AI and Intellectual Property: Lessons from Getty Images and Stability AI
With the rapid advancement of artificial intelligence comes significant litigation risks which are often overshadowed by the hype around its transformative benefits. The scarcity of updated laws and regulations has left many startups and established companies inadvertently exposed to intellectual property (IP) litigation.
For organisations, the fallout can include substantial monetary losses and reputational damage. Creators face revenue erosion as their original works lose commercialisation opportunities in an AI-flooded market.
A landmark UK case illustrates these tensions. In Getty Images v Stability AI (November 2025), Getty accused Stability AI of using millions of its copyrighted images (via URLs in training datasets like LAION) to develop Stable Diffusion, an open-source generative AI model. Getty alleged infringement of copyright, database rights, and trade marks. It pointed to outputs that reproduced Getty watermarks.
Stability argued that training/development occurred outside the UK (limiting UK rights’ applicability) and denied secondary infringement, arguing the model’s weights do not store or reproduce infringing copies.
Getty dropped its primary copyright and database claims. The High Court rejected the secondary copyright infringement claim entirely, finding the model itself was not an “infringing copy.” However, it upheld limited trade mark infringement for early versions (v1/v2) due to watermark-bearing outputs deemed “historic and extremely limited.”
Getty has permission to appeal the copyright ruling (decision pending in 2026). It was refused permission to appeal against the findings on trade mark infringement concerning Stable Diffusion v3.x model.
This case, along with new developments in the legal, creative, and technological sector will continue shaping how UK law treats AI training data and model outputs. Organisations developing artificial intelligence models face choices in acquiring and managing training data such as licensing datasets, ensuring no substantial reproduction of originals in model outputs, and limiting data mining to publicly available (and ideally opt-out-respecting) sources.
Disclaimer: This publication is for educational and informational purposes only and does not constitute formal legal advice or create an attorney-client relationship.