Getty Images V Stability AI – Court Considers AI Training And Copyright

The High Court of England and Wales has delivered one of the first major decisions on the use of copyrighted material to train artificial intelligence models.

In Getty Images v Stability AI*, the Court dismissed Getty Images' copyright claims relating to the training of the Stable Diffusion AI model (the AI Model), but upheld a limited finding of trade mark infringement.

Although based on UK law, the decision adds an important perspective to the growing debate around how copyright and trade mark principles apply to AI training and AI generated outputs.

What happened

Getty Images alleged that Stability AI used millions of Getty Images' photographs, metadata and captions to train the AI Modelwithout a licence. Getty Images pursued direct and secondary copyright infringement claims, as well as trade mark infringement.

Direct infringement and location of training

Initially, Getty Images' argument was that Stability AI reproduced its photographs without permission in the course of training the AI Model. Stability AI successfully argued that the training was carried out on computers located in the United States rather than the United Kingdom. Accordingly, the Court found that UK copyright law did not apply to Stability AI's training activities and the Court had no jurisdiction to consider the direct infringement claim.

Secondary copyright infringement

Getty Images also argued that Stability AI had imported unlawful copies of its images into the UK by making the AI Model available to UK users. Getty Images claimed the model itself was an "article" that Stability AI knew, or ought to have known, contained infringing copies of its works.

The Court accepted that an article may include intangible electronic items. However, to amount to secondary infringement under UK copyright law, the article must contain a reproduction of the copyrighted work.

The Court found that the AI Model does not store or reproduce Getty Images' photographs. Instead, the AI Model learns from patterns, colours and composition in the training data and generates new images that are not copies. As the AI Model did not retain or reproduce Getty Images' works, the secondary infringement claim was dismissed.

Trade mark infringement

Getty Images also alleged that Stability AI infringed its trade marks because some AI generated outputs displayed the Getty Images watermark. The Court found the trade mark claim succeeded in part. These findings were described as "both historic and extremely limited in scope", with no evidence of widespread or ongoing misuse.

Why this decision matters

1. The Court recognised AI training as transformative

The decision suggests that training an AI model on copyrighted works may be lawful where the model does not reproduce those works. In this instance, the Court accepted that the training process involved a form of transformative learning rather than the storage or reproduction of the underlying images. This characterisation may influence how courts in other jurisdictions approach the distinction between learning from data and copying data.

2. Location of training is critical

The case highlights the importance of where training occurs. For AI developers operating across borders, the physical location of training infrastructure may affect which copyright laws apply and the risk of litigation.

3. Trade mark risk for AI generated outputs

The decision confirms that trade mark infringement may arise where AI generated content contains brand elements such as watermarks, even if generated inadvertently.

Key takeaways

1. The decision provides no blanket approval for training AI on copyrighted content; it reinforces the importance of understanding how the model stores and processes information.
2. Model outputs containing watermarks or protected brand elements may give rise to trade mark liability for developers or users.
3. IP rights holders considering enforcement should assess where model training occurred and whether the model retains or reproduces copyrighted content.
4. AI developers should review their training practices, data governance and output monitoring to minimise copyright and trade mark risk.

How we can help

If you would like guidance on copyright compliance or protecting your brand in the age of AI, please contact a member of our Intellectual Property + Technology Team.

*Getty Images v Stability AI [2025] EWHC 2863 (Ch)

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.