Jude Copeland

On 14 January 2025, Mrs Justice Joanna Smith DBE delivered a reserved judgment in Getty Images (US) Inc and Ors v Stability AI Ltd [2025] EWHC 38 (Ch), offering a glimpse into the critical issues at the heart of the UK’s most significant legal dispute involving artificial intelligence (AI) and intellectual property (IP).

This procedural ruling, while not a final determination, highlights the complexities of managing large-scale copyright claims in the context of generative AI (GenAI).

Background of the litigation

In January 2023, Getty Images, a leading provider of visual and digital media content, initiated legal proceedings against Stability AI. Getty alleges that Stability AI unlawfully “scraped” 12 million photographs, videos, and illustrations from its website, over half of which are original copyright-protected works, to train its text-to-image GenAI model, Stable Diffusion.

Getty claims that this unauthorised use of its material infringed copyright and that the synthetic images generated by Stable Diffusion further infringe its rights as they reproduce substantial parts of its works. Stability AI has admitted that “at least some images” from Getty’s website were used during training but has not disclosed the specific images or their scope.

The judgment

The judgment arose from an application made by Stability AI during a Case Management Conference held in November 2024. Stability AI sought to exclude Thomas M Barwick Inc, the sixth claimant, from acting as a representative for a class of copyright owners whose works had been exclusively licensed to Getty.

Key issues: representative claim and class definition

The court addressed whether the sixth claimant had sufficiently similar interests to the represented parties and whether the class was adequately defined. Mrs Justice Smith found the proposed class definition impermissible as it relied on determining whether copyright infringement had occurred, a key issue to be resolved in the litigation.

The claimants’ alternative proposal, which sought to proceed without joining all exclusive licensors, was also rejected. The court held that the claimants failed to provide sufficient evidence to ensure that Stability AI would not face future claims from unrepresented licensors. Consequently, the representative claim was dismissed.

The court encouraged the claimants to refine their approach by either submitting additional evidence to support a representative claim or narrowing the scope of the class to facilitate effective case management.

Evidential challenges in identifying infringed works

The court acknowledged significant evidential hurdles in identifying the copyrighted works allegedly used to train Stable Diffusion. Stability AI holds knowledge of its training datasets but has yet to disclose which or how many Getty images were used.

To address these challenges, Stability AI proposed resolving issues of authorship, subsistence, and infringement through sampling, a process where a subset of works would be analysed, and findings extrapolated. However, no consensus has been reached on implementing this approach, leaving critical evidentiary issues unresolved.

Technical and practical complexities

Mrs Justice Smith highlighted the immense practical difficulties of interrogating datasets of this scale. Stability AI argued, and the court agreed, that attempting to identify specific copyrighted works within a vast training dataset would be “wholly disproportionate and practically impossible without significant resources.”

The nascent nature of many AI businesses compounds these challenges. Smaller companies may lack the infrastructure or resources to comply with extensive data interrogation and sampling orders, especially in cases alleging millions of infringements.

With less than five months until the trial on liability, scheduled for 9 June 2025, the sampling and extrapolation approach appears to be the most pragmatic solution. However, this remains untested in the UK legal system and could set a significant procedural precedent.

Significance of the proceedings

These proceedings are critical for the future of both AI development and copyright law. A final judgment will likely address fundamental legal and ethical questions regarding the use of copyright-protected material to train GenAI models, particularly in the absence of rights-holder consent.

The outcome will have wide-reaching implications:

1. For AI Firms: The decision could prompt AI developers to reassess their data utilisation practices, including securing permissions from rights-holders and improving transparency in data use. Companies operating in England and Wales may need to invest in more robust compliance systems or reconsider their operations in this jurisdiction.

2. For Creators: The ruling may redefine how creators protect and monetise their works, encouraging stronger licensing frameworks and more vigilant monitoring of how their content is used in AI training.

Conclusion

The procedural judgment in Getty Images v Stability AI underscores the challenges of managing copyright disputes in the context of generative AI. While the substantive issues remain unresolved, the case demonstrates the UK courts’ efforts to balance the interests of rights-holders, AI developers, and broader technological innovation.

As the trial approaches, key questions surrounding the legality of AI training practices and the evidentiary standards required in such disputes will take centre stage. The final judgment will undoubtedly shape the intersection of AI and intellectual property law, establishing precedents with profound implications for both industries.

• Jude Copeland is a legal technology specialist in Belfast-based Cleaver Fulton Rankin’s legal technology group.