creates multiple layers of IP.

Layer 1 — Algorithms

Patents and trade secrets may protect technical innovations.

Layer 2 — Models

Model architecture, weights and implementation may receive different forms of protection.

Layer 3 — Training datasets

Copyright, database rights, contractual rights and confidentiality may apply.

Layer 4 — Outputs

Questions arise concerning ownership and infringement.

Layer 5 — Research infrastructure

Cloud environments, APIs and evaluation platforms can themselves be proprietary.

This creates a potential IP stacking problem.

One company may control multiple layers simultaneously.

17. Open Science Versus Commercialisation

AI governance must balance two competing objectives.

Open research model

Promotes:

• reproducibility;
• peer review;
• scientific collaboration;
• broad innovation;
• educational access.

Closed model

Can provide:

• investment incentives;
• security controls;
• protection of trade secrets;
• controlled deployment;
• safety management.

Neither extreme is automatically optimal.

A sensible regulatory framework may use graduated openness.

For example:

18. Researcher Concentration

Innovation competition also depends upon human capital.

Suppose a small number of companies employ most leading AI researchers.

This can create:

• talent concentration;
• wage-setting power;
• reduced researcher mobility;
• intellectual-network concentration;
• dependency on a few laboratories.

Competition law can therefore intersect with labour-market regulation.

Authorities may examine:

• wage-fixing;
• no-poach agreements;
• restrictive employment clauses;
• coordinated hiring restrictions.

19. AI Safety as Innovation Steering

Safety regulation can improve society, but poorly designed regulation can unintentionally favour incumbents.

For example, suppose a regulator requires:

• expensive model testing;
• extensive documentation;
• costly certification;
• continuous monitoring;
• specialised audit infrastructure.

A large company may easily absorb these costs.

A startup may not.

The result can be:

Safety regulation → higher fixed costs → startup exit → greater concentration.

Therefore, AI regulation should consider regulatory proportionality.

20. Export Controls and Global AI Research

Export controls over advanced chips and AI technologies increasingly influence global innovation.

They may affect:

• availability of GPUs;
• research collaboration;
• cloud computing;
• semiconductor development;
• international scientific cooperation.

From a competition perspective, restrictions may have two contradictory effects.

Positive

They may protect national security and prevent strategic technologies from reaching prohibited users.

Negative

They may fragment global research markets and reinforce technological blocs.

This creates the emerging concept of:

AI research sovereignty.

21. Global Fragmentation

AI research is increasingly divided among regulatory systems.

Major governance approaches include:

• United States;
• European Union;
• United Kingdom;
• China;
• India;
• Japan;
• Canada;
• international organisations.

Different jurisdictions may impose different requirements concerning:

• safety;
• data;
• research transparency;
• model deployment;
• export controls;
• competition;
• intellectual property.

This creates regulatory arbitrage.

Researchers and companies may relocate activities to jurisdictions with lower compliance burdens.

22. International Coordination

A global AI research-governance framework could establish common principles.

Principle 1 — Research neutrality

Public funding should not unnecessarily favour particular firms.

Principle 2 — Competitive access

Publicly supported AI infrastructure should provide fair access.

Principle 3 — Interoperability

Researchers should be able to move between compatible infrastructure where technically feasible.

Principle 4 — Transparency

Government-funded AI programmes should disclose:

• funding allocation;
• selection criteria;
• conflicts of interest;
• major commercial partnerships.

Principle 5 — Innovation protection

Merger control should consider potential innovation competitors.

Principle 6 — Proportional safety regulation

Compliance requirements should reflect actual risk.

Principle 7 — Open scientific access

Fundamental research should remain as accessible as reasonably possible.

23. Competition Between States

There is also a geopolitical dimension.

Countries increasingly compete for:

• AI researchers;
• semiconductor capacity;
• cloud infrastructure;
• university talent;
• venture capital;
• scientific datasets.

This creates a new form of innovation competition between states.

Government policy can therefore resemble industrial policy.

The legal challenge is to distinguish:

legitimate strategic investment

from

discriminatory market protectionism.

24. Regulatory Capture

AI research governance can become vulnerable to regulatory capture.

Large companies may possess:

• greater technical expertise;
• greater lobbying capacity;
• greater access to policymakers;
• greater ability to participate in standards organisations;
• greater ability to fund research institutions.

Consequently, technical standards may unintentionally reflect incumbent interests.

A robust governance model therefore requires:

• independent experts;
• disclosure of conflicts;
• participation by startups;
• academic representation;
• civil-society participation;
• transparent standard-setting.

25. AI Innovation Steering and Constitutional/Public Law

Government steering also raises public-law questions.

Where public funds or regulatory decisions determine technological priorities, authorities may need to satisfy:

• legality;
• rationality;
• proportionality;
• equality;
• procedural fairness;
• non-discrimination;
• transparency.

A government cannot necessarily designate a particular company as the national AI champion without considering whether the programme is legally authorised and rationally designed.

26. Emerging Legal Doctrine: Innovation as a Competitive Parameter

Traditional competition law often focuses on:

• price;
• output;
• quality.

AI markets require a fourth dimension:

innovation trajectory.

A firm may harm competition even where prices remain low if it:

• prevents development of rival models;
• acquires emerging research competitors;
• monopolises AI researchers;
• restricts access to essential compute;
• controls critical datasets;
• forecloses interoperability.

Thus, innovation competition becomes a central metric.

27. A Practical Legal Test

A regulator assessing an AI research-governance arrangement can use the following framework:

Step 1 — Identify the research resource

Is it:

• compute?
• data?
• talent?
• funding?
• infrastructure?
• IP?
• distribution?

Step 2 — Identify the controller

Is control exercised by:

• government;
• university;
• dominant technology firm;
• consortium;
• cloud provider?

Step 3 — Determine dependence

Can competitors reasonably obtain substitutes?

Step 4 — Examine exclusion

Does the governance arrangement exclude:

• startups;
• independent researchers;
• rival models;
• foreign researchers?

Step 5 — Examine innovation effects

Does it reduce:

• research diversity;
• technological experimentation;
• entry;
• future competition?

Step 6 — Consider legitimate objectives

Are restrictions justified by:

• national security;
• privacy;
• AI safety;
• intellectual property;
• research integrity?

Step 7 — Apply proportionality

Could the same objective be achieved through a less restrictive mechanism?

28. Key Legal Tensions

29. Overall Legal Position

The major cases demonstrate several enduring principles:

1. Microsoft shows that technological ecosystem control can protect market power.
2. Google Shopping demonstrates the importance of self-preferencing and algorithmic visibility.
3. IMS Health establishes the exceptional nature of compulsory access to protected resources.
4. Bronner provides a restrictive framework for essential-facility access.
5. Microsoft interoperability demonstrates that technical incompatibility can have competitive significance.
6. Meta/Instagram/WhatsApp litigation highlights the importance of potential and nascent competition.
7. Dow/DuPont demonstrates why innovation itself can be a competitive parameter.

Together, these principles are highly relevant to AI research governance.

30. Conclusion

Global research governance of AI systems and innovation steering is becoming a core competition-law and public-law problem.

The fundamental challenge is not simply whether governments should regulate AI. It is how they regulate AI without allowing regulation, funding, infrastructure ownership or strategic industrial policy to determine the winners and losers of technological innovation.

The most important future legal questions will concern:

• equitable access to compute;
• public AI infrastructure;
• university–industry research arrangements;
• AI researcher concentration;
• foundation-model mergers;
• research-data access;
• interoperability;
• government AI subsidies;
• export controls;
• AI safety compliance costs;
• open versus closed research;
• and international coordination.

The emerging principle should therefore be:

AI governance should protect safety and public interests while preserving pluralism in research, contestability in AI markets and diversity in technological innovation.