High Court refuses to throw out Getty Images’ IP infringement claim against generative AI provider Stability AI

As 2023 draws to a close, generative AI remains a hot topic. Not only can you now look the term up in the dictionary, but we also have the first decision from a UK court looking at IP infringement in the context of generative AI. In this decision, the High Court rejected Stability AI’s attempt to throw out some of Getty Images’ claims relating to training and use of the generative AI tool known as Stable Diffusion. Whilst the focus wasn’t directly on questions of whether use of IP protected works in training generative AI can infringe IP rights, the case highlights some of the potential battlegrounds we can expect to see if this case does progress all the way to trial.   

What is the case about?

Stability AI has developed a generative AI tool known as “Stable Diffusion”, which creates synthetic images in response to word commands entered by users. Getty allege that Stable Diffusion was trained using millions of copyright-protected images scraped from its websites without its permission (see here). Those actions, Getty assert, infringed their copyright and database rights. Getty also claim that the outputs (i.e. the images) produced by Stable Diffusion infringe their rights by reproducing substantial parts of their copyright protected works or by bearing Getty’s trade marks (in the form of its watermark). 

In these particular proceedings, the court had to consider various applications that had been made by each of the parties. The most interesting of these was Stability AI’s application for reverse summary judgment or strike out in respect of Getty’s claims that: 

1. the training and development of Stable Diffusion infringed Getty’s IP rights in the UK because Getty’s works were downloaded onto servers or computers in the UK; and
2. importing the pre-trained Stable Diffusion software into the UK amounted to secondary infringement of copyright in the UK.

To succeed, Stability AI had to convince the court that Getty had no realistic prospect of success and that there was no other compelling reason for trial. In assessing this, the court not only had to take account of the evidence before it, but also evidence that could be expected to be available at trial (e.g. following disclosure). 

The training and development claim 

Stability AI’s main line of argument in asking the court to dismiss Getty’s training and development claim was that no training or development of Stable Diffusion had been carried out in the UK. As copyright is a territorial right, there would be no infringement of UK copyright unless Getty could show that an infringing act had been committed here. As a result, whether or not this aspect of Getty’s claim would be struck out ultimately came down to whether there was any prima facie evidence that any training and development of Stable Diffusion had taken place in the UK. 

Whilst Stability AI put forward evidence that no such training or development had taken place here, discrepancies were identified, including:

• statements by the Stability AI CEO in a 2022 interview about Stable Diffusion that the company had fast tracked the UK residency of some developers;
• evidence (based on posts from X (formerly Twitter)) that the CEO, who resides in London, had been personally involved in developing Stable Diffusion; and
• unexplained evidence of data transfers to London from AWS in the US. 

This left the court unable to conclude that Getty’s claim had no real prospect of success.

The court also noted that although Getty had only sought to connect this claim with the UK by inviting an inference that during the development and training of Stable Diffusion, Getty’s images were downloaded onto servers or computers in the UK, more than this couldn’t have been expected from Getty at this stage in proceedings (at least until disclosure has taken place).  

Secondary infringement claim

Stability AI’s application in relation to the secondary infringement claim stood or fell on the meaning of “article” in sections 22 and 23 of the Copyright, Designs and Patents Act 1988 (which provide that importing an article known to be an infringing copy of a work is an act of secondary infringement). The key question was whether that term is limited to physical items or if it also covers intangibles (e.g. software on a website)?  The court concluded that this was a novel point of law that should be decided by the trial judge following full and comprehensive argument from both sides and so, again, refused to strike it out or grant summary judgment. 

Comment

As this is the only live IP infringement case in the UK relating to generative AI, many will be watching it with interest. Whilst the decision didn’t focus directly on any of the big questions around whether training generative AI using third party content could amount to IP infringement, it is helpful in highlighting some of the key battlegrounds we can expect to see as this case progresses.  However, it does highlight the distinct possibility that the main copyright and database right claims may ultimately be dismissed on the grounds that no infringing acts (whether of primary or secondary infringement) were committed in the UK. We can only hope that if that does turn out to be the case, the courts will still opine on the question of whether training generative AI can give rise to IP infringement and, if so, in what circumstances.

Any examination of the secondary infringement claim at trial will also be of general interest to IP practitioners given the potential consequences for copyright law beyond the context of generative AI.