AI Credit Allocation Systems and Capital Access Gatekeeping

Introduction

AI credit-allocation systems use machine learning, automated decision engines, alternative-data analytics, behavioural scoring, and algorithmic risk models to determine who receives credit, how much credit is available, at what price, and on what conditions. They may be used by banks, fintech lenders, credit-information companies, digital-wallet providers, BNPL platforms, marketplace lenders, and institutional investors.

The competition-law significance arises when an AI system becomes a gatekeeper to capital access. A dominant lender, credit-data provider, scoring platform, cloud/AI infrastructure provider, or vertically integrated financial platform may use algorithmic control to:

• exclude competing lenders;
• deny competitors access to essential credit data;
• favour affiliated borrowers or platforms;
• discriminate against competing businesses;
• impose tying or exclusivity conditions;
• use sensitive transaction data to disadvantage rivals;
• coordinate lending or pricing decisions through algorithms;
• make entry into credit markets prohibitively difficult; or
• reinforce an existing dominant position through automated decision-making.

Importantly, using AI or having a high rejection rate is not itself an antitrust violation. The legal issue normally concerns market power, exclusionary conduct, discriminatory access, foreclosure, collusion, tying, abuse of dominance, or an unlawful discriminatory lending practice.

I. Meaning of AI Credit Allocation

An AI credit-allocation system can be represented as:

Applicant/Data → Data Collection → Feature Engineering → AI Risk Model → Credit Score → Eligibility Decision → Credit Limit → Interest/Fees → Monitoring

The system may analyse:

• income;
• repayment history;
• bank-account transactions;
• employment;
• business turnover;
• purchasing patterns;
• mobile-device information;
• location information;
• social or behavioural data;
• platform activity;
• cash-flow information;
• existing debt;
• collateral;
• transaction histories; and
• alternative credit data.

Unlike traditional credit scoring, AI systems may identify complex correlations that are difficult for human decision-makers to understand.

This creates a competition concern when the institution controlling the model also controls an important input into the downstream market.

II. Capital Access Gatekeeping

Capital-access gatekeeping occurs when a private or public intermediary effectively determines whether another undertaking can obtain the financial resources necessary to participate in a market.

Examples include:

1. Credit-data gatekeeping

A dominant credit-information platform refuses or restricts access to important credit information.

2. Lending-platform gatekeeping

A dominant digital marketplace determines which merchants receive working capital.

3. AI-scoring gatekeeping

A proprietary scoring system becomes the de facto standard for determining creditworthiness.

4. Bank-fintech gatekeeping

A major bank restricts fintech competitors' access to transaction or account information necessary to compete.

5. Platform-finance gatekeeping

An e-commerce platform provides preferential financing to its own sellers while imposing worse terms on sellers using competing platforms.

6. Infrastructure gatekeeping

A dominant cloud, payments, identity, or data infrastructure provider conditions access to credit-related services on use of its affiliated products.

III. Competition-Law Issues

A. Abuse of Dominant Position

A dominant undertaking controlling a critical credit input may potentially engage in abuse by:

• refusing access;
• imposing discriminatory access terms;
• applying different algorithmic standards to similarly situated rivals;
• degrading data quality;
• delaying access;
• charging excessive access fees;
• imposing exclusivity;
• tying credit access to another service; or
• using proprietary information to disadvantage competitors.

The key question is not merely whether the AI model is powerful, but whether the undertaking has market power and uses that power in an exclusionary manner.

IV. AI as an Essential Input

Suppose a dominant credit-data company possesses the largest and most comprehensive dataset concerning small-business repayment behaviour.

A competing AI lender requests access.

The dominant company refuses access while simultaneously using that data to operate its own lending service.

Potential questions include:

1. Is the data indispensable?
2. Can competitors reasonably reproduce it?
3. Is alternative data available?
4. Does refusal eliminate effective competition?
5. Is there an objective business justification?
6. Does the refusal protect a legitimate investment?
7. Is the refusal discriminatory?
8. Does the refusal foreclose downstream competition?

This connects AI credit markets with the essential-facilities and refusal-to-deal doctrines.

V. Six Important Case Laws

1. United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001)

Principle

Microsoft is one of the foundational modern cases concerning leveraging market power in one technological market to protect dominance in another.

Microsoft possessed substantial power in operating systems and used contractual and technical strategies concerning browsers to restrict competitive threats.

Relevance to AI credit

The analogy is important where an undertaking controls:

Credit Data + AI Scoring + Lending Platform

and uses control over one layer to restrict competition in another.

For example, a dominant credit-data provider could potentially:

• restrict access to its scoring API;
• degrade interoperability with rival lenders;
• require lenders to use its affiliated lending platform; or
• use contractual restrictions to prevent competing AI-credit systems from accessing important inputs.

Legal lesson

AI integration does not immunize conduct from competition scrutiny. The question is whether technological integration is being used to preserve or extend market power through exclusionary mechanisms.

2. Aspen Skiing Co. v. Aspen Highlands Skiing Corp., 472 U.S. 585 (1985)

Principle

The U.S. Supreme Court recognized that, under exceptional circumstances, a dominant firm’s termination of a previously profitable course of dealing can constitute unlawful exclusionary conduct.

AI-credit relevance

Consider a dominant financial-data company that historically supplies transaction or credit information to competing lenders.

It subsequently:

• terminates access;
• continues using the information itself;
• refuses commercially reasonable alternatives; and
• materially weakens downstream competitors.

The Aspen Skiing reasoning becomes relevant because the issue is not simply "refusal to deal." The surrounding circumstances may demonstrate an exclusionary strategy.

Important limitation

Aspen Skiing is an exceptional case and does not establish a general duty for dominant firms to deal with every competitor.

3. Verizon Communications Inc. v. Law Offices of Curtis V. Trinko, LLP, 540 U.S. 398 (2004)

Principle

Trinko significantly limited expansive refusal-to-deal theories.

The Supreme Court emphasized that competition law generally does not require firms to share their assets with competitors merely because sharing would facilitate competition.

AI-credit relevance

This is particularly important for proprietary AI models and credit datasets.

A company may legitimately argue:

"Our credit model and dataset are proprietary assets developed through substantial investment."

A competitor cannot automatically demand access merely because the dataset would make competition easier.

Legal lesson

AI credit gatekeeping must be analysed carefully:

Proprietary asset ≠ automatically essential facility

and

competitor disadvantage ≠ automatically antitrust violation.

4. MCI Communications Corp. v. AT&T Co., 708 F.2d 1081 (7th Cir. 1983)

Principle

MCI is a leading U.S. essential-facilities case. The Seventh Circuit articulated factors relevant to determining whether a refusal to provide access to a facility could constitute unlawful monopolization.

The traditional framework considers matters such as:

1. control of the facility by a monopolist;
2. inability of competitors reasonably to duplicate it;
3. denial of use to a competitor; and
4. feasibility of providing access.

AI-credit application

Imagine that a dominant credit-information infrastructure contains a dataset that:

• cannot reasonably be recreated;
• is indispensable for accurate credit risk assessment;
• is controlled by one undertaking; and
• is simultaneously used by that undertaking to compete downstream.

The MCI framework provides a useful analytical structure.

Modern qualification

Courts have subsequently approached essential-facilities theories cautiously, particularly after Trinko.

5. United States v. Terminal Railroad Association, 224 U.S. 383 (1912)

Principle

Terminal Railroad concerned control over critical railroad infrastructure and the ability of competing businesses to access an indispensable transportation gateway.

The case is historically important to the development of the essential-facilities/access principle.

AI-credit relevance

A modern equivalent could theoretically arise where one undertaking controls an infrastructure layer necessary for competitors to reach customers.

For example:

AI credit infrastructure → lenders → borrowers

If the infrastructure becomes indispensable and access is selectively denied to competing lenders, the Terminal Railroad principle helps explain why control over a critical bottleneck can create competition concerns.

Important distinction

A credit dataset does not automatically constitute an essential facility. Indispensability and the absence of reasonable alternatives must be established.

6. FTC v. Surescripts, LLC, No. 1:19-cv-01080 (D.D.C. 2019)

Principle

The Surescripts litigation concerned alleged exclusionary conduct in electronic prescription markets, particularly contractual restrictions and practices designed to maintain market power.

Although it did not concern AI credit allocation directly, it provides a useful modern technology-market analogy.

AI-credit relevance

Similar issues can arise when a dominant financial-data or AI platform uses:

• exclusivity;
• loyalty-inducing arrangements;
• contractual restrictions;
• customer lock-in; or
• restrictions on multi-homing

to prevent competing credit platforms from gaining scale.

Legal lesson

In digital markets, exclusion may occur through contracts and platform architecture, rather than simply through explicit denial of access.

VII. Additional Relevant Case Laws

7. FTC v. Facebook, Inc. / Meta Platforms, Inc.

The FTC's litigation concerning Facebook's alleged maintenance of monopoly power is relevant to understanding how digital platforms may use acquisitions, interoperability restrictions, and platform control to preserve market power.

AI-credit relevance

A credit platform that controls:

• borrower data;
• merchant data;
• payments;
• credit scoring; and
• lending

may create similar vertical-integration questions.

The central issue would be whether integration produces legitimate efficiencies or instead protects market power by restricting rivals.

8. Qualcomm Inc. v. FTC, 969 F.3d 974 (9th Cir. 2020)

Principle

The case concerned competition issues involving standard-essential patents and licensing practices in the semiconductor industry.

The Ninth Circuit ultimately rejected the FTC's particular theory of anticompetitive harm.

AI-credit relevance

It illustrates the importance of distinguishing:

• conduct that harms competitors; from
• conduct that harms the competitive process.

An AI lender may legitimately outperform competitors because its model is better. That alone does not establish antitrust liability.

9. United States v. Grinnell Corp., 384 U.S. 563 (1966)

Principle

Grinnell is a foundational U.S. monopolization case concerning:

1. possession of monopoly power; and
2. acquisition or maintenance of that power through exclusionary conduct.

AI-credit relevance

For AI credit markets, the analysis should therefore separate:

Market Power

from

Exclusionary Conduct.

A highly accurate AI credit model could create substantial competitive advantages without necessarily creating unlawful monopoly power.

VIII. Algorithmic Discrimination

AI credit allocation creates another important legal problem: algorithmic discrimination.

An AI model may not explicitly use a protected characteristic but may rely upon variables that operate as proxies.

For example:

ZIP code → neighbourhood characteristics → income correlation → predicted default risk

or

employment history → occupation → demographic correlation → credit decision

The system may therefore produce discriminatory outcomes even without an explicit discriminatory instruction.

IX. Competition Law and Discrimination Are Related but Distinct

It is important to distinguish two legal questions.

Question 1 — Competition

Does the conduct exclude or disadvantage competitors in a manner that harms competition?

Question 2 — Equal-access / lending discrimination

Does the system unlawfully discriminate against applicants or classes of borrowers?

The two can overlap but are not the same legal claim.

For example, a dominant credit platform might discriminate against a competing lender by restricting API access. That is primarily a competition issue.

Conversely, an AI model that systematically produces unlawful disparities between borrower groups may raise lending-discrimination concerns even if no competitor is being excluded.

X. Data Advantage and Feedback Loops

AI credit markets are particularly susceptible to data-network effects.

The cycle may look like:

More customers

↓

More transaction data

↓

Better AI model

↓

More accurate credit decisions

↓

Lower default rates

↓

More customers

↓

Even more data

This creates a feedback loop.

A dominant undertaking may consequently develop a data-based competitive advantage that is difficult for new entrants to replicate.

However, superior performance generated through innovation is not automatically unlawful.

Competition concerns become stronger where the incumbent additionally:

• prevents data portability;
• blocks interoperability;
• restricts access to APIs;
• imposes exclusivity;
• prevents customers from multi-homing;
• acquires emerging competitors; or
• uses its data advantage to foreclose rivals.

XI. Self-Preferencing in AI Credit

Suppose an e-commerce platform operates:

1. a marketplace;
2. a seller-credit business; and
3. an AI risk-scoring system.

The platform's AI determines which merchants receive working capital.

It gives favourable credit scores to merchants that:

• use its payment system;
• advertise on its platform;
• use its logistics service; or
• purchase other affiliated products.

This can raise self-preferencing and leveraging concerns.

The important factual questions would include:

• Is the platform dominant?
• Are the criteria objectively related to credit risk?
• Are competing services disadvantaged?
• Can sellers realistically obtain alternative financing?
• Does the practice foreclose rival providers?
• Are there efficiencies?
• Is the conduct transparent and consistently applied?

XII. Tying and Bundling

An AI credit provider could potentially condition access to financing upon purchasing another product.

Example:

"A merchant can receive favourable working-capital financing only if it uses our payment-processing service."

This potentially creates a:

Credit Market → Payment Market

tie.

The analysis would ordinarily examine:

1. separate products;
2. market power in the tying product;
3. coercion or practical conditioning;
4. foreclosure;
5. competitive effects; and
6. legitimate business justification.

XIII. Algorithmic Collusion

One of the most significant emerging risks is the use of AI systems by competing lenders to determine prices.

Imagine several lenders independently deploy algorithms that monitor:

• competitors' interest rates;
• loan approvals;
• credit limits;
• borrower behaviour; and
• market demand.

If algorithms merely independently respond to market conditions, that does not automatically establish collusion.

But competition concerns become much more serious where firms:

• exchange competitively sensitive information;
• deliberately configure algorithms to coordinate;
• use a common intermediary to facilitate coordination; or
• communicate through an algorithmic mechanism designed to implement an agreement.

Thus:

Autonomous parallel pricing ≠ automatically cartel

but

algorithmically implemented agreement = potentially conventional cartel conduct.

XIV. Common Algorithmic Gatekeeping Mechanisms

XV. Market Definition

Market definition becomes particularly complicated.

Possible relevant markets include:

A. Credit-information market

Provision of credit histories and financial information.

B. AI credit-scoring market

Provision of automated credit-risk assessment.

C. Digital lending market

Online provision of loans.

D. SME financing market

Credit specifically directed toward small and medium enterprises.

E. Consumer credit market

Personal loans, cards, BNPL and similar products.

F. Credit infrastructure market

APIs, identity verification, transaction-data infrastructure and related services.

The same undertaking may therefore occupy multiple vertically related markets.

XVI. Market Power Indicators

AI-credit market power may be assessed through:

• market share;
• control over unique data;
• number of borrowers;
• network effects;
• switching costs;
• API dependence;
• access to banking data;
• model accuracy;
• customer lock-in;
• interoperability;
• regulatory barriers;
• economies of scale;
• financial resources;
• vertical integration; and
• availability of alternative credit providers.

Traditional market-share analysis may therefore be insufficient where data and network effects are substantial.

XVII. The Role of Explainability

AI credit decisions can create a competition problem when customers and competitors cannot understand the decision architecture.

For example, a platform may state:

"The applicant was rejected because the algorithm determined that the applicant was high risk."

If the model is effectively a black box, it becomes difficult to determine whether:

• legitimate risk factors were used;
• discriminatory proxies were used;
• competing-platform usage was penalised;
• affiliated services were favoured;
• sensitive information was improperly incorporated; or
• the model was deliberately designed to exclude competitors.

Therefore, explainability, auditability and documentation become important elements of competition-law compliance.

XVIII. Remedies

Competition authorities may potentially consider several remedies depending upon the violation.

1. Non-discriminatory access

Require access to relevant APIs or datasets on objectively determined terms.

2. Interoperability

Require technical compatibility with competing credit providers.

3. Data portability

Permit users or businesses to transfer relevant financial information.

4. Non-exclusivity

Prohibit contracts that prevent customers from using competing credit providers.

5. Structural separation

In exceptional circumstances, separate the credit-data infrastructure from downstream lending operations.

6. Algorithmic auditing

Require independent examination of potentially exclusionary models.

7. Transparency requirements

Require disclosure of relevant decision criteria or access conditions.

8. Behavioural commitments

Prohibit discriminatory treatment of competing platforms.

XIX. Compliance Framework for AI Credit Providers

A competition-compliance programme should include:

Step 1 — Identify markets

Determine the relevant credit, data and infrastructure markets.

Step 2 — Identify gatekeeper assets

Map proprietary datasets, APIs, scoring models and infrastructure.

Step 3 — Test dominance

Assess market shares, barriers, network effects and switching costs.

Step 4 — Audit exclusion

Examine refusals, restrictions, exclusivity and discriminatory access.

Step 5 — Audit algorithms

Check whether competing-platform usage or unrelated commercial relationships influence credit decisions.

Step 6 — Test tying

Determine whether access to credit is conditioned upon purchase of another product.

Step 7 — Examine information flows

Prevent exchange of competitively sensitive information between competitors.

Step 8 — Document objective criteria

Record legitimate risk-management reasons for access and pricing decisions.

Step 9 — Establish human review

Particularly for unusual or contested automated decisions.

Step 10 — Monitor continuously

AI systems change through retraining and new datasets, so competition compliance must be continuous rather than a one-time assessment.

XX. Comparative Doctrinal Framework

XXI. Key Case-Law Principles at a Glance

Conclusion

AI credit-allocation systems occupy a strategically important position between data markets, financial markets and digital-platform markets. Their competitive significance is greatest when a single undertaking controls multiple layers:

Financial Data → AI Scoring → Credit Decision → Payment Infrastructure → Customer Platform

Such vertical integration can produce genuine efficiencies, including faster underwriting, lower transaction costs, improved fraud detection and greater access to credit. At the same time, the same architecture can create competition concerns if market power is used to deny rivals access to critical data, impose exclusivity, favour affiliated businesses, tie products, restrict interoperability or facilitate coordination.

The central legal distinction is therefore:

AI-based competitive advantage is not itself unlawful; the competition issue arises when market power is combined with exclusionary or discriminatory conduct that materially restricts the competitive process.