Do European copyright rules also apply to large non-European generative AI models?

📢 Yes, as soon as these models are offered, marketed or made available in the EU.

This is the key conclusion of a report by the CSPLA (Conseil supérieur de la propriété littéraire et artistique, an advisory body to the French Ministry of Culture on copyright issues), written by Tristan Azzi, professor at the École de droit de la Sorbonne, and Yves El Hage, Maître de conférences at the Université Jean Moulin Lyon 3. This approach was recently confirmed in German law by a ruling of the Munich Higher Regional Court (OLG München 11 November 2025, Ref. 42 O 14139/24, GEMA v OpenAI), according to which the operator of a large generative AI model accessible in the EU is directly liable, regardless of the location of the training or the company's place of business.

🔍 Why?

• The location of servers or training operations outside the EU is legally irrelevant. The CJEU has already rejected this criterion in copyright infringement cases (CJEU, 19 Apr. 2012, Case C-523/10, Wintersteiger; and CJEU, 27 Sept. 2017, Joined Cases C-24/16 and C-25/16, Nintendo v BigBen). • Generative AI operates as a global and indivisible process: no output without input, no data collection without results.

• In the case of complex offences, the law of the place of exploitation is decisive. If access or use takes place in the EU, European law applies.

⚖️ A framework reinforced by the EU AI Act

Recital 106 requires providers of generative AI models to implement a strategy to comply with EU copyright law, thereby supporting this approach.

📌 Practical consequence

National courts must apply European copyright law to generative AI models when they are accessible in the EU—particularly with regard to:

• exclusive rights,

• the opt-out mechanism provided for under the 2019 Directive on text and data mining.

📄 Read the report (Ministry of Culture): urlr.me/4UtWfz

  Image : Pixabay