Explainability As A Market Access Requirement For AI Systems

Introduction

Explainability as a market-access requirement refers to the use of legal or regulatory rules requiring an AI system to provide sufficient information about how it reaches, supports, or produces decisions before that system may lawfully enter or remain in a market.

The concept is particularly important where AI systems operate in credit, insurance, employment, healthcare, public administration, advertising, online platforms, autonomous systems, and other economically significant markets. A regulator may require a developer or operator to demonstrate that the system's decisions are sufficiently understandable, auditable, contestable, and reproducible.

Explainability therefore operates at the intersection of:

1. Competition law;
2. AI regulation;
3. Data protection law;
4. Consumer protection;
5. Administrative law;
6. Sector-specific regulation; and
7. Fundamental rights and procedural fairness.

The important competition-law question is whether explainability is merely a compliance obligation, or whether it can become a condition of market participation that affects entry, innovation, interoperability and competitive access.

1. Meaning of Explainability

AI explainability concerns the ability to provide an intelligible account of:

• what information the system considered;
• what factors materially influenced the output;
• how the model was trained or configured;
• why a particular decision was generated;
• whether human intervention occurred;
• what uncertainty or limitations exist;
• whether discriminatory or prohibited variables influenced the result; and
• how the decision can be challenged or corrected.

Explainability does not necessarily require disclosure of the entire source code or model weights.

A regulatory system may instead require:

• decision-specific explanations;
• feature-level explanations;
• documentation;
• audit trails;
• model cards;
• testing results;
• risk-management documentation;
• human-review mechanisms;
• logging;
• reproducibility;
• explanation of adverse decisions.

Thus:

Explainability is best understood as a spectrum rather than a binary requirement.

2. Explainability as a Market-Access Requirement

A market-access requirement exists when compliance with an explainability standard is necessary to:

• obtain regulatory authorization;
• obtain a licence;
• register an AI system;
• supply services to regulated customers;
• participate in public procurement;
• deploy an AI system in a regulated sector;
• access critical infrastructure;
• continue commercial deployment; or
• avoid prohibition or suspension.

For example, suppose an AI credit-scoring provider wishes to enter a banking market.

A regulator could require the provider to demonstrate that:

1. credit decisions can be explained;
2. relevant factors can be identified;
3. adverse decisions can be challenged;
4. discriminatory effects can be tested;
5. decisions can be audited;
6. records can be preserved.

The provider's ability to compete therefore becomes partly dependent upon its ability to satisfy the explainability requirement.

3. Why Explainability Can Affect Competition

Explainability requirements can have pro-competitive and anti-competitive consequences.

Pro-competitive effects

Explainability can:

• increase consumer trust;
• reduce information asymmetry;
• facilitate switching;
• permit comparison between competing AI products;
• expose discriminatory practices;
• facilitate regulatory supervision;
• prevent hidden exclusionary algorithms;
• make automated decisions contestable;
• reduce uncertainty for business customers.

Potential restrictive effects

Conversely, poorly designed requirements may:

• increase fixed costs of entry;
• favour large incumbents;
• disadvantage open-source developers;
• protect established vendors with compliance departments;
• create certification bottlenecks;
• require disclosure of commercially sensitive information;
• discourage experimentation;
• make complex machine-learning systems commercially impractical.

The competition-law challenge is therefore to distinguish necessary explainability from excessive explainability burdens.

4. Explainability and Barriers to Entry

A particularly important issue is whether explainability requirements become a regulatory barrier to entry.

Consider two firms:

Incumbent A

• has thousands of employees;
• possesses extensive legal resources;
• operates existing compliance infrastructure;
• can conduct large-scale model auditing.

Entrant B

• has a technically superior AI model;
• has limited financial resources;
• lacks an established compliance department.

If the regulator requires extensive documentation, continuous testing and expensive certification, Entrant B may be unable to enter.

The requirement may therefore have the following effect:

Explainability obligation → compliance cost → increased fixed cost → reduced entry → greater incumbent advantage

This does not automatically make the regulation unlawful. The central question becomes whether the burden is necessary, proportionate and competitively justified.

5. Explainability and Regulatory Certification

One possible regulatory model is a certification system.

An AI provider could be required to demonstrate:

Before market entry

• model documentation;
• risk assessment;
• training-data governance;
• testing;
• explanation methodology;
• human oversight;
• logging capability.

During operation

• continuous monitoring;
• incident reporting;
• explanation of material decisions;
• periodic independent auditing.

After an adverse event

• regulator access to logs;
• reconstruction of the decision;
• explanation of causation;
• corrective measures.

This converts explainability from an abstract ethical principle into a market-entry condition.

6. Explainability and Proprietary Technology

A major difficulty arises where explainability conflicts with intellectual property.

AI firms may argue that disclosure requirements could reveal:

• source code;
• model architecture;
• proprietary datasets;
• training techniques;
• parameter structures;
• trade secrets.

Regulators therefore increasingly need to distinguish between:

Internal regulatory explainability

Information supplied confidentially to regulators or auditors.

User-facing explainability

Information supplied to affected individuals.

Public explainability

Information made publicly available.

These should not automatically be treated as equivalent.

A competition-sensitive regulatory framework should normally require the minimum information necessary to establish accountability, rather than indiscriminate disclosure of proprietary technology.

7. Explainability and Data Protection

Explainability is particularly significant where AI processes personal data.

Automated decisions may affect:

• credit;
• employment;
• insurance;
• housing;
• education;
• public benefits.

Data-protection law may consequently require information about automated decision-making and meaningful safeguards against harmful automated outcomes.

The competition issue arises where compliance with these obligations determines whether an AI provider can commercially operate.

Thus:

Data-protection explainability can indirectly become a market-access condition for AI suppliers.

8. Explainability and Contestability

Explainability also promotes contestability.

Suppose an AI system rejects a loan application.

If the individual receives only:

"Application rejected by automated assessment."

there is little meaningful opportunity to challenge the decision.

But if the system provides:

"The decision was materially influenced by debt-to-income ratio, recent repayment history and insufficient verified income."

the individual can potentially:

• correct inaccurate information;
• challenge the assessment;
• provide additional evidence;
• seek human review.

Explainability therefore supports both individual rights and competitive discipline.

9. Explainability and Algorithmic Discrimination

Explainability can expose discriminatory mechanisms.

An apparently neutral AI system may produce systematically different results because of:

• proxy variables;
• geographic information;
• historical datasets;
• correlated attributes;
• biased labels;
• feedback loops.

A sufficiently explainable system allows regulators to investigate whether the discriminatory outcome results from:

data → model design → feature selection → decision rule → output

This is particularly important where AI becomes a gatekeeper for access to economically important services.

10. Explainability as an Essential Regulatory Capability

In highly automated markets, regulators themselves may suffer from an epistemic disadvantage.

A regulator may be unable to determine:

• why a dominant platform changed rankings;
• why an algorithm excluded competitors;
• whether an algorithm discriminates;
• whether pricing algorithms coordinated;
• whether an AI recommendation system favoured affiliated services.

Explainability requirements can therefore function as a form of regulatory infrastructure.

The requirement does not merely protect consumers. It enables the regulator to exercise competition-law powers effectively.

11. Explainability and Dominant Firms

The issue becomes particularly significant where an AI system is operated by a dominant undertaking.

A dominant platform could potentially claim:

"The model is too complex to explain."

If regulators accept this argument without qualification, algorithmic complexity could become a shield against competition-law scrutiny.

A competition authority may instead require the dominant firm to provide sufficient information to establish:

• ranking criteria;
• exclusionary effects;
• discrimination;
• self-preferencing;
• access restrictions;
• interoperability decisions;
• algorithmic changes.

In this context, explainability can operate as a condition for exercising market power responsibly.

12. Explainability and Self-Preferencing

Suppose a dominant platform uses AI to rank products.

Its algorithm repeatedly places its own products above competing products.

The platform claims that the ranking is produced by an opaque machine-learning model.

A regulator may require sufficient explainability to determine:

1. whether self-preferencing occurred;
2. whether competing products were disadvantaged;
3. which ranking variables mattered;
4. whether commercial relationships influenced the model;
5. whether the algorithm changed following competitive threats.

Explainability thus becomes relevant to abuse-of-dominance analysis.

13. Explainability and Algorithmic Collusion

Another application concerns pricing algorithms.

Suppose competing firms use AI pricing systems.

The systems repeatedly produce:

• parallel price increases;
• rapid responses to competitors;
• stable supra-competitive prices.

The authority needs to determine whether the outcome reflects:

• independent optimisation;
• explicit communication;
• algorithmic coordination;
• hub-and-spoke arrangements;
• conscious parallelism.

Explainability and auditability may provide evidence about the algorithm's:

• objectives;
• inputs;
• constraints;
• learning process;
• reaction functions.

This can significantly affect enforcement.

14. Explainability and Public Procurement

Governments can use explainability as a procurement-access requirement.

For example, a public authority purchasing AI for:

• tax administration;
• policing;
• healthcare;
• public benefits;
• immigration;
• education;

could require bidders to demonstrate:

• explainability;
• auditability;
• human oversight;
• bias testing;
• logging;
• accountability mechanisms.

This can improve public-sector accountability but can also raise competition concerns if requirements are unnecessarily tailored to incumbent technologies.

A procurement authority should therefore ensure that requirements are:

• objective;
• technology-neutral;
• proportionate;
• transparent;
• equally available to new entrants.

15. Explainability and Interoperability

Explainability can also support interoperability.

A business using an AI system may need to understand:

• why an API produces particular outputs;
• how data are processed;
• what limitations apply;
• how decisions can be reproduced.

Without sufficient documentation, customers can become dependent on the provider.

Thus explainability can sometimes reduce technological lock-in.

However, regulators should distinguish explainability from full technical disclosure. Requiring complete disclosure of proprietary systems could itself undermine innovation.

16. Explainability and Switching Costs

Suppose a business adopts an AI compliance system.

Over time it becomes dependent on:

• proprietary data structures;
• proprietary model outputs;
• undocumented decision rules;
• inaccessible historical records.

When the business attempts to switch suppliers, it cannot understand or reproduce previous decisions.

Explainability and documentation requirements can reduce this dependency by requiring:

• accessible logs;
• decision records;
• standardized documentation;
• exportable information;
• reproducible outputs.

Thus explainability can indirectly promote multi-homing and switching.

17. Six Important Case Laws

1. Google Spain SL v AEPD and Mario Costeja González (CJEU, 2014)

This case concerned search-engine indexing and personal-data rights.

The Court recognized significant obligations surrounding the processing and presentation of personal information by search engines.

Relevance

Although not an AI explainability case in the modern sense, it demonstrates the broader principle that powerful information intermediaries can be subjected to legal obligations concerning:

• transparency;
• accountability;
• individual rights;
• control over information processing.

Competition significance

For AI search and recommendation systems, the case supports the proposition that technological complexity does not automatically eliminate legal accountability.

2. SCHUFA Holding AG case (CJEU, 2023)

The SCHUFA litigation concerned automated credit scoring and the GDPR's rules governing automated decision-making.

The Court examined circumstances in which a score generated through automated processing could effectively determine an individual's outcome.

Importance for explainability

Credit scoring is a paradigmatic example of an AI-like system acting as a market-access gatekeeper.

If a score determines whether a person receives credit, then the operation of the scoring system can materially affect access to an economic opportunity.

Competition significance

The case illustrates why regulators may demand sufficient transparency around automated scoring systems where opaque systems determine access to economically significant services.

3. State of Wisconsin v. Loomis (Wisconsin Supreme Court, 2016)

The case concerned the use of the COMPAS risk-assessment system in criminal sentencing.

The defendant challenged the use of a proprietary algorithm because he could not fully examine the underlying methodology.

Importance

The case is particularly significant because it demonstrates the tension between:

algorithmic opacity + proprietary secrecy + consequential decision-making.

The court permitted use of the assessment subject to limitations and cautions.

Competition significance

The case illustrates why regulators may distinguish between:

• disclosure of the complete algorithm;
• disclosure of relevant methodology;
• disclosure of limitations;
• meaningful explanation of the resulting decision.

This is directly relevant to market-access regulation of proprietary AI.

4. R (Bridges) v Chief Constable of South Wales Police (Court of Appeal, 2020)

This case concerned the use of automated facial-recognition technology by police.

The Court of Appeal found problems concerning the legal framework governing deployment and compliance with equality and data-protection requirements.

Importance

The decision demonstrates that automated technology deployed by powerful institutions must operate within a sufficiently defined legal framework.

Market-access significance

For AI providers supplying public authorities, compliance with transparency, legality and accountability requirements can become a condition of deployment.

The case therefore illustrates how public-law requirements can indirectly regulate market access for AI technology.

5. R (Privacy International) v Investigatory Powers Tribunal (UK Supreme Court, 2019)

The case concerned judicial review and the limits of executive power under statutory schemes involving national-security surveillance.

Relevance

The broader principle is that the exercise of technologically sophisticated governmental power cannot simply escape judicial scrutiny because the underlying processes are complex or institutionally sensitive.

AI relevance

As governments increasingly deploy AI-assisted surveillance and decision-making, explainability and reviewability become important components of lawful governance.

Competition significance

Where governments purchase or deploy AI systems, suppliers may need to satisfy requirements that permit meaningful legal scrutiny of automated processes.

6. United States v. Google LLC — Search / Search Advertising Litigation

The major US antitrust litigation involving Google's search and search-advertising practices illustrates the broader competition-law importance of understanding algorithmic systems used by dominant digital platforms.

The litigation has involved issues concerning:

• ranking;
• defaults;
• distribution;
• search-quality mechanisms;
• advertising systems;
• data advantages;
• exclusionary conduct.

Explainability significance

Where a dominant platform controls an important digital gateway, understanding how its systems operate may be essential to determining whether apparently technical decisions have exclusionary consequences.

Market-access significance

An explainability requirement can therefore become an enforcement mechanism ensuring that dominant AI-driven platforms cannot use algorithmic complexity to prevent regulators or competitors from understanding potentially exclusionary conduct.

18. Additional Relevant Authorities

Several other authorities provide useful doctrinal support.

Google Shopping (European Commission / General Court)

The Google Shopping litigation demonstrates the importance of understanding ranking and visibility mechanisms when assessing self-preferencing and exclusionary conduct.

Intel v Commission

The case illustrates the importance of economic analysis when evaluating exclusionary practices, particularly where complex commercial mechanisms cannot simply be characterized from their form.

Bronner

The essential-facilities doctrine provides a useful conceptual comparison where access to an important technological infrastructure becomes necessary for competitors.

IMS Health

The case demonstrates how intellectual property rights and market access can intersect where proprietary assets become indispensable for competition.

Microsoft

The Microsoft litigation illustrates the importance of interoperability information and access to technical information in maintaining competitive markets.

19. Explainability and Article 102 TFEU

Under Article 102 TFEU, explainability may become relevant where an AI system is controlled by a dominant undertaking.

Potential theories include:

A. Refusal to provide necessary information

A dominant undertaking could potentially restrict competitors' access to information required for interoperability or effective competition.

B. Discriminatory algorithmic treatment

AI systems could produce systematically different treatment for competitors.

C. Self-preferencing

The dominant platform could configure ranking or recommendation algorithms to favour its own services.

D. Margin squeeze

AI-driven pricing or allocation systems could contribute to discriminatory upstream/downstream conditions.

E. Exploitative opacity

In certain circumstances, opaque automated systems could contribute to unfair or discriminatory conditions.

Explainability would not independently establish an Article 102 infringement. It would instead operate as an evidentiary and regulatory instrument.

20. Explainability and UK Competition Law

Under UK competition law, similar questions arise under:

• Chapter II Competition Act 1998;
• Digital Markets, Competition and Consumers Act 2024;
• sector-specific regulation;
• consumer protection legislation.

The UK framework is particularly relevant because powerful digital firms can be subjected to pro-competition interventions and conduct requirements.

An explainability obligation could potentially require a designated firm to:

• document algorithmic changes;
• provide information to regulators;
• maintain audit trails;
• explain ranking decisions;
• facilitate independent testing;
• provide information necessary to evaluate discriminatory or exclusionary conduct.

The crucial safeguard is proportionality.

21. Explainability and EU AI Regulation

The EU AI regulatory architecture makes explainability especially important for high-risk AI systems.

Relevant obligations can concern:

• transparency;
• documentation;
• record keeping;
• human oversight;
• accuracy;
• robustness;
• cybersecurity;
• risk management.

This means that explainability can become functionally connected to the ability to place certain AI systems on the European market.

The important conceptual development is:

AI regulation can convert transparency from a voluntary governance principle into a condition of lawful commercialization.

22. Can Explainability Requirements Be Anti-Competitive?

Yes, potentially.

A requirement could become problematic if it:

1. Disproportionately burdens entrants

Large firms can absorb compliance costs more easily.

2. Protects incumbents

Established firms may already possess certification infrastructure.

3. Favours particular technology

A regulator might unintentionally design requirements around one type of AI architecture.

4. Requires excessive disclosure

Unnecessary disclosure could destroy legitimate trade-secret protection.

5. Creates certification monopolies

If only a small number of approved auditors can certify AI systems, certification itself may become a bottleneck.

6. Prevents experimentation

Start-ups may be unable to conduct limited pilots because they cannot satisfy full-scale compliance requirements.

23. Proportionality Test

A competition-sensitive explainability regime should ask four questions.

Question 1: Is the requirement legitimate?

Examples:

• preventing discrimination;
• protecting consumers;
• ensuring safety;
• facilitating competition;
• protecting fundamental rights.

Question 2: Is explainability suitable?

Will explanation actually help achieve the regulatory objective?

Question 3: Is it necessary?

Could the same objective be achieved through:

• auditing;
• outcome testing;
• human review;
• logging;
• statistical monitoring?

Question 4: Is the burden proportionate?

Does the benefit justify the cost imposed on:

• entrants;
• SMEs;
• open-source developers;
• consumers;
• innovators?

This proportionality analysis is critical to prevent explainability from becoming unnecessary regulatory protection for incumbents.

24. Outcome-Based Versus Process-Based Explainability

A particularly important regulatory choice is between:

Process-based model

The firm must explain how the AI works.

Advantages:

• deeper transparency;
• better auditing;
• greater regulatory understanding.

Disadvantages:

• expensive;
• technically difficult;
• potential trade-secret disclosure.

Outcome-based model

The firm must demonstrate that the system produces:

• accurate outcomes;
• non-discriminatory outcomes;
• auditable results;
• contestable decisions.

Advantages:

• potentially lower compliance burden;
• technology-neutral;
• more accessible to SMEs.

Disadvantages:

• may fail to identify hidden mechanisms causing harmful outcomes.

A sophisticated regime should normally combine both approaches according to risk.

25. Explainability and Trade Secrets

A balanced framework should create different disclosure levels.

This avoids the false choice between:

complete secrecy and complete public disclosure.

26. Explainability as a Competition-Enabling Requirement

Explainability can actually promote competition.

It can facilitate:

• switching;
• interoperability;
• independent auditing;
• benchmarking;
• consumer comparison;
• regulatory enforcement;
• detection of self-preferencing;
• detection of discriminatory access;
• investigation of algorithmic coordination.

Consequently, explainability should not be viewed solely as a compliance cost.

It can also be understood as a form of competitive infrastructure.

27. Risks of Regulatory Overreach

There are nevertheless significant risks.

An excessively detailed explainability requirement could:

increase compliance costs → discourage entry → reduce innovation → consolidate incumbent market power.

This creates a paradox:

A regulation designed to make AI markets more accountable could unintentionally make them less competitive.

This is particularly significant in fast-moving AI markets where:

• model architectures evolve rapidly;
• open-source alternatives compete with proprietary models;
• smaller firms innovate quickly;
• explainability techniques remain imperfect.

28. Recommended Regulatory Model

A competition-sensitive framework should adopt a tiered explainability obligation.

Tier 1 — Low-risk AI

Require:

• basic documentation;
• transparency notice;
• basic logging.

Tier 2 — Commercially consequential AI

Require:

• meaningful decision explanations;
• auditability;
• data-quality documentation;
• human review.

Tier 3 — High-risk AI

Require:

• comprehensive documentation;
• independent testing;
• detailed logging;
• human oversight;
• bias assessment;
• regulator access.

Tier 4 — Systemically important AI

Require:

• continuous monitoring;
• independent audits;
• regulator-access mechanisms;
• incident reporting;
• algorithmic change notifications;
• enhanced interoperability and contestability safeguards.

This avoids imposing the same compliance burden on every AI provider.

29. Competition-Law Decision Framework

A competition authority examining an explainability requirement could use the following framework:

AI system

↓

Does it materially affect market access or competitive conditions?

↓

What risk does opacity create?

↓

Consumer harm / discrimination / exclusion / collusion / safety

↓

Is explainability necessary?

↓

Can a less restrictive measure achieve the same objective?

↓

What compliance costs fall on entrants?

↓

Does the requirement favour incumbents?

↓

Can confidential disclosure protect trade secrets?

↓

Is independent auditing sufficient?

↓

Apply proportionate, technology-neutral requirement

30. Key Legal Issues

The major legal questions surrounding explainability as a market-access requirement are therefore:

1. Whether explainability is legally mandatory.
2. Who must provide the explanation.
3. To whom the explanation must be provided.
4. How technically detailed the explanation must be.
5. Whether trade secrets must be protected.
6. Whether the requirement is proportionate.
7. Whether it creates barriers to entry.
8. Whether certification creates a regulatory bottleneck.
9. Whether dominant firms should face heightened obligations.
10. Whether SMEs require differentiated compliance requirements.
11. Whether regulators themselves have sufficient technical expertise.
12. Whether explanation must be continuous after market entry.

31. Key Case-Law Principles

32. Conclusion

Explainability can legitimately function as a market-access requirement for AI systems where opacity creates substantial risks to competition, consumer protection, fundamental rights or regulatory accountability.

Its significance extends beyond simply asking whether an AI system is understandable. Explainability can determine whether:

• an AI provider may enter a regulated market;
• a platform may deploy an AI system;
• a supplier can participate in public procurement;
• a dominant undertaking must disclose information to a regulator;
• consumers can challenge automated decisions;
• competitors can understand discriminatory access conditions.

The competition-law tension is fundamental.

On one side:

Opacity can conceal exclusionary conduct, discrimination, algorithmic coordination and abusive market power.

On the other:

Excessive explainability obligations can increase entry costs, expose trade secrets and strengthen established incumbents.

Accordingly, the preferred legal approach is risk-based, proportionate, technology-neutral and confidentiality-sensitive. Explainability should be sufficient to permit meaningful accountability and competitive scrutiny, but should not automatically require disclosure of every aspect of a proprietary AI system.

The emerging principle can therefore be stated as: