The High Court’s recent judgment in Getty Images v Stability AI [2025] EWHC 2863 (Ch) marks one of the first major UK decisions to consider how copyright and trade mark law apply to artificial intelligence models trained on copyrighted content. The ruling provides helpful, but cautious, clarity for technology companies developing or deploying AI systems.
Case background
Getty Images claimed that Stability AI’s image-generation model, Stable Diffusion, had been trained on millions of Getty’s copyrighted images without authorisation. It alleged that Stability was importing and/or distributing the model in the UK with the knowledge of the infringement, (otherwise known as the “secondary copyright infringement”), and also brought trade mark and passing off claims relating to the appearance of Getty’s watermark in some AI-generated images.
The court’s findings
On copyright, the Court rejected Getty’s secondary infringement claim. It held that the Stable Diffusion model is not an “infringing copy” under the Copyright, Designs and Patents Act 1988. The model’s weights, being the model’s learned parameters, encode statistical information rather than reproducing the original images themselves. As such, importing or distributing the model in the UK did not constitute infringement.
Importantly, the Court did not rule on whether the process of training a model on copyrighted material, if done within the UK, would amount to direct infringement. Getty had withdrawn this part of its claim during proceedings, partly due to difficulties showing that the training had occurred in the UK. This leaves open key questions around the legality of AI training activities in domestic and international contexts.
The Court did, however, uphold parts of Getty’s trade mark claim. It found that the appearance of Getty’s watermark in certain AI-generated outputs could mislead users as to the source or endorsement of those images, constituting trade mark infringement. However, there was insufficient evidence that this occurred on a widespread or continuing basis, and the passing off claim was dismissed.
What this means for AI developers
The judgment offers measured reassurance to AI developers. Merely making available a trained model, provided it does not store or reproduce protected works, is unlikely to breach UK copyright law. Nonetheless, the decision does not provide a blanket exemption. Developers must remain alert to potential infringement risks in their training datasets, the jurisdictions in which training takes place, and the possibility of trade mark misuse through generated content.
Looking ahead
The case underscores the growing tension between innovation and intellectual property protection in the AI era. While the Court avoided expanding copyright liability to the underlying architecture of AI models, it left key issues unresolved particularly around data-mining exceptions and the legality of large-scale scraping for training. As AI adoption accelerates, legislative reform rather than litigation may ultimately be needed to provide clearer boundaries.
