UK judge delivers a 'damp squib' in Getty AI training case, no clear precedent set
Experts disagree about what the ruling means for AI training on copyrighted material
London's High Court has dismissed the major portions of Getty Images' lawsuit against generative AI firm Stability AI for training its image-generation model on copyrighted images, which some legal experts say could weaken intellectual property laws. However, others saw daylight for trademark and copyright protection in the judge's ruling.
The High Court of Justice handed down its decision on Tuesday. Justice Joanna Smith affirmed just a portion of one of several arguments that Getty made in the lawsuit it filed in 2023, namely that the presence of Getty watermarks on certain images generated by Stability's Stable Diffusion models constituted trademark infringement. As for everything else, Smith found Getty failed to prove its claims.
Getty sued Stability for copyright infringement in part because those watermarks showed up in AI-generated images, which the photo agency argued was clear evidence that London-based Stability had scraped its massive library of photographs to train its AI models.
But Getty couldn't prove that any of that training had taken place in the UK, forcing it to drop its more general claim of copyright infringement in the middle of the trial. "Getty Images may be able to maintain such a case in the jurisdiction where the Model was in fact trained, but there is no basis for that case in this jurisdiction," Smith wrote in her decision.
Getty did try to make a case of secondary infringement – that is, arguing that even if the training took place elsewhere, importing infringing content was still illegal. But Smith dismissed that claim as well. Because generative models like Stable Diffusion don't store actual copies of the content they're trained on, instead recording the weights, secondary infringement doesn't apply, Smith explained. "While it is true that the model weights are altered during training by exposure to Copyright Works, by the end of that process the Model itself does not store any of those Copyright Works; the model weights are not themselves an infringing copy and they do not store an infringing copy," Smith wrote.
Given the high profile of the case, both Getty and Stability claimed victory.
Stability told The Register in an email that the ruling "ultimately resolves the copyright concerns that were the core issue," even though Getty dropped some of its claims during the trial.
Getty, for its part, said Smith's decision rejected Stability's argument that individual users should be held liable for the production of infringements like Getty watermarks, "confirming that responsibility for the presence of such trademarks lies with the model provider."
The image publisher also declared victory on copyright despite its loss on a jurisdictional technicality, saying the court had confirmed that "wherever the training and development did take place, Getty Images' copyright‑protected works were used to train Stable Diffusion." Getty said it would make that argument in the United States, where it has also sued Stability on similar grounds.
Getty voluntarily dismissed its lawsuit filed in Delaware District Court against Stability in August, but filed a separate action in California.
Implications uncertain
Intellectual property lawyers have been coming out of the woodwork to weigh in on the case, as, in the words of Freeths partner and head of IP law Simon Barker, the Getty v. Stability case "is likely to influence both future litigation and policy debates on AI and intellectual property, not just in the UK but internationally."
Barker told us that the case was an interesting one that tested the collision of AI and IP, and did make it clear that AI-generated outputs reproducing protected trademarks will be a liability going forward.
"Each case will turn on its own facts and rights holders will need to evidence a likelihood of confusion or association with the relevant trade mark to succeed," Barker added.
Michelmores IP partner Iain Connor, on the other hand, described the case as a "massive damp squib" because Getty dropped its most important claim, giving Smith "no opportunity to rule in general terms on the lawfulness of AI's use of copyright protected 'input materials' and whether an AI model's 'output' infringed such copyrights."
"The decision leaves the UK without a meaningful verdict on the lawfulness of an AI model's process of learning from copyright materials," Connor told us in an emailed statement. "The case does nothing to answer the 'big tech vs creative industries' argument."
Rebecca Newman, legal director at law firm Addleshaw Goddard, had a far more pessimistic take for intellectual property holders, arguing that the case sets a bad precedent in the United Kingdom as a whole.
"Despite the protection UK copyright purports to offer, Stability have got away with exploiting authorial works for their huge value in training model weights," Newman told us. "The texture of the end product should be irrelevant – extracting value from protected works is an act reserved to the copyright owner."
Newman added that the court's decision in the Getty v. Stability case avoided having to rule on the copyright holders' rights on a technicality that has considerable implications.
"In practice, models trained on infringing data outside of the UK can be imported into the UK without legal repercussions," Newman said. "The UK's secondary copyright regime is not strong enough to protect its creators."
The implications of the ruling will next be tested in California as Getty brings its case against Stability to the US. A jury trial has been demanded in the California case, but a date has not been set. ®