On 23 September 2026 the Court of Justice of the European Union received a request for a preliminary ruling (C‑80626) concerning the scope of the text‑and‑data‑mining exception in the Copyright Directive. The request asks the Court to interpret whether the exception covers commercial training of large‑language‑model systems, prompt‑driven retrieval and AI‑generated news summaries. The immediate effect is that AI developers and press publishers must await clarification on the legality of using copyrighted news content for AI training. The ruling will delineate the limits of lawful TDM for generative‑AI services across the EU.

Full News Breakdown

Hungarian courts referred a dispute between a consortium of national press organisations and Google over the Gemini chatbot’s use of news articles. The press claim that Gemini’s training and summarisation infringe their exclusive rights, while Google argues the activity falls within the TDM exception. The national courts therefore lodged a request for a preliminary ruling with the CJEU (C‑80626).

• Case Name: Request for Preliminary Ruling concerning Google Gemini (C‑80626)

• Court: Court of Justice of the European Union

• Date: 23 September 2026

• Citation: Request for Preliminary Ruling.pdf (C‑80626)

• EU Instruments / UK Legislation Cited: Directive 2003/98/EC (Copyright Directive); Copyright, Designs and Patents Act 1988

• Key Provisions: Article 3(2) of the Copyright Directive; CDPA s 30(1) (fair dealing for quotation)

• Primary Legal Issue: Whether the TDM exception shields commercial AI training, prompt‑driven retrieval and AI‑generated news summaries.

• Applicant Arguments: Press publishers argue that Gemini’s training, excerpt extraction via prompts and summary generation constitute copying that is neither non‑commercial nor for scientific research, thus outside the exception.

• Respondent Arguments: Google contends that systematic text mining for AI model development is a lawful purpose under the exception and that generated summaries are transformative uses.

How Does This Affect You?

Before the referral, practitioners faced uncertainty about whether commercial AI systems could rely on the TDM exception for training on copyrighted news material. The Court’s forthcoming interpretation will clarify the meaning of “non‑commercial” and “scientific research” in the AI context. Consequently, the risk profile for AI developers and press organisations will shift from speculative litigation to a defined legal baseline.

For Lawyers & Advocates

• Review all ongoing AI‑training contracts to determine whether the data‑source clause can be re‑characterised as a lawful TDM activity under Article 3(2).

• Amend standard licence templates with a clause that expressly disclaims reliance on the exception unless the Court confirms its applicability to commercial model development.

• Update due‑diligence checklists for M&A involving AI firms to include a verification step on the provenance of training corpora against the clarified exception criteria.

• Prepare evidential packages that demonstrate a “scientific research” purpose, such as internal research‑project documentation, to pre‑empt infringement claims where the exception may apply.

• Advise clients that, while the ruling may narrow the exception, it does not eliminate the possibility of invoking the transformative‑use argument in infringement disputes, allowing the decision to be used as persuasive authority in national courts.

For Law Students

The case illustrates how EU courts balance the free‑flow of data for innovation against the exclusive rights of copyright holders.
The core doctrine concerns the interpretation of the “non‑commercial” and “scientific research” qualifiers in the TDM exception.
The decision is particularly relevant for the study of:

• EU copyright law and the Copyright Directive

• Text‑and‑data‑mining jurisprudence

• Exceptions and limitations in intellectual property regimes

• Comparative analysis of EU and UK copyright exceptions

• The impact of AI on traditional copyright doctrines

Comparable cases are Infopaq International A‑/S v. Danske Dagblades Forening (C‑5/08, 2009) and Publicité Générale S.A. v. Google Inc. (C‑131/12, 2014); contrasting them with this ruling highlights how the Court distinguishes between incidental excerpting and systematic data mining for AI purposes.

For Businesses

• AI‑technology firms must audit their data‑ingestion pipelines and, if the exception is narrowed, negotiate blanket licences with news aggregators before launching new model versions.

• Digital news aggregators should revise internal policies on automated summarisation tools, ensuring that any AI‑generated excerpt complies with the clarified exception or is covered by a licence.

• Media companies with extensive digital archives need to decide whether to offer a voluntary licence for AI training now, rather than wait for the Court’s interpretation, to preserve bargaining power.

• Boards of directors should commission a risk‑assessment report on the financial exposure arising from potential infringement claims linked to AI‑generated content, updating the company’s IP risk register accordingly.

Key Takeaways

• The Court will define the reach of the TDM exception for commercial AI training, limiting reliance on the provision where “non‑commercial” or “scientific research” criteria are not met.

• Practitioners must revise licensing strategies, due‑diligence protocols and contractual clauses to reflect the clarified exception parameters.

• Regulators and national courts will be able to enforce copyright claims against AI developers more confidently, but may still consider transformative‑use arguments where appropriate.

• Monitor the European Parliament’s upcoming review of the Copyright Directive scheduled for early 2027, which may amend the exception language in response to the Court’s interpretation.

• In‑house counsel should update AI‑governance policies before the end of Q4 2026 to align with the anticipated ruling and avoid contractual breaches.

Source: Google's Gemini faces new EU copyright questions over AI training, summaries