District Court of Amsterdam Prohibits Meta from Using Personal Data for AI Training – Private Enforcement of DMA?

by Sebastian Louven//

An interim injunction by the District Court of Amsterdam against Meta prohibits Meta from using the applicant’s personal data for training its AI models.¹ The decision is available in the Dutch judicial database.²

Context: Private Enforcement of the DMA

The decision adds a relevant layer to the ongoing legal discussion in the EU. It aligns with two earlier German decisions, also concerning Meta’s plan to train AI models using user data.

• In one case, the Cologne Higher Regional Court (OLG Köln) rejected the application for an interim measure on substantive grounds, holding that there was no data combination in regard to Article 5(2) DMA and was permissible under data protection law.

• In another case, the Schleswig Higher Regional Court (OLG Schleswig) rejected a similar application on procedural grounds, finding a lack of urgency because the request was filed only after Meta had already initiated the contested action. However, in an obiter dictum, the court noted potential violations of Article 9 GDPR.

The Amsterdam Decision – Procedural Framework

The Amsterdam decision was handed down on 19 May 2025 and published on 11 June 2025 – it took some time to reach the broader legal community. The application was filed on 30 April 2025, shortly after Meta’s announcement of its plan to train the AI with user data. It appears to be a small claims procedure, which Meta challenged as inappropriate. It remains unclear whether the decision is final.

The claimant was an individual and not a collective body. Thus, the ruling applies strictly to the individual case and does not offer precedent for public enforcement or collective redress actions.

The court denied a request for an English version of the judgment, stating that it lacked a legal basis for issuing such a translation.

The Court’s Injunction (translated)

The Subdistrict Judge:

Prohibits Meta from combining the personal data of [Claimant], or at least his “information”, between Facebook and/or Instagram and/or other Meta services used for Meta Ads, Meta AI, “AI tools for advertising materials” and/or for the (generation of) models, including those used in Meta AI and provided via an open platform, for the duration of the main proceedings (CV FORM 23-14577), under penalty of a fine of €100,000 for each violation of this prohibition;

Orders Meta to bear the costs of the proceedings, estimated at €100.00 for the claimant, plus post-judgment costs, including service fees;

Declares the judgment provisionally enforceable;

Rejects all other or further claims.

The Applicant’s Legal Arguments

The claimant sought to prohibit Meta from combining personal data – particularly data from Facebook and/or Instagram – with data from other Meta services for the purpose of training AI models. Crucially, the claimant invoked the need for express user consent under Article 5(2) DMA, arguing that an opt-out would not suffice.

Another key argument concerned the user journey: The opt-out could only be exercised by first agreeing to Meta’s terms and conditions, a step the applicant refused to take. Once again, it is clear that the user could not opt-out because he already did not accept the necessary terms and conditions. This pointed to a structural imbalance: Users cannot meaningfully object to data use without accepting the very terms they wish to contest. This highlights Meta’s continued control over the contractual framework.

Key Findings of the Court

The court found no procedural barrier to granting interim relief. The main proceedings were considered sufficiently related, as the claimant is also seeking clarification of the legal relationship with Meta regarding Facebook and Instagram usage. The court noted that the data combination process appeared irreversible, establishing the urgency required for interim measures.

Meta had been granted sufficient time to respond but focused solely on procedural objections. Notably, Meta had prior knowledge that the claimant opposed the data use for AI purposes and had explicitly objected – essentially a valid opt-out.

Substantively, the court accepted that training AI on user data constitutes a form of data combination, thereby falling under Article 5(2) DMA. Unlike the Cologne decision, the Dutch court did not attempt to reinterpret the provision narrowly.

This is indisputable. The scope of Article 5(2) DMA is broader than a simple aggregation of data about the same user. It targets the consolidation of data across services, even if those data concern different users. That is precisely the point: the competitive advantage lies in the aggregation of large, diverse datasets, not just personalisation. As such, each user has a right to object, regardless of whether the specific dataset relates to them or not.

Structural Dimensions

This aligns with the purpose and structure of the DMA:

• Article 13(3) DMA requires gatekeepers to ensure full and effective compliance with their obligations.

• Article 8(1) sentence 1 DMA states that compliance measures must effectively achieve the objectives of the regulation.

• Article 13(4) prohibits any circumvention, whether by contractual, commercial, technical, or design-based means.

• Article 13(6) forbids gatekeepers from hindering users in exercising their rights, particularly through interface design that undermines autonomy or free choice.

Meta’s opt-out process does precisely that – it effectively nullifies the requirement of prior consent by embedding it in a default-driven, opt-out-only mechanism. This contradicts the DMA’s regulatory logic.

Enforcement and Penalty

A striking part of the ruling concerns the penalty. Normally, this would not have been necessary, since Meta reportedly honors objections, according to the Dutch DPA. However, Meta’s procedural posture in this case led the court to conclude that Meta might not comply. This justified a conditional penalty of €100,000 per violation – reminiscent of the German concept of “Klaglosstellung”, where a party seeks to evade judgment by offering late-stage informal compliance.

Conclusion

This case highlights the practical relevance of private enforcement under the DMA. While limited to an individual case, the ruling confirms that courts across Member States may interpret AI training on user data as unlawful data combination. Meta’s reliance on opt-out mechanisms may increasingly come under pressure – especially where such structures are incompatible with the design-based fairness required under Articles 5 and 13 DMA.

More broadly, the ruling underscores the contestability logic at the heart of the DMA: gatekeepers cannot unilaterally dictate how users participate in the very ecosystems designed to serve them.

Dr. Sebastian Louven, Antitrust & TelCo Lawyer | Partner at louven.legal (Detmold, Germany)