Case background
Getty Images – the Seattle-based stock photo agency – sued Stability AI in London, alleging its Stable Diffusion image generator had been “trained” on Getty’s copyrighted works without permission. Getty claimed that roughly 12 million Getty images were copied (or "scraped") from its websites to train the AI model. The suit asserted three main infringements: that copying Getty’s images for training violated copyright; that certain AI-generated outputs allegedly reproduced Getty images (and even Getty's watermarks); and that distributing the Stable Diffusion model in the UK (via online downloads) amounted to secondary infringement. Stability AI countered that all training took place outside the UK, so UK law did not apply, and argued that any infringement (if at all) would be attributable to end users or was protected as fair dealing.
High Court ruling
With Getty having withdrawn its primary copyright allegations, Justice Joanna Smith’s judgment of 4 November 2025 considered only the remaining issues of secondary infringement and trade marks. The court held that Stable Diffusion does not infringe Getty’s copyright, since it “does not store or reproduce any Copyright Works (and has never done so)”. In other words, the AI model itself could not be an infringing “copy” under UK law. The only partial success for Getty was on trade marks: some Stable Diffusion outputs containing Getty’s watermark infringed Getty’s registered marks. Even then, the judgment emphasised this was “both historic and extremely limited in scope”. The findings related exclusively to a very small number of outputs generated by early versions of Stable Diffusion; there was no evidence before the Court that the current versions of the system produce such watermarked images. All other claims (including a passing-off claim) were either withdrawn before trial or did not succeed at trial.
Implications and next steps
The ruling is widely seen as a mixed victory. AI companies hailed it as clarification that model training (outside the UK) does not, by itself, breach UK copyright. Getty and its supporters view it as a warning: “copyright owners’ exclusive right to reap what they have sown has been avoided on a technicality,” as one lawyer noted. Commentators agree the overarching legal question – can using third party works to teach an AI model infringe copyrights – remains unanswered. As Justice Smith noted, the court could only decide the “diminished” case that remained and could not consider any related issues.
In practice, the decision puts further pressure on policymakers. In a post-judgment statement, Getty urged governments to impose stronger transparency rules for AI training data. The case will proceed on other fronts – Getty has a related lawsuit pending in the U.S. – but for UK IP law there are now (narrow) precedents:
Where AI-generated outputs reproduce registered trade marks, conventional principles of trade mark infringement will apply.
By contrast, if an intangible AI model does not store or reproduce protected creative works (or databases), both copyright and database infringement claims are likely to face significant hurdles; mere use of such materials for training outside the UK, alone, will not be sufficient.
Businesses and creators will be closely watching this shifting and uncertain legal landscape. Recent debates on the UK Data (Use and Access) Bill (the ‘AI Bill’) – including repeated government defeats in the House of Lords over AI data scraping – underline how contested this question now is. With Parliament, Government and the courts sending slightly different signals, the UK position on AI training and scraping remains fluid and unpredictable.
Key takeaways
- A win for Stability AI. Getty Images dropped its core copyright claims mid-trial, and the High Court ruled that Stable Diffusion’s AI “does not store or reproduce any copyright works”. In short, Getty’s attempt to treat the AI model itself as an infringing copy was rejected.
- The court found limited trade mark infringement: some AI-generated images carrying Getty’s watermark breached Getty’s marks. However, Justice Smith stressed these findings were “historic and extremely limited in scope” – affecting only early versions of the model’s outputs. All other claims (including passing off and a secondary copyright claim) were dismissed or abandoned.
- This was not the landmark case to clarify the law that many were hoping for. With Getty’s primary claims withdrawn, the court was unable to rule on whether using copyrighted images to train an AI model within the UK infringes copyright. Commentators are calling for (and continue to await) clarity from the Courts and/or Parliament.
For responsive and clear advice on where English law on generative AI and copyright looks likely(!) to settle, and what that might mean for your organisation, please contact our experts.
