Navigating Copyright in the Age of Generative AI: EU, French, and UK Developments and Approaches

Detailed Analysis

The EU Resolution and the UK Report address several overlapping themes at the intersection of copyright law and generative AI. The following comparative analysis highlights both convergences and significant divergences between the two approaches:

The Copyright Framework for AI Training—Divergent Philosophies

• The EU Resolution considers that the current copyright legislation, including the Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC (the "CDSM Directive"), is insufficient to address the challenges of licensing copyrighted material for generative AI and calls on the Commission to urgently propose a supplementary legal framework (paragraphs 1 and 2 of the Resolution).
• The UK Report, by contrast, takes a more restrained stance. The UK Government rejected its own earlier preferred option of introducing a broad copyright exception with opt-out, following strong opposition from the creative industries. Instead, it proposes to gather further evidence on how copyright laws are impacting AI development, consider alternative targeted approaches, and monitor developments in litigation and international contexts before legislating. In short, the EU demands new rules now, while the UK prefers to wait and see.

Transparency Obligations—Mandatory vs. Voluntary

• The EU Resolution insists on full mandatory transparency and source documentation regarding the use of copyrighted works by providers and deployers of general-purpose AI models placed on the EU market, including detailed lists of each copyrighted work used for training, inference, retrieval-augmented generation ("RAG"), or fine-tuning (paragraphs 12 and 13 of the Resolution). It further requires crawlers to identify themselves and AI providers to maintain detailed records of their scraping activities, and establishes that mere information about third-party content used by AI providers does not constitute a trade secret under EU law.
• The UK Report, while recognizing strong stakeholder support for transparency (over 90% of consultation respondents favored disclosure), stops short of introducing mandatory requirements. Instead, the UK Report proposes to work with industry and experts to develop best practice on input transparency to help rights holders assert their rights, while continuing to monitor the effects of transparency rules in other countries, including the EU AI Act. The UK also explicitly proposes not to create any new regulator to oversee transparency compliance and not to impose regulatory duties on existing regulators at this time.

Rights Holders' Control, Opt-Out and Licensing

• The EU Resolution reaffirms that rights holders must have full control over the digital use of their content by AI systems for training purposes, supported by full transparency and an effective ability to exclude such use. It proposes that the EUIPO serve as a trusted intermediary to manage and catalogue opt-outs, and calls for voluntary sector-by-sector collective licensing agreements to create an operational licensing market with fair remuneration, firmly opposing any blanket licence in exchange for a flat-rate payment (paragraphs 5, 9, 10, 19, 20, and 21 of the EU Resolution).
• The UK Report, while sharing the principle that rights holders should be fairly remunerated, takes a markedly less interventionist stance. It proposes not to intervene in the licensing market at this stage, citing insufficient evidence to justify government intervention. The UK relies on the existing copyright framework and the emerging voluntary licensing market, noting encouraging trends in licensing deals globally. The UK also rejected the opt-out exception model (similar to the EU's Article 4 CDSM Directive approach), finding that both creative industries and AI developers had significant concerns about its practical implementation.

Territorial Application and Models Trained Abroad

• The EU Resolution affirms that EU copyright law should apply to all generative AI models and systems placed on the EU market, regardless of the jurisdiction in which training took place, and that non-compliant models should be prohibited from the EU market (paragraph 16 of the Resolution). This represents a strong extraterritorial assertion.
• The UK Report acknowledges the issue—recognizing that most frontier AI models are trained outside the UK, primarily in the USA—but proposes a significantly more cautious approach. It declines to amend copyright law in respect of systems developed outside the UK at this time, instead relying on the existing law as interpreted in the Getty Images v Stability AI judgment (where the primary copyright infringement claims relating to AI training were withdrawn at trial due to insufficient evidence that the relevant training activity had taken place in the UK; suggesting that AI training conducted overseas (as is usually the case in practice) would be unlikely to infringe UK copyright law). The UK proposes to keep this issue under review, engage bilaterally with other countries, and work through forums such as the World Intellectual Property Organization ("WIPO") to support closer international alignment.

Enforcement and Presumptions

• The EU Resolution recommends a rebuttable presumption that, for any generative AI model placed on the EU market, copyrighted works have been used for training when transparency obligations have not been fully complied with. When a rights holder prevails on this basis, all reasonable and proportionate costs should be borne by the AI provider (paragraph 24 of the Resolution). This represents a significant shift in the burden of proof in favor of rights holders.
• A notable national development in this area is the French Bill adopted by the Senate on 8 April 2026, which would insert a new Article L. 331-4-1 into the French Intellectual Property Code. Under this provision, in any civil proceedings, a work protected by copyright or related rights is presumed to have been used by an AI provider, provided that an indication relating to the development, deployment, or output of the AI system makes such use plausible. Unlike the EU Resolution's presumption—which is triggered by non-compliance with transparency obligations—the French presumption would apply more broadly, requiring only a plausible indication of use. The French Bill would also apply to proceedings pending at the date of its entry into force. This text remains subject to the parliamentary process and has not yet been finally adopted.
• The UK Report does not propose any equivalent presumption. Instead, it relies on the existing civil enforcement framework, which the UK Government considers to be generally effective and capable of adapting to AI-related challenges. The UK proposes further work to identify and address enforcement barriers, including considering the likely effects on SMEs and individual creators. It also notes the potential role of the Getty Images v Stability AI appeal in clarifying the law on secondary infringement for AI models trained abroad. Notably, the UK explicitly proposes that no new regulator should be created specifically to oversee AI-copyright matters.

AI-Generated Content and Computer-Generated Works

• The EU Resolution reaffirms that wholly AI-generated content that does not meet the established criteria for copyright protection should be ineligible for such protection, with its public domain status clearly determined (paragraph 25 of the Resolution). It also calls for clear labelling of exclusively AI-generated content and measures against deepfakes (paragraphs 27 and 28 of the Resolution).
• The UK Report reaches a remarkably similar conclusion on substance: it proposes to remove the UK's unique copyright protection for wholly computer-generated works ("CGWs"), which has existed since 1988, finding minimal evidence that it is being actively used or has material economic impact. However, the UK again diverges on approach: while the EU calls for mandatory labelling requirements, the UK proposes only to work with industry to explore best practice on labelling AI-generated content, continuing to monitor international developments rather than legislating.

Digital Replicas and Deepfakes

• The EU Resolution calls on the Commission to examine measures to protect individuals against AI-generated deepfakes imitating their personal characteristics without consent, and calls on digital service providers to act against such illegal use (paragraphs 27 and 28 of the Resolution).
• While the UK Government agrees that existing protections—a patchwork of IP, data protection, online safety, and criminal law—may not be sufficient, it stops short of proposing specific legislation. Instead, it proposes to explore a range of options, including the potential introduction of a new personality or image right, while considering the complex interplay between IP, privacy, and other areas of law. The government has stated that it will "launch a consultation in the summer to seek views on how we address these harms, while protecting legitimate innovation."

Conclusion

For stakeholders in the AI and creative industries, the practical takeaway is clear: the EU standard is likely to become the baseline for any company seeking access to the European market, while the UK's evolving position offers both flexibility and uncertainty. Meanwhile, national initiatives such as the French Bill demonstrate that EU Member States may not wait for EU-level legislation and could adopt their own rules on presumptions of use, further complicating the compliance landscape for AI providers. The coming months will be critical, as the EC responds to the Resolution with legislative proposals, the UK Getty Images appeal concludes (although a decision is not expected until 2027), and international discussions at WIPO and beyond continue to shape the global framework. Companies and creators should prepare for a world in which copyright compliance in the AI context is not a single standard but a multijurisdictional puzzle requiring careful, jurisdiction-specific strategies.