First High Court Judgement In The UK On Copyright Claims Relating To Use Of Works In AI Training And Outputs

On 4 November 2025, the judgment of Getty Images v Stability AI was handed down by the High Court in a first of its kind decision.

The High Court held that Stability had not committed copyright infringement in training its AI on copyright materials outside of the UK, however Stability was found to have committed limited infringement of Getty's trade marks through the output of its generative AI Stable Diffusion. Helen Brain and Dylan Hooper examine the judgement and give their thoughts on whether legislation is keeping up to date with rapidly advancing cross-jurisdictional technology.

Case Summary

Getty Images is a well know content provider, licensing copyright works (predominantly photographs) throughout the world. Stability AI, a company founded in 2019, is the provider of Stable Diffusion a generative AI which produces synthesised images in response to user prompts. By Stability's own admission, its AI was trained in the US and that training had included exposing the AI to significant amounts of content. Getty alleged Stability had used millions of Getty images to train and develop the AI.

Although Getty initially argued that Stability AI had committed primary copyright infringement in the UK, Getty had to drop this claim due to lack of evidence that any training or other primary acts of infringement by Stability involving Getty owned copyright works had taken place in the UK. Because of the lack of evidence of acts infringing UK copyright, the only matters to be determined by the judge by the end of the trial were on the limited claims of secondary infringement of copyright, through bringing an infringing article into the UK, and of trade mark infringement due to reproduction of the Getty trade mark in watermarks visible in a limited subset of results created by users of the Stable Diffusion AI.

It is this lack of evidence and the limits on the claims eventually decided by the judge which really narrows the scope and importance of this judgment, though the judgment remains interesting for being the first judgment in the UK high court on claims by a copyright owner relating to use of works in training of AI and in outputs generated by AI.

Impact in the UK

One interpretation of the judgment that AI developers may be taking away is that they can reduce their risk of successful UK copyright claims by content owners if they:

1. Do not carry out any development or training activities in the UK using copyright works, and;
2. provided the AI offered in the UK also does not contain any of the copyright works, then the AI may be commercialised in the UK without committing any copyright infringement in the UK (but there could be claims in other jurisdiction).

This case could become a reference point for AI developers when testing their technology in the future, as they may choose to manage their risk in the UK by training their AI in jurisdictions with a more permissive intellectual property regime, then commercialise the AI in more IP friendly jurisdictions such as the UK (with key technical precautions in place to prevent the AI generating infringing outputs in response to user prompts).

The decision shows that the UK's intellectual property legislation needs updating to more clearly define what amounts to infringement in the face of advancing technologies.

While this decision stands it could potentially put rightsholders in the UK in a worse position regarding AI models trained outside the UK as they arguably have no mechanism to prevent them being commercialised in the UK if the AI provider can offer the AI in such a way that it also does not "contain" or generate outputs containing any third party IP rights. But that is effectively just saying that there were no infringing acts in the UK. Where companies have the resources, they may be able to bring claims in other jurisdictions in which the training occurred using their copyright works – but inpidual content creators are likely to find enforcement too costly.

As we've noted, the judgment did not ultimately decide whether the act of developing and training an AI model with existing works is copyright infringement. If Getty had been able to show in evidence that the AI had been trained in the UK, using Getty owned images, there may well have been a good primary copyright claim by Getty against Stability. A key point in the judgment clarifies that secondary infringement can be committed by the import of intangible (e.g., AI) as well as tangible goods, but again on the facts Getty's claim was not made out since the Stable Diffusion AI offered in the UK did not itself currently (nor had it in the past) actually contain any of Getty's copyright works.

Trade mark

This case found that Stability AI had infringed Getty Images' trade mark as their watermark was present in some of the content produced by Stable Diffusion. Because the Getty watermarks were generated and visible only on images produced in response to the users prompts, Stability tried to argue that the infringement was "performed" by the AI user themselves, not by Stability as the provider of the AI. The Court found that the training of the AI was Stability AI's responsibility and found against Stability's "blame the user" argument. This outcome is significant for AI developers as this judgment could mean that developers cannot escape liability for the outputs where trade marks belonging to others are reproduced as inpidual users are required to monitor and prevent infringing outputs generated by their system.

Conclusions

Getty Images will likely be taking some of the findings from the UK Courts' ruling for their case in the US against Stability AI. Tellingly, Getty have since the judgment was delivered reiterated requests for the British Government to help protect creators' work. Getty Images could potentially appeal the judgment, but it may well decide instead to focus on lobbying for legislative change and pursuing its claims against Stability AI in other jurisdictions.

Training and development of AI models using intellectual property belonging to others will no doubt lead to future litigation, and AI developers and rightsholders still require clarification on the issues left undecided by this case.

Helen Brain, Head of Commercial had this view on the judgement:

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.