Global Disease Modeling AI Systems and Policy Dependency Risks

1. Introduction

Global disease-modeling AI systems are computational systems that use epidemiological, clinical, mobility, genomic, demographic, environmental, economic, and behavioural data to forecast disease transmission, estimate hospital demand, identify high-risk populations, simulate interventions, and support public-health decision-making.

Artificial intelligence can significantly improve disease surveillance and modeling. However, concentration of these capabilities in a small number of technology companies, research institutions, cloud providers, data platforms, or model developers can create policy dependency risks. Governments may become dependent on privately controlled models for decisions concerning lockdowns, vaccination allocation, border restrictions, hospital capacity, disease surveillance, or emergency procurement.

The central competition-law and governance question is therefore:

When an AI disease-modeling system becomes an indispensable input into public-health policymaking, can control over the model, data, compute, interfaces, standards, or scientific infrastructure become a source of economic and institutional power?

The issue sits at the intersection of competition law, public procurement, digital-platform regulation, data governance, AI regulation, administrative law, public health, and fundamental rights.

2. What Is a Disease-Modeling AI System?

A disease-modeling AI system may perform several functions:

1. Disease forecasting – predicting infection rates or outbreaks.
2. Transmission modeling – estimating reproduction rates and transmission pathways.
3. Hospital-demand prediction – forecasting ICU beds, ventilators, medicines, or staffing requirements.
4. Intervention simulation – modeling the effects of vaccination, masking, distancing, travel restrictions, or school closures.
5. Genomic prediction – identifying potential variants or mutation patterns.
6. Early-warning surveillance – detecting abnormal disease signals.
7. Resource optimization – allocating scarce healthcare resources.
8. Population-risk scoring – identifying regions or groups requiring intervention.
9. Policy optimization – recommending combinations of public-health measures.
10. Automated decision support – continuously updating government recommendations as new data arrive.

The technological stack can therefore extend far beyond an individual algorithm.

Typical dependency chain

Health data → data infrastructure → cloud compute → AI model → epidemiological forecast → policy recommendation → government decision → population outcome

Control at any point in this chain may create strategic leverage.

3. Why Globalization Creates Competition Concerns

Disease modeling is increasingly global.

A model may be:

• developed in one country;
• trained using data from multiple jurisdictions;
• hosted on cloud infrastructure in another country;
• supplied through an API;
• integrated into a national health ministry's dashboard;
• continuously updated using international datasets; and
• relied upon by several governments simultaneously.

This creates cross-border dependency.

A government may technically own its health data while nevertheless lacking control over:

• the model architecture;
• training methodologies;
• compute infrastructure;
• proprietary datasets;
• model weights;
• APIs;
• validation protocols;
• software updates;
• cybersecurity infrastructure;
• technical personnel; or
• interpretability tools.

Consequently, data sovereignty does not necessarily equal computational sovereignty.

4. Policy Dependency as a Competition-Law Problem

Policy dependency becomes particularly significant when an AI system is no longer merely an analytical tool but becomes a critical input into governmental decision-making.

For example:

Government → relies on AI model → model forecasts severe outbreak → government imposes restrictions → economic activity changes → businesses adapt → government continues relying on same model.

This can create a feedback loop.

If the model provider controls the underlying infrastructure, the provider may obtain significant institutional bargaining power.

The concern is not that every disease model constitutes a monopoly. Rather, the concern arises where concentration + indispensability + switching costs + opacity + public dependence combine.

5. Relevant Markets

Several relevant markets could potentially emerge.

A. Disease-modeling software

The relevant market could consist of specialized epidemiological modeling systems.

B. AI forecasting services

Governments might purchase disease forecasts as a service rather than software.

C. Health-data infrastructure

A firm controlling access to important health datasets could possess significant upstream power.

D. Cloud compute

Large-scale epidemiological AI may depend upon GPU/TPU-intensive computing.

E. Health-data analytics

The market could encompass analytics and prediction services using government and private healthcare data.

F. Public-health decision-support systems

The ultimate market may involve integrated systems combining:

• data;
• analytics;
• forecasting;
• dashboards;
• policy simulation; and
• automated recommendations.

6. Sources of Market Power

6.1 Data advantages

Disease modeling depends heavily upon high-quality data.

A firm with access to:

• hospital records;
• genomic databases;
• vaccination records;
• mobility data;
• laboratory results;
• demographic information; and
• historical epidemiological data

may obtain substantial informational advantages.

The competitive problem becomes more serious where competitors cannot replicate the dataset.

6.2 Network effects

More users can generate more information about:

• model performance;
• disease trends;
• geographical variation;
• intervention outcomes.

That information can improve the model, creating a self-reinforcing cycle:

More governments → more data → better model → greater adoption → more governments.

6.3 Switching costs

Once a government integrates a disease-modeling platform into:

• emergency dashboards;
• procurement systems;
• hospital planning;
• vaccination systems;
• surveillance infrastructure;

switching providers may become extremely expensive.

The resulting lock-in can arise even without contractual exclusivity.

6.4 Technical interoperability

If a dominant system uses proprietary:

• APIs;
• data formats;
• model interfaces;
• validation standards; or
• dashboards,

competitors may face significant barriers to entry.

Interoperability therefore becomes a competition-policy issue.

7. Compute Dependency

Advanced disease-modeling AI can require substantial computational resources.

If only a small number of cloud providers control relevant:

• GPUs;
• TPUs;
• high-performance computing;
• cloud clusters;
• specialized accelerators;

then the disease-modeling ecosystem may become dependent upon upstream compute suppliers.

This produces a vertical dependency chain:

Chip → cloud → AI infrastructure → disease model → government policy.

A bottleneck at the compute level can consequently influence downstream public-health technology.

8. Model Opacity and Policy Dependency

Traditional epidemiological models can generally be scrutinized through published:

• assumptions;
• equations;
• parameters;
• datasets;
• methodologies.

Some AI systems may be much more difficult to audit.

A government could therefore receive a recommendation such as:

"The model predicts a 73% probability of severe regional transmission."

But policymakers may not know precisely:

• why the prediction was produced;
• which variables mattered most;
• whether the model has geographic bias;
• whether training data were representative;
• whether the model changed after an update;
• how sensitive the result is to missing data.

This creates an important governance principle:

Policy dependency without epistemic transparency is particularly risky.

9. Model Updates as a Governance Problem

AI systems can change over time.

A disease-modeling provider may alter:

• model weights;
• training datasets;
• parameters;
• safety filters;
• forecasting methodology;
• data sources.

If government agencies rely continuously upon the system, a provider could effectively alter the analytical foundation of public policy without a new legislative or procurement decision.

This creates what may be called algorithmic delegated policymaking.

10. Risk of Scientific Centralization

If a single model becomes the dominant reference point for global disease forecasts, governments and researchers may begin using it as the benchmark against which other models are evaluated.

This can produce:

Model dominance → scientific standardization → reduced methodological diversity → increased dependency.

Scientific consensus is valuable, but excessive dependence upon a single computational infrastructure can create systemic vulnerability.

A model failure, data corruption, cyberattack, or methodological error could consequently propagate across multiple jurisdictions.

11. Competition Between Models and Epistemic Diversity

Competition law normally protects rivalry between firms.

In disease modeling, however, model diversity itself can have public value.

Multiple independent models may reveal:

• different assumptions;
• different uncertainty ranges;
• alternative causal explanations;
• different intervention outcomes.

A dominant provider that acquires or eliminates competing modeling platforms could therefore reduce not merely commercial competition but also epistemic diversity.

This creates a distinctive competition concern:

Competition may protect the diversity of knowledge used to make public-health decisions.

12. Public Procurement Risks

Governments frequently procure digital systems through contracts.

Potential concerns include:

• long-term exclusive contracts;
• proprietary formats;
• automatic renewal;
• dependence on one cloud provider;
• restrictions on independent validation;
• insufficient data portability;
• lack of source-code or model-access rights;
• restrictive licensing;
• bundled services.

A procurement contract can therefore create market power even where the initial tender was competitive.

13. Tying and Bundling

A dominant technology provider might offer:

cloud infrastructure + health-data platform + AI model + dashboard + cybersecurity + analytics

as a single package.

Bundling can create efficiency benefits, but competition concerns arise where a dominant supplier uses strength in one market to foreclose competitors in another.

For example:

Dominant cloud infrastructure → mandatory use of proprietary AI analytics → exclusion of independent disease-modeling competitors.

This resembles classic leveraging theory.

14. Exclusive Dealing

Exclusive government contracts could potentially prevent competing AI developers from obtaining:

• disease datasets;
• hospital data;
• government validation;
• reference contracts;
• cloud capacity;
• public-health integrations.

Because government datasets may be uniquely valuable, exclusivity can create substantial foreclosure.

15. Refusal to Supply Critical Data

Suppose an incumbent controls a uniquely important dataset necessary to develop accurate disease forecasts.

If access is denied to competitors, the issue may potentially resemble an essential-facilities/refusal-to-deal problem.

However, competition authorities would need to establish the relevant legal criteria, including:

• indispensability;
• lack of realistic alternatives;
• competitive foreclosure;
• justification;
• proportionality.

16. Data Portability

Government agencies should ideally be capable of exporting:

• raw data;
• metadata;
• historical predictions;
• model outputs;
• validation records;
• audit logs.

Without portability, a government may technically own its data but remain practically locked into the provider.

Thus:

Data ownership without data portability may provide insufficient competitive autonomy.

17. Interoperability

Interoperability obligations can reduce dependency.

Possible requirements include:

• standardized APIs;
• machine-readable outputs;
• open data schemas;
• interoperable dashboards;
• model-output portability;
• standardized epidemiological metadata.

This can lower switching costs and encourage multi-provider ecosystems.

18. Algorithmic Collusion and Public-Health Markets

Disease-modeling AI can also indirectly affect private markets.

Suppose AI forecasts are used to determine:

• pharmaceutical procurement;
• vaccine allocation;
• hospital purchasing;
• medical-supply inventories;
• insurance pricing.

If multiple firms use a common AI system, its recommendations could potentially coordinate market behaviour.

The competition concern therefore moves from government dependency into private-market coordination.

19. Pharmaceutical and Healthcare Effects

Disease forecasts can influence:

• drug demand;
• vaccine demand;
• hospital capacity;
• medical-device procurement;
• pharmaceutical production.

A dominant disease-modeling provider could potentially obtain valuable information concerning expected demand before market participants do.

If the same corporate group also operates in healthcare or pharmaceutical markets, vertical integration becomes especially significant.

20. Cross-Border Regulatory Conflicts

Different jurisdictions may impose different rules concerning:

• health data;
• privacy;
• AI transparency;
• medical devices;
• public procurement;
• national security;
• cybersecurity.

A global provider may therefore face conflicting obligations.

For example:

Country A: requires data localization.

Country B: requires cross-border interoperability.

Country C: requires algorithmic transparency.

Country D: treats the model as critical infrastructure.

This creates regulatory fragmentation and may reinforce the position of large firms capable of complying with multiple regimes.

21. Relevant Case Laws

The following cases provide useful legal analogies for analyzing disease-modeling AI dependency.

1. United Brands Co. v Commission, Case 27/76

The Court of Justice examined dominance and the possibility of abusive conduct by a dominant undertaking.

Relevance

The case establishes the broader principle that dominance is not itself unlawful, but a dominant undertaking bears special responsibilities concerning its market conduct.

Applied to AI disease modeling, a provider with substantial dominance over a critical forecasting ecosystem could face heightened scrutiny concerning:

• discriminatory access;
• exclusionary contracts;
• refusal to supply;
• unfair conditions;
• leveraging.

2. Commercial Solvents Corp. v Commission, Joined Cases 6/73 and 7/73

The case concerned refusal to supply an input to downstream competitors.

Relevance

It is important for disease-modeling ecosystems where an upstream firm controls a critical:

• dataset;
• computational resource;
• API;
• technical service.

If withholding the input eliminates effective downstream competition, refusal-to-supply principles become relevant.

3. Bronner v Mediaprint, Case C-7/97

The Court imposed a demanding test for compulsory access to an infrastructure controlled by a dominant undertaking.

Relevance

This is particularly important for AI infrastructure.

A disease-modeling platform should not automatically be characterized as an essential facility merely because it is useful.

A competition authority would need to examine whether:

• access is indispensable;
• replication is realistically possible;
• exclusion would eliminate effective competition;
• there is no objective justification.

4. Microsoft Corp. v Commission, Case T-201/04

The European General Court considered Microsoft's refusal to provide interoperability information and the leveraging of dominance into adjacent markets.

Relevance

This is highly relevant to:

• proprietary AI APIs;
• interoperability;
• closed health-data platforms;
• model integration;
• technical interfaces.

A dominant disease-modeling platform that deliberately prevents interoperability could potentially raise analogous concerns.

5. Google Shopping, Case T-612/17

The EU courts considered Google's treatment of competing services within its dominant search ecosystem.

Relevance

The case illustrates how a dominant digital platform can use control over an important gateway to advantage its own downstream service.

Analogously:

dominant health-data gateway → preferred disease-modeling service → foreclosure of rival models

could create competition concerns.

6. Google Android, Case T-604/18

The case concerned Google's use of contractual arrangements and dominance in mobile operating systems to strengthen related services.

Relevance

The case provides an important analogy for ecosystem leverage.

A dominant technology provider could potentially use:

• cloud dominance;
• operating-system control;
• health-data infrastructure;
• authentication;
• identity systems

to advantage its own disease-modeling product.

7. Google AdSense, Case T-334/19

The case concerned contractual restrictions that could limit competing advertising services.

Relevance

Its broader significance lies in the analysis of contractual restrictions and foreclosure.

In public-health AI, similar concerns could arise from contractual provisions restricting governments or hospitals from using competing models.

8. IMS Health GmbH & Co. OHG v NDC Health, Case C-418/01

The case concerned access to a protected information structure and the circumstances under which refusal to license can raise competition concerns.

Relevance

Disease-modeling systems may depend upon proprietary data structures and datasets.

The case is therefore relevant to the question of whether intellectual-property rights over critical data or technical structures can coexist with competition-law obligations in exceptional circumstances.

22. Lessons from the Cases

Taken together, these cases establish several useful principles.

23. Merger-Control Risks

The most important long-term risk may be concentration through acquisitions.

Suppose:

Company A controls disease data.

Company B controls AI modeling.

Company C controls cloud compute.

A merger between two or more could create a vertically integrated disease-modeling ecosystem.

Traditional turnover thresholds may fail to capture some strategically important acquisitions where the target has:

• valuable data;
• scientific talent;
• proprietary models;
• intellectual property;
• few revenues but high future potential.

Therefore, merger control may require attention to innovation competition and data concentration, not merely present revenue.

24. Killer Acquisitions

A large technology company could acquire a promising disease-modeling startup before it becomes a major competitor.

Potentially important indicators include:

• unique epidemiological datasets;
• superior predictive accuracy;
• specialized researchers;
• proprietary model architecture;
• government contracts;
• strong scientific partnerships.

Even where the target has low turnover, its competitive significance may be substantial.

25. Global Coordination Problems

Disease modeling naturally crosses borders.

Yet competition authorities operate primarily within territorial legal systems.

One authority may examine:

• cloud foreclosure;

another:

• data concentration;

another:

• public procurement;

another:

• privacy;

another:

• national security.

The absence of coordination can allow a global platform to exploit regulatory gaps.

26. Risk of Regulatory Capture

A particularly important concern is algorithmic regulatory capture.

If governments rely heavily upon a particular provider for disease forecasts, that provider may become deeply embedded in:

• advisory committees;
• emergency planning;
• procurement standards;
• technical regulations;
• international health coordination.

Over time, policymakers may begin designing regulation around the capabilities of the incumbent system.

This can produce:

Private technological architecture → public regulatory architecture.

That is considerably more consequential than ordinary vendor dependence.

27. Epistemic Lock-In

A government may initially select a model because it performs well.

Later, officials may continue using it because:

• historical datasets are stored within it;
• staff are trained around it;
• policy workflows depend on it;
• alternative systems lack comparable validation;
• changing systems would create political risk.

The original technological advantage can therefore become institutional lock-in.

28. Accountability Problem

If a policy decision is challenged, responsibility may become fragmented:

AI developer → government contractor → health ministry → epidemiologists → political authority.

Each actor may claim that another actor was responsible for the ultimate decision.

This creates a serious accountability problem.

AI should therefore generally remain a decision-support instrument rather than an unreviewable substitute for public authority.

29. Recommended Competition and Governance Safeguards

A. Multi-model procurement

Governments should avoid unnecessary dependence upon a single forecasting system.

B. Open standards

Use interoperable:

• APIs;
• data formats;
• reporting standards.

C. Data portability

Government data and historical outputs should be exportable.

D. Independent validation

Models should be independently tested.

E. Model documentation

Authorities should have access to:

• assumptions;
• methodological documentation;
• version histories;
• performance metrics.

F. Audit rights

Government contracts should include meaningful technical audit rights.

G. Exit provisions

Contracts should facilitate migration to competing providers.

H. Competition-sensitive procurement

Procurement authorities should examine whether contracts create long-term foreclosure.

I. Merger scrutiny

Competition authorities should consider:

• data;
• innovation;
• talent;
• infrastructure;
• potential competition.

J. Cross-border cooperation

Authorities should coordinate where the same global provider operates across multiple jurisdictions.

30. A Competition-Law Test for Disease-Modeling AI

A useful analytical framework is:

Step 1 — Identify the bottleneck

Is the critical input:

• data?
• compute?
• model?
• API?
• infrastructure?
• scientific validation?

Step 2 — Define the market

Determine whether the relevant market is:

• disease modeling;
• AI forecasting;
• health analytics;
• cloud compute;
• health-data infrastructure.

Step 3 — Assess dominance

Consider:

• market share;
• data advantages;
• switching costs;
• network effects;
• interoperability;
• entry barriers.

Step 4 — Assess dependency

Ask:

Can governments realistically replace the system?

Step 5 — Identify conduct

Examine:

• tying;
• bundling;
• exclusivity;
• discrimination;
• refusal to supply;
• self-preferencing;
• interoperability restrictions.

Step 6 — Examine efficiencies

The provider may legitimately argue that integration produces:

• greater accuracy;
• lower costs;
• faster forecasting;
• improved cybersecurity;
• better emergency coordination.

Step 7 — Apply proportionality

Any intervention should preserve legitimate technological efficiencies while preventing unjustified foreclosure.

31. Broader Concept: Digital Public-Health Infrastructure

The most important conceptual development is that disease-modeling AI may evolve from a software product into digital public-health infrastructure.

Once governments use it continuously, the system can become analogous in functional importance to:

• telecommunications;
• payment systems;
• cloud infrastructure;
• electricity grids;
• identity infrastructure.

This raises the possibility of applying infrastructure-oriented competition principles to certain AI ecosystems.

32. Global Systemic Risk

The ultimate risk is not merely that one company becomes dominant.

The more serious possibility is:

Data concentration + compute concentration + model concentration + policy dependence

creating a globally interconnected system.

A common model used by many governments could generate correlated errors.

For example:

1. Model makes an erroneous assumption.
2. Multiple governments rely upon it.
3. Similar policies are adopted.
4. Private markets respond simultaneously.
5. Alternative models lose influence.
6. Error becomes system-wide.

Thus, competition among independent models can function as a resilience mechanism.

33. Conclusion

Global disease-modeling AI systems can deliver enormous public-health benefits, but their increasing importance creates a novel form of infrastructural and policy dependency.

The competition-law problem is not simply whether an AI developer has a large market share. It is whether control over data, compute, models, interfaces, scientific standards, and public-sector procurement allows a small number of firms to become indispensable intermediaries between scientific information and governmental decision-making.

The principles emerging from United Brands, Commercial Solvents, Bronner, Microsoft, IMS Health, Google Shopping, Google Android and Google AdSense provide useful analytical foundations for examining these risks.

The central policy objective should therefore be:

AI-enabled public-health systems should be innovative and globally scalable without becoming closed technological chokepoints through which essential scientific information and public policy must pass.

A resilient regulatory model should preserve multi-model competition, interoperability, data portability, independent validation, transparent procurement, contestable infrastructure, merger scrutiny, and meaningful governmental exit rights. This approach treats competition not merely as a mechanism for lowering prices, but as a safeguard for scientific diversity, institutional autonomy, public accountability, and systemic resilience.