Mandatory Openness of Algorithms and Trade Secret Conflicts

Introduction

Mandatory openness of algorithms refers to legal or regulatory requirements that compel a dominant platform, AI provider, automated decision-maker, or other technologically significant undertaking to disclose some information about the functioning of its algorithms. The disclosure may range from broad explanations of objectives and parameters to source code, model documentation, training-data information, audit access, testing results, or even direct inspection of the algorithm.

The central legal conflict is between two legitimate interests:

1. Transparency and accountability — regulators, competitors, courts, and affected users may need sufficient information to detect discrimination, exclusion, manipulation, algorithmic collusion, self-preferencing, or other anticompetitive conduct.
2. Trade-secret protection — source code, model architecture, training methods, ranking mechanisms, parameters, datasets, and technical know-how may constitute commercially valuable confidential information.

Competition law therefore faces an important proportionality problem: how much algorithmic openness is necessary to make market power accountable without effectively transferring a firm's competitive technology to rivals?

The issue is particularly significant for AI systems, search-ranking algorithms, recommender systems, ad auctions, pricing algorithms, credit-scoring systems, marketplaces, and automated compliance or enforcement systems.

1. Meaning of Mandatory Algorithmic Openness

Mandatory openness does not necessarily mean publication of source code.

It can exist at several levels:

A. Outcome transparency

The undertaking explains:

• why a particular decision was made;
• what factors materially influenced the outcome;
• whether human review was involved;
• what rights of appeal exist.

This is generally the least intrusive form.

B. Process transparency

The undertaking provides information concerning:

• ranking criteria;
• weighting methodologies;
• decision rules;
• data categories;
• model objectives;
• relevant constraints;
• governance procedures.

C. Audit transparency

A regulator or independent auditor receives access to:

• model documentation;
• logs;
• testing records;
• performance metrics;
• model versions;
• relevant datasets;
• controlled testing environments.

This is often preferable to public disclosure.

D. Interface/API transparency

A regulator or competitor may obtain controlled access to:

• APIs;
• interoperability interfaces;
• data feeds;
• technical documentation.

E. Source-code transparency

The most intrusive form involves access to:

• source code;
• model architecture;
• algorithms;
• proprietary technical documentation.

This creates the strongest trade-secret conflict.

2. Why Competition Law May Require Algorithmic Openness

Algorithms can become an important source of market power.

A platform may use an algorithm to:

• rank its own products above rivals;
• exclude competing services;
• discriminate against particular sellers;
• manipulate advertising auctions;
• impose personalised prices;
• favour affiliated businesses;
• suppress interoperability;
• coordinate prices;
• restrict access to essential data;
• make switching difficult.

Traditional competition-law investigation may therefore be impossible without access to the relevant technological system.

Algorithmic transparency can help establish:

Conduct → mechanism → competitive effect → harm.

Without access to the mechanism, an authority may see only the final market outcome.

3. The Trade-Secret Problem

Trade secrets protect commercially valuable confidential information that derives value from not being generally known and is subject to reasonable measures of secrecy.

Algorithms can contain precisely this type of information.

Examples include:

• source code;
• proprietary ranking formulas;
• recommendation logic;
• machine-learning architecture;
• feature engineering;
• model weights;
• optimisation techniques;
• fraud-detection rules;
• pricing models;
• training techniques;
• proprietary datasets.

Mandatory disclosure can therefore create several risks.

Competitive appropriation

A rival may learn how to reproduce the undertaking's technology.

Reverse engineering

Disclosure can enable reconstruction of commercially valuable systems.

Cybersecurity risk

Publishing security-sensitive algorithms may facilitate circumvention.

Innovation disincentive

Businesses may reduce investment in R&D if regulatory disclosure effectively transfers technological advantages.

Strategic gaming

Users and competitors may manipulate a disclosed algorithm.

For example, complete disclosure of a search-ranking system could enable systematic manipulation of rankings.

4. Core Legal Principle: Transparency Must Be Proportionate

The strongest legal approach is generally graduated transparency.

The regulator should ask:

1. What regulatory objective requires disclosure?
2. What information is genuinely necessary?
3. Can the objective be achieved without revealing source code?
4. Can disclosure be restricted to a regulator?
5. Can an independent auditor conduct the examination?
6. Can confidentiality rings protect the information?
7. Can sensitive portions be redacted?
8. Can access be provided through controlled testing rather than publication?

Thus:

Regulatory necessity should not automatically become unrestricted technological disclosure.

5. Mandatory Openness and Dominant Platforms

The issue becomes especially significant where an undertaking possesses substantial market power.

A dominant platform may argue:

"Our algorithm is our intellectual property and trade secret."

The competition authority may respond:

"The algorithm is also the mechanism through which exclusionary conduct is occurring."

Neither proposition automatically determines the case.

The legal question becomes whether confidentiality can coexist with effective competition-law enforcement.

6. Six Important Case Laws

1. Microsoft Corp. v Commission

European Union — General Court, T-201/04

The Microsoft litigation is highly important for understanding the relationship between technological interoperability and intellectual-property protection.

Microsoft was required to disclose interoperability information relating to its server operating systems. Microsoft argued, among other things, that the information involved intellectual-property interests.

The EU courts nevertheless accepted that intellectual-property protection does not automatically immunise conduct from competition-law scrutiny.

Significance

The case demonstrates that:

• proprietary technology may have competition significance;
• intellectual-property rights are not absolute;
• interoperability can justify compulsory access in exceptional circumstances;
• disclosure remedies can be legitimate where necessary to eliminate competitive foreclosure.

Relevance to algorithmic openness

A modern competition authority could analogise proprietary algorithmic interfaces to interoperability information.

However, Microsoft does not establish a general rule requiring disclosure of source code.

The stronger lesson is that proprietary technological information can become subject to carefully defined access obligations when competition law requires it.

2. IMS Health GmbH & Co. OHG v NDC Health GmbH & Co. KG

Court of Justice of the European Union — Joined Cases C-418/01

IMS Health concerned access to a proprietary system used for pharmaceutical sales data.

The CJEU established demanding conditions concerning when refusal to license intellectual property could constitute an abuse of dominance.

The Court's reasoning is particularly relevant because it demonstrates the exceptional character of compulsory access.

Significance

The case emphasises that competition law should not casually transform intellectual-property rights into compulsory licensing obligations.

The circumstances must justify intervention.

Relevance to algorithms

An algorithmically controlled infrastructure may sometimes become indispensable for competitors.

But regulators must establish something stronger than:

"The algorithm is useful."

They may need to demonstrate necessity, competitive foreclosure, lack of realistic alternatives, and other applicable conditions.

3. Bronner v Mediaprint

CJEU — Case C-7/97

Bronner concerned access to a newspaper distribution system.

The CJEU adopted a restrictive approach to compulsory access under Article 102 TFEU.

The Court emphasised the importance of avoiding situations where competition law forces successful businesses to subsidise competitors or share assets merely because those assets are commercially valuable.

Significance

Bronner is fundamental to understanding essential-facility-type algorithmic access claims.

A competitor cannot simply say:

"The dominant firm's algorithm is better, therefore I should have access."

The legal threshold for compulsory access is significantly higher.

Algorithmic application

Suppose a dominant platform possesses a proprietary recommendation algorithm.

A competitor seeking access would need to establish why ordinary market competition cannot realistically occur without access and why compelling disclosure is legally justified.

4. Google Shopping

Google Search (Shopping) — Commission Decision AT.39740; General Court Case T-612/17

The Google Shopping litigation is particularly important for algorithmic competition.

The European Commission found that Google systematically gave prominent placement to its own comparison-shopping service while demoting competing comparison-shopping services.

The General Court substantially upheld the Commission's findings.

Significance

The case demonstrates that:

Algorithmic ranking can itself constitute a mechanism of exclusionary conduct.

The competition issue was not simply Google's possession of an algorithm.

Rather, it concerned how the algorithmic ranking and placement system operated within Google's dominant search ecosystem.

Trade-secret relevance

The case does not establish that Google had to publish its complete search algorithm.

Instead, competition authorities can investigate algorithmic conduct through:

• ranking evidence;
• traffic data;
• experiments;
• internal documents;
• monitoring;
• statistical evidence;
• technical investigation.

This supports a functional transparency model rather than automatic source-code publication.

5. Google Android

European Commission Decision AT.40099; General Court Case T-604/18

The Google Android proceedings concerned Google's conduct relating to the Android ecosystem and various contractual restrictions.

The case illustrates the importance of examining technological ecosystems rather than considering each contractual restriction in isolation.

Significance

Digital competition can depend upon:

• default settings;
• technical integration;
• access conditions;
• APIs;
• distribution agreements;
• interoperability;
• ecosystem architecture.

Algorithmic relevance

A regulator may therefore need technical information about how a digital ecosystem operates without necessarily requiring public release of the underlying source code.

The case supports a distinction between:

technical transparency necessary for competition enforcement

and

unrestricted disclosure of proprietary technology.

6. Bundeskartellamt v Facebook

Germany — Higher Regional Court/Düsseldorf proceedings concerning Facebook's data practices

The Facebook proceedings concerned the relationship between data collection, platform power, user conditions and competition law.

The German competition authorities treated data-related conduct as potentially connected with Facebook's dominant position.

Significance

The case demonstrates the increasing overlap between:

• competition law;
• data governance;
• platform architecture;
• user choice;
• privacy;
• technological design.

Algorithmic relevance

Modern algorithms depend heavily upon data.

Therefore, algorithmic transparency cannot always be reduced to source code.

A regulator may need to understand:

• what data is collected;
• how data is combined;
• how data affects ranking;
• whether users can meaningfully refuse particular processing;
• whether data advantages reinforce dominance.

The broader lesson is that algorithmic accountability may require information about inputs and governance, not merely code.

7. Additional Important Authorities

Several other authorities strengthen the legal framework.

Google Search (Shopping)

Shows that discriminatory algorithmic treatment can become an Article 102 problem.

Amazon Marketplace

EU and national investigations into Amazon's use of marketplace data demonstrate how algorithmic and data-driven systems can raise concerns about self-preferencing and competitive neutrality.

Apple App Store

The Apple investigations illustrate how technical architecture, app-store rules, ranking, payment systems and access conditions can collectively influence competition.

Facebook/Meta

Demonstrates the importance of data architecture in assessing platform dominance and competitive effects.

8. Algorithmic Transparency Versus Source-Code Disclosure

A critical distinction should be maintained.

The most proportionate remedy will therefore frequently be controlled access rather than publication.

9. Confidentiality Rings

One particularly useful mechanism is a confidentiality ring.

Sensitive algorithmic information can be disclosed only to:

• regulators;
• approved experts;
• external auditors;
• specially authorised lawyers;
• courts.

Competitors and the public do not receive unrestricted access.

This can reconcile two objectives:

Competition enforcement

with

trade-secret protection.

10. Independent Algorithm Audits

An alternative is mandatory independent auditing.

An undertaking could be required to permit an authorised auditor to examine:

• ranking logic;
• training data;
• model performance;
• discriminatory outcomes;
• self-preferencing;
• pricing behaviour;
• model updates;
• manipulation vulnerabilities.

The auditor could then provide the competition authority with findings without releasing the underlying source code.

This is especially appropriate for:

• AI recommendation systems;
• algorithmic pricing;
• ad auctions;
• credit scoring;
• automated procurement;
• marketplace ranking.

11. Algorithmic Transparency and AI

AI makes the trade-secret conflict substantially more complicated.

A modern AI system may contain:

• source code;
• model architecture;
• weights;
• prompts;
• training datasets;
• fine-tuning data;
• reinforcement methods;
• safety mechanisms;
• evaluation systems.

Requiring publication of all of these elements could destroy the economic value of the model.

Therefore, AI regulation should distinguish between:

"Explainability"

Why did the system produce this outcome?

"Auditability"

Can an authorised party verify how the system operates?

"Reproducibility"

Can another party reproduce the model?

"Source-code disclosure"

Can another party inspect the actual code?

These are not legally equivalent obligations.

12. Trade Secrets as a Legitimate Regulatory Consideration

Trade-secret protection should not be treated as merely an obstacle.

It can serve legitimate public-policy objectives.

Strong protection encourages firms to:

• develop innovative technologies;
• invest in R&D;
• maintain cybersecurity;
• develop proprietary AI models;
• experiment with new algorithms.

A competition regime that routinely forces disclosure of commercially valuable technology could unintentionally weaken innovation.

Therefore:

Competition law should prevent the misuse of secrecy, not eliminate legitimate secrecy.

13. When Mandatory Openness Is More Justified

Algorithmic disclosure becomes stronger where:

1. the undertaking possesses substantial market power;
2. the algorithm determines access to an important market;
3. competitors cannot realistically reproduce the relevant function;
4. there is credible evidence of exclusionary conduct;
5. algorithmic opacity prevents effective enforcement;
6. less intrusive investigative methods are inadequate;
7. disclosure is narrowly tailored;
8. strong confidentiality protections exist.

14. When Mandatory Openness Is Less Justified

Disclosure should generally be resisted where:

• the undertaking is not dominant;
• competitors have alternative technologies;
• the information is merely commercially useful rather than indispensable;
• the authority has alternative evidence;
• disclosure would reveal unrelated trade secrets;
• publication would create cybersecurity risks;
• disclosure would permit gaming;
• the requested information is disproportionate to the suspected infringement.

15. Algorithmic Openness as a Competition Remedy

Where an authority establishes anticompetitive conduct, possible remedies include:

Behavioural remedies

• disclose ranking criteria;
• prohibit discriminatory ranking;
• establish transparent access criteria;
• provide explanations for exclusion.

Technical remedies

• API access;
• interoperability;
• data portability;
• audit interfaces;
• logging requirements.

Monitoring remedies

• independent monitoring trustee;
• periodic algorithmic audits;
• regulator access;
• compulsory reporting.

Structural remedies

In exceptional cases, separation of:

• platform functions;
• ranking functions;
• marketplace operations;
• data infrastructure.

16. The "Minimum Necessary Disclosure" Principle

A sound regulatory framework should proceed through a hierarchy:

Level 1: outcome explanation
↓
Level 2: decision factors
↓
Level 3: technical documentation
↓
Level 4: regulator-controlled audit
↓
Level 5: confidential expert inspection
↓
Level 6: limited competitor access
↓
Level 7: public disclosure/source-code publication

The regulator should ordinarily stop at the lowest level capable of achieving the regulatory objective.

This is the most important safeguard against unnecessary destruction of trade-secret protection.

17. Competition-Law Significance

Mandatory algorithmic openness can address several modern competition problems.

Self-preferencing

Transparency can reveal whether a platform systematically advantages its own services.

Algorithmic discrimination

Audits can reveal whether competitors receive systematically inferior treatment.

Algorithmic collusion

Access to logs and model behaviour can help determine whether algorithms facilitate coordination.

Personalised pricing

Transparency can reveal whether pricing algorithms exploit data advantages.

Predatory or exclusionary strategies

Historical algorithmic records can demonstrate whether prices or rankings were manipulated to exclude rivals.

Data-driven dominance

Transparency can reveal whether superior data access creates a self-reinforcing competitive advantage.

18. Central Legal Tension

The central conflict can be expressed as:

Competition law requires sufficient transparency to make market power reviewable, while trade-secret law protects the very technological knowledge that may generate competitive advantage.

Neither absolute secrecy nor absolute openness is satisfactory.

Absolute secrecy can make powerful algorithmic systems effectively unaccountable.

Absolute openness can convert competition enforcement into compulsory technology transfer.

The better solution is targeted, confidential and proportionate transparency.

19. Case-Law Principles at a Glance

Conclusion

Mandatory openness of algorithms should not be equated with mandatory publication of source code. Competition law has a legitimate interest in examining algorithmic systems where they determine market access, ranking, pricing, interoperability or competitive conditions. However, trade-secret law protects genuine technological investment and innovation.

The strongest legal model is therefore proportionate algorithmic transparency:

Disclosure to the extent necessary for competition enforcement, protected to the extent necessary to preserve legitimate commercial secrecy.

In practice, independent audits, regulator access, confidentiality rings, algorithmic logs, technical documentation, controlled APIs and reasoned explanations will often be preferable to unrestricted publication of source code.

The jurisprudence of Microsoft, IMS Health, Bronner, Google Shopping, Google Android and Facebook collectively supports an important proposition: proprietary technology is not immune from competition-law scrutiny, but compulsory technological disclosure must remain exceptional, necessary and proportionate.