The CJEU reference in Like Company v Google (C-250/25) is set to deliver the Court's first authoritative ruling on whether the training and output of generative AI systems engages copyright liability under EU law, and on the precise scope of the text and data mining exceptions introduced by the Digital Single Market Directive. Businesses and rights-holders operating across the European Union face genuine legal uncertainty in the interim: recent national decisions in Germany and the United Kingdom have reached different conclusions on key questions, and the Advocate General's opinion is not expected until September 2026. Understanding the current state of the law, and the choices available before the Court speaks, is therefore a strategic priority for any organisation whose commercial model touches AI-generated content, training data, or rights-managed material.
The Legal Framework: DSM Directive, TDM Exceptions and the Gaps
The principal legislative instrument governing AI training in the EU remains Directive (EU) 2019/790 on copyright in the Digital Single Market (the DSM Directive). Articles 3 and 4 of that Directive introduced two text and data mining (TDM) exceptions: a mandatory exception for research organisations and cultural heritage institutions (Article 3), and a broader general exception for any lawful user (Article 4) that rights-holders may opt out of by means of a machine-readable reservation. Those exceptions were transposed by Member States at different points following the Directive's 2021 implementation deadline, and the precise procedural requirements for an opt-out reservation vary across jurisdictions.
The AI Act (Regulation (EU) 2024/1689), which entered into force in August 2024 and is being applied in phases, separately requires providers of general-purpose AI models to document and publish summaries of training data and to comply with EU copyright law – including Article 4 opt-outs. That obligation does not resolve the underlying copyright questions; it assumes they will be answered elsewhere, presumably by courts including the CJEU. Organisations planning AI Act compliance in France and other Member States will find that copyright compliance forms an inseparable part of the broader regulatory exercise.
What the DSM Directive does not address expressly is: whether the generation of text, images or music that reproduces protected expression constitutes an independent act of reproduction; whether model weights themselves constitute an infringing copy; and how territorial rules apply when training occurs outside the EU but the output is consumed within it. These are precisely the questions that C-250/25 and the national cases described below are beginning to expose.
Recent National Decisions: Divergence Before Harmonisation
GEMA v OpenAI – LG München I (11 November 2025)
In its judgment of 11 November 2025, the Landgericht München I held that GPT-4 and GPT-4o had infringed the rights of song-lyric authors represented by GEMA by memorising and reproducing protected lyrics. The court found infringement of paragraphs 15, 16 and 19a of the German Copyright Act (UrhG). Critically, it held that the TDM exception under paragraph 44b UrhG – the German transposition of Article 4 of the DSM Directive – did not apply to the conduct in question. The court did not publish a fully reasoned explanation of its territorial analysis in the initial decision, leaving open how it would treat training that occurs on servers located outside Germany or the EU. OpenAI has lodged an appeal, and the outcome of that appeal will bear directly on the reasoning the appellate court applies to the opt-out mechanism and to the memorisation theory of liability.
A further decision by the same court, reported on 31 July 2026 and described as non-final at the time of writing, indicated that training on protected musical works may constitute infringement even where the training takes place outside the European Union, on the basis that the relevant act of exploitation occurs when the model is used within the EU. That finding, if confirmed on appeal, would have significant implications for the extraterritorial reach of EU copyright law and for the corporate structures that AI developers have adopted to date. It should be read as a reported, non-final indication only: it has not been confirmed by a higher court, and it sits in tension with general principles of EU private international law.
Getty Images v Stability AI – England and Wales High Court (4 November 2025)
In Getty Images v Stability AI [2025] EWHC 2863 (Ch), handed down on 4 November 2025, the High Court of England and Wales reached a notably different conclusion on one key point: model weights do not store copies of protected works in a form that engages the reproduction right. The secondary infringement claim therefore failed. The court did make limited findings in relation to trade mark rights, though those findings were narrow in scope. The decision is instructive for EU practitioners even though it applies UK law post-Brexit: it illustrates that courts in comparable common-law and civil-law systems can assess the same underlying technology and reach structurally different outcomes on questions of storage, reproduction and trade mark infringement.
The divergence between the München I approach (memorisation as reproduction) and the English court's approach (weights do not store copies) reflects an unresolved technical and legal question that the CJEU will need to address directly or indirectly in C-250/25. The answer matters enormously: if weights do not store protected expression, the entire theory of direct infringement at the training stage becomes significantly harder to sustain in court.
C-250/25 Like Company v Google: What the Court Is Being Asked
The reference in Like Company v Google (C-250/25) was formally before the CJEU for a hearing on 10 March 2026. The Advocate General's opinion is expected in September 2026, after which the Grand Chamber or a chamber of the Court will deliberate before delivering its judgment. The precise questions referred are not reproduced in full in publicly available pre-hearing documentation at the time of writing, but the case concerns the use of protected content in connection with Google's AI-related services, and the referring court has invited the CJEU to address the interaction between the reproduction right, the TDM exceptions, and the conditions under which an Article 4 opt-out reservation is effective.
The questions likely to be answered – whether explicitly or by implication – include the following:
- Whether automated ingestion of web-accessible material for AI training constitutes an act of reproduction within the meaning of Article 2 of the Information Society Directive (2001/29/EC), even where no human reads the material.
- Whether an Article 4 DSM Directive opt-out reservation expressed in a
robots.txtfile, terms of service, or technical metadata is legally sufficient to exclude the general TDM exception. - Whether AI-generated outputs that are substantially similar to protected training data infringe the reproduction right, the right of adaptation, or neither, in circumstances where the developer cannot demonstrate that the specific work was absent from the training corpus.
- Whether the territorial scope of EU copyright protection extends to training acts performed on infrastructure located in third countries where the output is distributed within the EU.
The Court's answers will bind national courts across all 27 Member States and will resolve, at least in principle, the divergence between decisions such as those of the LG München I and the English High Court's approach in Getty Images. The Commission and several Member States are expected to submit observations; ESMA and EBA, while not parties, will monitor the outcome closely given its implications for AI-generated financial information and model documentation requirements under EU financial regulation.
If your organisation is engaged in litigation or arbitration in the EU concerning AI-generated content, our IP litigation and arbitration practice can assist in framing arguments that account for the evolving CJEU jurisprudence. We are happy to discuss the specific procedural and substantive options available in your jurisdiction: contact us at info@ferrazwhitmore.com or visit our contacts page.
Practical Impact: Rights-Holders, AI Developers and Intermediaries
Rights-holders
Copyright owners – whether individual creators, collecting societies such as GEMA, or institutional rights-holders such as image libraries – currently face a fragmented enforcement landscape. The LG München I approach offers a route to relief based on memorisation and reproduction, but that theory depends on being able to demonstrate that a specific protected work was in the training corpus and was reproduced in output, which is technically demanding to establish. Design registration can assist brand owners and visual artists by creating registered rights that are arguably easier to enforce against specific outputs, though design infringement analysis for AI-generated images raises its own open questions.
Rights-holders should also audit their opt-out mechanisms. If the CJEU confirms that a machine-readable reservation is sufficient to exclude the Article 4 TDM exception, then the technical implementation of that reservation – and its date of publication relative to any alleged training crawl – will become significant evidence in litigation. Those who have not yet implemented structured opt-outs should do so promptly; those who have done so should verify that their reservations satisfy the requirements of each relevant Member State's transposition.
AI Developers and Deployers
Developers of general-purpose AI models training on EU-accessible or EU-hosted content face the most direct exposure. The LG München I decisions indicate that German courts are willing to apply broad theories of reproduction liability and to scrutinise whether the TDM exception applies on the specific facts. The reported 31 July 2026 non-final indication on extraterritoriality – that training outside the EU may still engage EU copyright where output is exploited within it – if ultimately confirmed, would significantly narrow the utility of offshore training structures as a risk-mitigation tool.
Deployers of third-party AI models (companies using API access to foundation models) also carry risk. Where a deployer customises or fine-tunes a model on proprietary or licensed data, it may itself become an author or a joint infringer depending on how the fine-tuning interacts with underlying training data. Supply chain due diligence – understanding the provenance of training data and the terms under which a model provider has licensed or otherwise obtained rights – is increasingly a standard component of AI procurement.
Intermediaries and Platform Operators
Platform operators that host AI-generated content or that provide AI generation tools to users must assess their position under Article 17 of the DSM Directive (upload filters and content-sharing service providers) alongside the general copyright framework. The CJEU's analysis in C-250/25 may touch on the relationship between Article 17 obligations and the use of protected content in AI training, though the core questions in that reference appear to focus on the training and output stages rather than on platform hosting obligations.
For further analysis of how the GEMA v OpenAI proceedings and the reported extraterritoriality finding interact with cross-border AI training structures, see our alert on GEMA, AI training and extraterritorial copyright in Germany.
Cross-Border Implications: Jurisdiction, Governing Law and Enforcement
The territorial dimension of AI copyright disputes is, in many respects, the most complex. EU copyright law governs acts of reproduction, communication and making available that occur within the EU, but AI training pipelines frequently involve data ingestion in one country, model training in another, and output delivery in a third. The reported LG München I indication of 31 July 2026 – that training outside the EU may still infringe where exploitation occurs within it – reflects an expansive reading of the lex loci protectionis principle: protection is claimed in the state where the protected expression is exploited, regardless of where the underlying act was performed.
That approach is not yet confirmed law and sits in some tension with the territoriality doctrine as generally understood under private international law. If the CJEU addresses territoriality in C-250/25, it will need to balance the legitimate interest of EU rights-holders in effective protection against the practical difficulty of applying EU copyright to acts performed by entities with no EU establishment. The European Commission has signalled awareness of this tension in its ongoing review of the DSM Directive implementation, though formal legislative proposals on AI and copyright had not been published at the time of writing.
For investors and multinationals with IP assets in multiple jurisdictions, the interaction between EU copyright law, UK copyright law post-Brexit (as illustrated by Getty Images v Stability AI), and the copyright frameworks of major training-data-producing countries creates a genuinely complex matrix. Trade mark rights add a further layer: both the CJEU reference and the English High Court's findings touch on trade mark issues, and the boundaries between copyright infringement, trade mark infringement and unfair competition in AI output cases remain contested across jurisdictions.
Strategic Recommendations: Acting Before the CJEU Speaks
The period between now and the expected Advocate General's opinion in September 2026 – and the subsequent CJEU judgment, which may follow in 2027 – is not a period in which businesses should take no action. The following steps are relevant depending on a party's position in the AI value chain:
- Audit training data provenance. Any organisation training or fine-tuning AI models on data that may include EU-protected works should document the source, the terms of access, and any opt-out reservations encountered. This documentation will be material in any future enforcement proceedings or regulatory inquiry.
- Implement and verify opt-out reservations. Rights-holders should ensure that their machine-readable opt-out reservations are in place and technically effective across all access points to their protected material, in anticipation of the CJEU confirming that a valid reservation excludes the Article 4 TDM exception.
- Review AI procurement contracts. Deployers should ensure that agreements with AI model providers include representations about copyright compliance, indemnities for third-party IP claims, and audit rights over training data documentation required under the AI Act.
- Monitor the appellate proceedings in GEMA v OpenAI. The outcome of OpenAI's appeal before the Oberlandesgericht München will clarify whether the LG München I's broad reading of the reproduction right and the narrow reading of the TDM exception survive higher-court scrutiny under German law, independently of the CJEU ruling.
- Consider design registration for visual AI outputs. Where organisations generate or commission AI-generated visual material for commercial use, design registration in relevant EU Member States can provide a registered right that is independent of the unresolved copyright questions and may be more straightforwardly enforced.
- Assess IP litigation strategy in light of territorial uncertainty. Rights-holders considering enforcement actions against AI developers should factor in the reported (non-final) extraterritoriality indication from LG München I and consider whether proceedings in multiple jurisdictions – or a coordinated cross-border strategy – better protects their position pending CJEU clarity.
If you are considering an enforcement action, structuring an AI compliance programme, or assessing the copyright exposure of an existing AI deployment in the EU, our team can assist with the analysis. Please reach out at info@ferrazwhitmore.com or through our contacts page.
Outlook: What the CJEU Judgment Could Change
The CJEU's eventual judgment in C-250/25 will be the most significant ruling on AI and copyright in EU legal history to date. Its impact will depend heavily on the precise questions answered and the breadth of the reasoning. A narrow ruling focused on the technical facts of the Like Company v Google reference may leave open many of the questions raised by the GEMA proceedings and by the broader debate about model weights and memorisation. A broader ruling that addresses the general framework for TDM opt-outs, the standard for reproduction in AI training, and the territorial scope of EU copyright would substantially reshape the compliance landscape for every organisation operating in this space.
Whatever the outcome, the judgment will not operate in isolation. National courts in Germany, France, the Netherlands and other Member States are already handling AI copyright disputes; the AG opinion in September 2026 will itself shift litigation strategy before the judgment is handed down. The Commission's review of the DSM Directive, combined with the phased application of the AI Act, ensures that the regulatory framework will continue to evolve in parallel with the case law. Practitioners and businesses should therefore treat the CJEU proceedings as one – albeit central – element of a broader, ongoing legal development rather than as a single event after which the position will be settled.
Ferraz & Whitmore monitors developments in CJEU proceedings, national copyright litigation and AI regulation across the EU on a continuing basis. If you would like to discuss how the evolving framework affects your organisation's specific position, please contact us at info@ferrazwhitmore.com or visit our contacts page to arrange a consultation.
Disclaimer: This publication is provided for informational purposes only and does not constitute legal advice. The information herein should not be relied upon as a substitute for professional legal counsel tailored to your specific circumstances. Ferraz & Whitmore assumes no liability for actions taken or not taken based on the contents of this material. For advice regarding your particular situation, please contact info@ferrazwhitmore.com.