The highly anticipated trial between Getty Images and Stability AI concluded on June 30, 2025. The case has gained national attention, as it represents a pivotal moment for the future of AI and copyright law in the UK.
On 16 January 2023, Getty Images brought proceedings against Stability AI, alleging that the AI company infringed Getty’s copyright by using millions of Getty’s images to train its generative AI model, Stable Diffusion, and that the outputs produced by the model reproduce substantial parts of these works.
The trial, which began on 9 June 2025, has attracted attention from both the technology and legal sectors, as it raises fundamental questions about how existing intellectual property laws should apply in view of modern generative AI systems.
In this article, Technical Assistant Egheosa Ogbomo and Partner Andrew White analyse the proceedings to date and the potential implications of any future developments.
Who are Stability AI and what is Stable Diffusion?
Stability AI is a UK-based artificial intelligence company which develops the family of Stable Diffusion AI models, open-source image generation tools capable of creating or altering images based on text or image prompts. Stability AI trained the original Stable Diffusion model on a subset of a dataset containing billions of images scraped from the internet.
Claims made by Getty
Copyright infringement claims
Getty alleges that Stability AI committed primary copyright infringement by reproducing substantial parts of millions of its images during Stable Diffusion’s training. This involved downloading, storing and augmenting them. They further claim infringement by making Stable Diffusion publicly available in the UK via Dream Studio and other open-source platforms, thus communicating significant parts of Getty’s works.
Getty alleges secondary infringement by authorising users to reproduce or communicate its works when outputs closely resemble Getty images. Secondary infringement due to the distribution of the trained model (an alleged infringing article) in the UK is another claim.
Trade mark infringement and passing off
In addition, Getty claims infringement of its trade marks, asserting that Stability AI used Getty’s marks without consent when generated outputs include Getty watermarks, causing confusion and exploitation of its reputation. They also allege passing off, arguing that generated images containing Getty logos misrepresent them as Getty-owned or licensed, implying endorsement.
Defences brought by Stability AI
Copyright infringement
Stability AI argues that any copying during data sourcing and training occurred entirely outside the UK, as they stored the datasets abroad. According to them, no infringing work was done in the UK. They claim output-stage infringements are the fault of users, since users control the input prompts and, in the case of image-prompts, the degree of input transformation.
Furthermore, they argue that any reproduced portions of Getty’s works are so minimal that they do not constitute a substantial part of copyrighted works and maintain that Stable Diffusion is not an infringing article. They may also rely on the pastiche defence, asserting that the extent of use of Getty’s work was necessary for pastiche, stating that this use does not affect the market for any originals.
Trade mark infringement and passing off
Stability AI claims that outputs containing Getty trade marks were only created through deliberate efforts by Getty’s legal team and do not reflect its normal commercial use. Furthermore, Stability AI denies any likelihood of confusion or unfair advantage. For passing off, it argues any misrepresentation arises from user actions and that outputs are not sufficiently similar to Getty’s works to mislead the public.
Key case developments so far
Since the trial began, Getty Images have dropped their claims for primary copyright infringement, citing a lack of evidence and knowledgeable witnesses to support the allegations. They also dropped their claims over the model’s training and development, maintaining that infringing acts occurred but that there were no witnesses from Stability AI who could provide clear and comprehensive evidence about the entire training process.
For the claims concerning AI-generated outputs, Getty stated that Stability AI has implemented measures preventing the reproduction of infringing outputs. Getty has been unable to establish that any outputs produced by the models reflect a substantial part of its protected images.
This illustrates the difficulty in proving exactly where the training of an ML model takes place for the purposes of determining infringement, as well as demonstrating that AI-generated images reproduce a ‘substantial part’ of protected original works. While this has narrowed the scope of the dispute, Getty’s claims of secondary infringement and trade mark misuse remain in contention.
What are the implications for IP law?
There remains significant uncertainty over the balance of power between AI developers and content creators on copyright licensing. A win for Stability AI in these proceedings could reduce the incentive for AI developers to seek licences in the UK and may lead some to continue developing models without securing permissions, or to do so outside of the UK. This has prompted questions about whether legislative reforms are needed to address potential gaps in protection. However, extending UK copyright law to cover acts abroad could create conflicts with foreign regimes such as US fair use, risking the UK’s attractiveness for AI research and model development.
There has already been a significant increase in the number of AI-related patent applications globally, with the European Patent Office observing a 45-fold increase in the annual number of AI-related European patent filings since 2015. How UK courts interpret copyright law for AI training and outputs could influence the approach to patent applications for AI-related inventions in the UK. Uncertainty over data use rights may consequently affect the development, disclosure and protection strategies for new AI technologies. Companies already face rising costs of data collection, particularly as new technologies such as Cloudflare’s tool will allow website owners to charge fees for access by web-scraping tools.
A final judgement on the remaining claims is likely to be handed down in the next few months. This will set an eagerly awaited precedent for how UK infringement laws should be interpreted in the age of AI, and large-scale text and data mining.
If you have any questions as to how the outcome of these proceedings may impact your IP strategy, please reach out to your usual contact at Mathys & Squire, or get in touch through a general enquiry and we would be happy to help.
