The rapidly evolving legal framework for artificial intelligence is being shaped by judicial decisions across the globe.
Two leading jurisdictions, namely the United Kingdom and Germany, have issued landmark rulings that offer a glimpse into the future of intellectual property law in the age of generative AI.
Getty Images v. Stability AI (the UK)
In the United Kingdom, the High Court delivered its long-awaited judgement in Getty Images v. Stability AI[1]. Getty Images, a global content provider, alleged that Stable Diffusion, Stability AI’s image generator, was illegally trained on its copyrighted images without obtaining any permission or consent from Getty Images.
The court’s findings are nuanced. Most importantly, Getty Images abandoned its primary copyright infringement claims due to evidential issues regarding where the training took place. In particular, Getty Images was unable to demonstrate that storage or reproduction took place in the UK.
The court turned to the question of secondary copyright infringement which constituted the core legal question before the court. Getty Images stated that the image generator contained copies of its images, and that making the model available for download in the UK amounted to copyright infringement.
In a decisive win for Stability AI, the court rejected the claim and determined that the AI model’s statistically derived parameters fall outside the definition of an “infringing copy” under the applicable law.
Nevertheless, Getty Images managed to secure a limited victory on its trademark claim, establishing that Stable Diffusion had, in the past, generated images bearing the Getty Images watermark.
GEMA v. Open AI (Germany)
Meanwhile, the German courts have taken a different stance on the matter. In a case brought by GEMA (a collecting society) the local court ruled that Open AI had infringed copyright by using song lyrics to train its ChatGPT models[2].
This marks the first European decision to directly tackle these questions, offering much-needed clarity after the UK ruling left key issues unresolved.
Interestingly, the court's judgment diverges from Stability AI's victory in the UK. Judges in Germany determined that the training process amounted to “memorisation” which they classified as a protected act of reproduction under the applicable law.
Furthermore, the court rejected Open AI's position that its use was allowed under the “text and data mining” exceptions for analytical purposes.
Conclusion
Both cases reveal growing European scrutiny of unlicensed AI training data and highlight a fundamental legal divergence. The UK decision avoided broader copyright issues and focused on technical aspects of AI models, while the German judgement offered a clearer approach, concluding that “memorisation” and ability to reproduce specific works amount to infringement.
Beyond their legal significance, these decisions emphasise that AI advancement must now co-exist with strong IP safeguards.
