German Court Exempts Meta AI Training Data From DMA Restrictions, Setting New Precedent for Tech Regulation

AI Training Exemption Shields Meta from DMA Data Combination Penalties

On May 23, 2025, the Higher Regional Court of Cologne ruled that Meta's use of partially de-identified data from two core platform services for AI training does not violate the Digital Markets Act (DMA) restrictions on data combination. This decision clarifies that AI development activities differ legally and technically from personalized advertising, allowing Meta to merge platform data for AI training without breaching DMA rules.

Key Points of the Ruling

• Case: 15 UKl 2/25, Higher Regional Court of Cologne
• Decision Date: May 23, 2025
• Finding: Incorporating partially de-identified, decomposed data into unstructured AI training datasets is not a prohibited data merger under Article 5(2)(1)(b) DMA.
• Scope: Applies to Meta's European market operations and sets precedent for AI training under DMA.

The court emphasized that neither the DMA nor the GDPR explicitly defines "data combination" in ways that restrict AI training practices. This stance contrasts with the Bundeskartellamt’s 2019 Facebook decision, which barred data merging across platforms without explicit user consent to prevent unfair competition advantages.

Legal and Regulatory Conflict

The Bundeskartellamt's longstanding position holds that dominant platforms like Meta cannot combine personal data from various services without explicit, voluntary user consent. Their 2019 decision found Facebook’s cross-platform data merging illegal under both competition and privacy laws, citing the risk of entrenching market dominance through data accumulation.

According to the regulator, Article 5(2)(1)(b) DMA applies broadly to all data combination purposes, including AI training. They argue that dominant companies cannot justify wide-ranging data processing practices by claiming performance benefits, as data protection rights are fundamental freedoms, not mere economic assets.

Conversely, the Cologne court distinguishes AI training from advertising, noting that AI development requires aggregated, partially de-identified data rather than individualized personal profiles. This technical distinction means AI training datasets inherently differ from the direct personal data combinations targeted by DMA provisions.

Regulatory Concerns on Data Combination

• Bundeskartellamt warns that gatekeepers’ access to cross-platform personal data can create unfair competitive advantages and raise barriers to market entry.
• They argue that even partially de-identified or decomposed data can yield insights that benefit dominant platforms disproportionately.
• Consent mechanisms are seen as insufficient when users face imbalanced power dynamics, making consent non-voluntary under GDPR.
• Allowing exemptions for AI training risks undermining the DMA’s core goal of limiting anticompetitive data accumulation.

Why the Court’s Ruling Matters

This ruling marks a significant divergence from regulatory enforcement in Germany, offering a new legal framework for evaluating AI data processing under competition law. It recognizes AI training as a distinct data activity requiring separate legal analysis from traditional data combination prohibitions in the DMA.

Technically, AI models learn from aggregated patterns in massive datasets rather than creating detailed individual user profiles. The court acknowledged that AI training involves embedding information into model weights, which does not preserve direct personal data in identifiable form.

For technology companies and legal professionals, this offers clearer guidance on how to handle data for AI development while staying within DMA compliance. It also signals that regulatory bodies and courts may need to refine their approaches to AI-related data practices as technology evolves.

Implications for Marketing and AI Development

Marketing technology providers can interpret this ruling as permission to develop AI-powered tools and optimization systems using aggregated data, provided they implement proper technical safeguards to prevent reconstructing personal profiles. The decision separates data use for personalized advertising—subject to strict DMA restrictions—from AI training, which requires large, diverse datasets for performance.

This clarity is especially relevant as European regulators increase scrutiny of gatekeepers under the DMA, including investigations into companies like Google and Meta on various compliance fronts.

Broader Regulatory Context

European data protection authorities are actively updating guidelines on AI and data use. For example, the Dutch Data Protection Authority has released GDPR preconditions for generative AI, and German authorities published comprehensive AI development guidelines in mid-2025.

Meta continues to face challenges regarding DMA compliance, including a formal appeal against a European Commission decision related to its ad-free subscription and personalized advertising offerings in Europe. The Court of Justice of the European Union’s 2023 Grand Chamber ruling supports subscription models as valid alternatives to consent for personalized advertising, which impacts Meta’s business model.

Timeline of Key Events

• May 23, 2025: Cologne court rules on Meta’s AI training data practices (case 15 UKl 2/25).
• July 4, 2023: European Court of Justice Grand Chamber supports subscription alternatives to personalized ads.
• March 2024: DMA obligations take effect for designated gatekeepers.
• April 2024: European Commission finds Meta’s European offering non-compliant with DMA.
• July 2, 2025: Meta files formal appeal against the Commission’s DMA decision.
• May 2025: Dutch Data Protection Authority publishes GDPR guidance for generative AI.
• June 2025: German authorities release comprehensive AI development guidelines.

Conclusion

The Higher Regional Court of Cologne’s ruling provides essential legal clarity on the intersection of DMA data combination rules and AI training. By distinguishing AI development data practices from personalized advertising data use, the court allows for continued innovation in AI while maintaining the DMA’s intent to prevent anticompetitive data hoarding.

Technology and legal professionals working on AI projects can leverage this precedent to better align data processing methods with European competition law. For practical AI training resources and further learning, explore comprehensive courses available at Complete AI Training.