On 15 December 2025, the French High Council for Literary and Artistic Property (Conseil supérieur de la propriété littéraire et artistique - CSPLA), an advisory body attached to the French Ministry of Culture, published a mission report addressing a core yet largely unresolved issue: the determination of the law applicable, under private international law rules, to generative artificial intelligence models commercialised within the European Union.
Chaired by Professor Tristan Azzi and reported by Yves El Hage, the report comes at a time of growing legal uncertainty surrounding the use of protected works to train generative AI models and the cross-border dissemination of AI-generated outputs, in a context marked by diverging national copyright regimes worldwide.
The report clearly identifies that, from a private international law perspective, the most complex issues arise not from AI outputs, but from the input phase, namely the reproduction of protected works and content for the purpose of training generative AI models.
Such training operations are inherently transnational: data may be collected, stored and processed across multiple jurisdictions, while the resulting models are commercialised and accessed by users located in different countries.
In this setting, the traditional application of the lex loci protectionis, the law of the country for which protection is claimed, raises significant difficulties, particularly where acts of reproduction are technically fragmented and legally difficult to localise.
The report expressly dismisses any approach based solely on technical criteria, such as the physical location of servers, data centres or copying operations.
Such an approach is considered unpredictable, unstable and ultimately unworkable, notably because of the multiplicity of infrastructures involved and the existence of so-called "intermediate inputs" occurring at various stages of a model’s lifecycle.
Instead, the report advocates a global legal analysis of the exploitation process. It emphasises the inseparable link between input and output, since training operations only make sense in light of the results generated for end users. This configuration is analysed through the well-established private international law concept of a “complex tort”, involving a chain of causally connected events occurring in different locations.
By characterising training and generation as two ends of a single exploitation process, the report favours an "ex post" localisation of the complex tort.
As a result, the law of the country where the damage materialises, understood as the country in which users access AI-generated results, should in principle apply.
In practical terms, this leads the authors to prioritise the law of the country where the AI model is commercialised, including for the assessment of the lawfulness of upstream reproductions carried out during the training phase. This solution is presented as the most coherent, foreseeable and consistent with established private international law principles governing online infringements of intellectual property rights.
The second part of the report examines the impact of the EU Artificial Intelligence Act of 13 June 2024 on conflict-of-laws analysis.
Although the Regulation does not formally establish a specific conflict rule, Recital 106 plays a decisive role: it requires providers of general-purpose AI models placed on the EU market to comply with EU copyright and related rights legislation regardless of the jurisdiction in which the underlying training acts take place.
The report concludes that this requirement necessarily influences private international law solutions, either through the characterisation of EU copyright rules as overriding mandatory provisions, or through the application of the international public policy exception, where the application of a foreign law would deprive right holders of any effective means of protection.
Conclusion
Through this report, the French CSPLA puts forward a structured and ambitious interpretation of the private international law framework applicable to generative AI. It clearly affirms that, where an AI model is commercialised within the European Union, national courts are strongly encouraged to apply EU copyright standards, even in respect of training acts performed outside the EU.
As such, this report constitutes a key reference for right holders, AI developers and legal practitioners alike, offering a coherent legal framework for an issue that is likely to become central in future AI-related litigation.
