Trending Articles
Trending Topics
Trending
Stability AI has scored a point against Getty Images in an ongoing legal battle centering on the use of copyright material as training data for AI models – but further cases are looming.
Stability AI is the London-based AI developer best known for the image-generation tool Stable Diffusion. Getty Images sued the AI maker back in 2023 in the UK – when do legal systems ever move fast enough to keep up with technology? – saying that it had used millions of its photographs to train Stable Diffusion’s models.
In response and in trial, Stability argued that training had taken place outside of the UK – so the High Court had no jurisdiction – and stressed that it had not reproduced Getty’s works.
Now, a UK High Court judge has found largely in favour of Stability AI – without offering much clarity to content creators or AI developers. “The most significant AI case to reach the English High Court has been decided and has turned out to be a massive damp squib,” said Iain Connor, intellectual property partner with Michelmores, in a statement.
He added: “The decision leaves the UK without a meaningful verdict on the lawfulness of an AI model’s process of learning from copyright materials.”
In June, Getty dropped two of the main claims, over primary copyright and database right infringement, due to concerns about lack of evidence and jurisdiction issues. It would be, after all, difficult to show specifically where training took place.
That still left a debate around secondary copyright infringement, which refers to enabling, contributing to, or benefiting from someone else’s infringement.
This week, the high court ruled that Stability AI didn’t infringe that secondary copyright. Though the judge admitted that it’s clear Getty images were used to train Stability’s models, the AI company didn’t reproduce those images or store them. (The court did find in favour of Getty on the trademark infringement issue and declined to rule on accusations of passing off.)
“An AI model such as Stable Diffusion which does not store or reproduce any copyright works (and has never done so) is not an ‘infringing copy’,” Mrs Justice Joanna Smith said, according to The Guardian, admitting that it was difficult to balance the interests of both.
Connor noted this secondary infringement allegation failed because there was no evidence that the content was stored or retained. “This should be contrasted with the US legal case involving Anthropic which settled for $1.5 billion because Anthropic admitted that it retained copies of author’s works without permission after it had trained its AI,” he added.
Rebecca Newman, a lawyer at Addleshaw Goddard, told Reuters: “Today’s finding means that copyright owners’ exclusive right to reap what they have sown has been avoided on a technicality.”
Getty was naturally unimpressed with the result. “We remain deeply concerned that even well-resourced companies such as Getty Images face significant challenges in protecting their creative works given the lack of transparency requirements,” the company said in a statement. “We invested millions of pounds to reach this point with only one provider that we need to continue to pursue in another venue.”
Getty added: “We urge governments, including the UK, to establish stronger transparency rules, which are essential to prevent costly legal battles and to allow creators to protect their rights.”
On the other hand, Stability AI’s general counsel Christian Dowell said in a statement that the company was pleased with the result. “Getty’s decision to voluntarily dismiss most of its copyright claims at the conclusion of trial testimony left only a subset of claims before the court, and this final ruling ultimately resolves the copyright concerns that were the core issue,” Dowell said.
The judge admitted her findings were “both historic and extremely limited in scope”.
Plenty of copyright cases remain – and for good reason. While in this instance Stability was found not to be making infringing copies, a key requirement in this case, there’s no question that AI companies have been mining all available sources of content.
The AI industry argues there’s no copyright infringement because nothing has been copied and direct copies aren’t reproducible. Plus, it adds, paying for all of this data would make the models unviable – a bold claim given the trillions of dollars being spent to build the infrastructure necessary for AI.
Of course, you have to wonder how Stability AI managed to train its model on data it didn’t download or store, and another case in the US centres on a file of pirated books being used to train AI.
Content creators note that their material is a necessary part of building and training these AI models, and they aren’t being asked nor are they being paid. This ruling doesn’t solve that quandary, though others might shed more light: Meta has been sued in France by publishers and OpenAI by The New York Times.
The UK government is consulting on how to update the copyright system for the AI age, and is considering a text and data mining exception that rights holders would have to opt out of. There’s also the option of paying content creators via a setup similar to the existing secondary copyright system, in which universities and schools pay for photocopying sections of books at universities or library lending.
Another route is AI companies paying rights holders via partnerships – and Getty just signed a licensing deal with AI search company Perplexity.
Announcing that deal, Jessica Chan, Head of Content and Publisher Partnerships at Perplexity, said: “Getty Images shares our belief that the future of AI‑powered discovery requires respecting the creators behind the content.”
Connor noted: “The case does nothing to answer the ‘big tech vs creative industries’ argument but Getty Images share price rose on Friday on the back of an AI licence deal it struck which suggests that both sides believe it is better to do commercial deals than seek to resolve the issues through the courts.”
Such partnerships might be easier – and cheaper – than lawsuits after all.
