The UK High Court case Getty Images v Stability AI was initially seen as a pivotal moment in defining how copyright law applies to artificial intelligence. Getty initially alleged that Stability AI unlawfully used its copyrighted images to train the Stable Diffusion model. However, the primary copyright and database rights claims were dropped due to jurisdictional and evidential challenges. The case now hinges on secondary copyright infringement, trade mark infringement, and passing off so it is likely that fewer key legal decisions will be made by the court. Nonetheless, the outcome on the secondary copyright infringement could still reshape legal standards for AI development and the legal status of AI-generated outputs. The trial ended at the end of June and a ruling is expected during the Autumn.
Background: The parties and the technology
Getty Images is a global leader in licensed stock photography. Stability AI is the UK-based developer of Stable Diffusion, an open-source generative AI model capable of producing images from text or image prompts. Getty alleged that Stability AI used millions of its copyrighted images to train Stable Diffusion without permission, and that the outputs generated by the model often resembled Getty’s content—including copying Getty’s watermarks.
Dropped claims: Primary copyright and database rights
Primary infringement due to training and developing the AI model
Getty initially claimed that Stability AI copied its images while training Stable Diffusion, breaching section 17 of the Copyright Designs and Patents Act 1988 (the Act) . However, the factual background was complex and it struggled to prove that any infringing acts occurred in the UK. Stability AI maintained that all training took place outside the UK, on non-UK cloud servers and had witness statements that backed up their position. As a result, this claim was dropped, not because it is bad in law but due to a lack of evidence as to where the AI model was trained.
Primary infringement by outputs
Getty alleged that the model’s generated images reproduced parts of its works, contrary to sections 16(2) and 20 of the Act by authorizing copying by end users or providing copies to the public. Stability‘s defence was that there was no substantial copying and/or that Getty did not own the original works and/or that any alleged infringements were the result of Getty using artificial prompts when using the model.
This claim faltered due to the nature of generative AI, which doesn’t produce identical images each time. Experts disagreed on whether “memorisation” had occurred or was likely to occur in practice (ie producing copies of the training works) and, in any event, Getty eventually dropped this claim after Stability agreed to block the disputed prompts which Getty claimed would produce infringing items.
Database rights infringement
Getty claimed its image collections were protected databases unlawfully used by Stability AI. These claims were considered “inherently linked” to the copyright claims and so were also withdrawn by Getty.
Remaining claims: Secondary copyright infringement
The most legally significant claim that remains is secondary copyright infringement, governed by sections 22 and 23 of the Act. This form of infringement focuses on acts like importing, possessing, or selling infringing articles. Getty’s claim is that Stability has imported the AI model into the UK which is an “article” under the Act and that Stability knows that it is an “infringing copy” because it was trained using Getty’s copyrighted images.
Novel legal questions
Is stable diffusion an “article” under the Act?
Stability AI contends that “article” only refers to a tangible object, and that intangible software or models don’t qualify. Getty counters that excluding intangible items would make a mockery of copyright law, especially in the digital age.
Is stable diffusion an “infringing copy under the Act?”?
Getty argues that the model was created through repeated exposure to its copyrighted images, which constitutes copying. Stability AI likens the situation to a USB stick that once contained infringing content but was later wiped clean-arguing that the model no longer retains the original images and thus cannot be infringing.
Did stability know the model was infringing?
Getty must also show that Stability AI had reason to believe the model was trained on infringing content. This may hinge on whether the presence of Getty watermarks or other indicators would have made the infringement obvious.
Implications for AI developers and copyright holders
This case is pivotal for several reasons:
- Territoriality of Copyright: The court’s decision on secondary copyright infringement will clarify whether AI models trained abroad but deployed in the UK can still infringe UK copyright law;
- Definition of “article” and “infringing copy”: If the court accepts Getty’s broader interpretation, it could open the door for more claims against AI developers whose models were trained on copyrighted content;
- AI developers need to ensure that no datasets are downloaded or stored in the UK and this message should be given to employees and consultants by clear written policies, with records kept to confirm this;
- AI developers should also note the importance of keeping clear and full documentation during AI model training to help show legal compliance and where the AI model was trained;
- In addition, AI developers should consider how they can block prompts which may lead to infringement risks by users;
- Potential claimants must consider which jurisdictions are best to bring an infringement claim;
- Disclosure requests need to be well thought through and, unlike Getty, pleadings must be amended promptly following the disclosure process;
- If Stability wins the secondary infringement claim, it could mean that in practice many UK SME copyright owners would not be able to enforce their copyright as they would be forced to start proceedings in foreign jurisdictions – expensive and risky;
- Potential claimants also need to ensure that they can prove that they own the works they allege have been infringed – this was problematic for Getty in the output claim;
- Legal Reform: The trial has intensified pressure on UK lawmakers to clarify how copyright law applies to AI with new rules relating to AI situations in the Act. A decision in favour of Getty could well accelerate future legislation, especially regarding text and data mining exceptions to make it easier to use copyrighted material to train AI models – the UK Government has made it clear that it wants to promote AI for economic growth.
How Moore Barlow can help
When you engage Moore Barlow’s Commercial and technology lawyers, you are working with a team recognised nationally for its technical excellence and commercially focused approach. We are consistently ranked by leading legal directories, including Chambers & Partners and The Legal 500, for our work in IT, intellectual property, and commercial law. These endorsements reflect not only our depth of expertise, but also our strong client relationships and ability to deliver practical, business-aligned legal solutions.
