The Claimants (collectively Getty), whose business centers on licensing photographic images, videos, and illustrations, alleged that Stability AI (Stability) had scraped millions of its images without consent to train various versions of its open-source AI image generator, Stable Diffusion. Getty claimed that this conduct infringed its copyright and database rights, constituted secondary copyright infringement through the importation of the pretrained software into the UK, and amounted to trademark infringement and passing off through the use of Getty’s marks in AI-generated outputs.
In late 2023, Stability sought reverse summary judgment/strikeout of Getty’s claims. The High Court dismissed the application, holding that there were reasonable grounds to believe further disclosure might clarify where training occurred and that novel questions of statutory interpretation raised by the secondary infringement claim should proceed to trial. However, during trial, Getty had to abandon its primary copyright infringement claim, accepting that there was no evidence that Stability had trained and developed Stable Diffusion in the UK.
The Court’s Decision
The key issue for the Court to decide became Getty’s secondary copyright infringement claim that Stable Diffusion was an “infringing copy” imported into the UK in breach of the Copyright, Designs, and Patents Act 1988 (CDPA). This in turn required the Court to decide whether Stable Diffusion was an “article” for those purposes. The CDPA does not define “article,” but the Court held that an article can be an infringing copy only if it actually contains or embodies the copyright work — at least transiently. The Court found that Stable Diffusion’s model weights did not store, reproduce, or contain any of Getty’s images; they were sets of numerical parameters derived from statistical training.
Accordingly, the Court held that merely using infringing copies in the course of creating another artifact does not make that artifact an infringing copy and that the Stable Diffusion model “has never consisted of or contained a copy” of Getty’s works.
Comment
With no equivalent to a U.S. fair-use defense, creators and AI companies had been waiting to see how the English High Court would apply English copyright law to the AI industry. But Getty’s inability to prove that any relevant acts of Stability took place in the UK left the Court with no need to apply copyright law to the alleged training and development acts of Stability. IP rights are territorial: no UK act, no UK infringement.
On secondary infringement, English High Court Judge, Mrs. Justice Joanna Smith, carefully considered the argument and rejected Getty’s claim: Merely exposing model weights to infringing copies during training does not render the resulting model an infringing copy.
AI companies and copyright owners alike may attempt to draw deeper conclusions from Mrs. Justice Smith’s strong judgment; some might even use the decision to justify calls to the UK Government for accelerations or pauses to AI copyright defenses. Ultimately, though, the case teaches nothing new about the application of English copyright law to AI training and development; it reminds copyright holders that they must prove that the alleged acts occurred in the UK, not just that they happened.
As to deployment and use of the final AI model, the judgment is emphatic: Given that the model never contained or stored an infringing copy, supplying that model is not secondary infringement.
Getty may decide to appeal. Until then, the law remains that AI companies training AI models outside the UK face little legal threat inside it.