Getty Images v Stability AI - UK trial begins... (Part 2 - copyright and database right)

This is the second instalment reporting on the opening of the Getty Images v Stability AI case. The first part, setting the scene and covering the trade mark and passing off claims, is available here. I had intended to cover the safe harbour defences in this post, but the copyright and database right claims are meaty enough for one day, so that will have to wait.

This Kat has a headache.

Copyright

This is probably the area of the case most closely watched.

 

There are broadly three copyright claims: copyright infringement by copying Getty’s works in the course of training Stable Diffusion; secondary copyright infringement by deploying Stable Diffusion in the UK; and copyright infringement in relation to the outputs created by Stable Diffusion.

Taking those in turn:

Training: Getty appears to have accepted now that some training of Stable Diffusion took place outside of the UK (the ‘Final Training’). Whether there was any reproduction of copyright works in the UK in earlier parts of the training is going to be a matter for evidence. If there was no downloading or storing copyright works in the UK, there is no infringement of UK copyright. Getty claims that the absence of evidence from key people means that the court should draw an inference; Stability says that it has spent approaching £3m on its disclosure, and has done everything it is required to do.

Deploying (secondary infringement): This is the bit of the judgment I will be scrolling to first, after the end to see who won. Getty’s case is that “Stable Diffusion is an ‘article’ which is, and which Stability knows or has reason to believe is, an infringing copy of the Copyright Works or each or any of them” (see ss 22 and 23 CDPA). Stability attempted to obtain summary judgment on this point earlier in the proceedings, but that application was refused ([2023] EWHC 3090 (Ch)). The question of whether an AI model is an ‘article’ or an ‘infringing copy’ is novel and will come down to both factual issues and statutory interpretation (including the so-called ‘always speaking’ principle). Getty’s case is that ‘article’ includes intangible articles, and that as the making of the model would have constituted an infringement of Getty’s copyright, then the model itself is an infringing copy, notwithstanding that it may not actually retain any copy of the works used to train it. Stability’s case is that articles must be tangible, and the model cannot be a copy if there is no copy of the works retained.

Outputs: According to Stability’s skeleton argument, only 13 images have been identified that appear to have been reproduced in outputs. Even then, Stability’s case is that the prompts used to generate those images (apparently verbatim or near-verbatim/reworded copies of the captions used on Getty’s database) would not have been inputted by real-world users, at least in sufficient quantities to justify the claim. Stability says they were ‘contrived’.

A neutral observer might hope that Getty does get past this hurdle (if it is actually a hurdle), so that the court can go on to consider whether Stability authorised the infringement (c.f CBS v Amstrad [1988] UKHL 15) and/or communicated the copies to the public. The authorisation claim is particularly interesting as Amstrad involved a very old world form of copying (double cassette decks which could be used for piracy), and how that case overlaps with AI could be fascinating.

Database right

The claim is under the sui generis database right, continuing to be available in the UK post-Brexit.

Subsistence

Stability’s attack on Getty’s database right claim starts with subsistence. As is common in many intra-group licensing structures, IP rights have (according to Stability) been assigned to a US entity. Readers may know that this quirk of database rights and collision of Brexit and EU law has come up before in Genius Sports Technologies Ltd v Soft Construct (Malta) Ltd [2021] EWHC 3200 (Ch).

Non-EU/UK readers may be unfamiliar with database rights as they are not the result of any global treaty. In short, they are a creation of EU law designed to protect investment in obtaining, verifying and presenting the contents of a database. Another key feature of the EU sui generis database right is that it is available only to individuals who are nationals of or resident in EEA member states and companies that are incorporated in EEA member states. The UK equivalent of this right, post-Brexit, is equivalent individuals or companies in the UK.

The right was also significantly limited by a CJEU decision in 2004  (BHB v William Hill Case C-203/02, EU:C:2004:695), where the CJEU held that the investment must be in the obtaining, verifying and presenting of pre-existing data, not the creation of it.

Stability’s position is that (a) by virtue of costs sharing agreements (the details of which are largely redacted from the non-confidential skeleton arguments),  the second claimant (an Irish company) did not make a qualifying investment, and (b) the Irish second claimant has assigned its database rights to the first claimant (a US company), thereby extinguishing any database rights. Getty disputes this interpretation, arguing that the Irish company has made sufficient investments and that its right has not been so transferred or extinguished. Getty however accepts that it does not have any rights in materials added to the database after Brexit. The question of whether the contract assigned the Irish company’s database rights to the US company is one for trial, and involves an assessment under Washington law.

Infringement

The first point, again, is where the training took place (training outside of the UK not being an infringement of this database right, which will need to be determined at trial). The second is whether the fairly limited number of copies in evidence (16) represent or are indicative of sufficiently substantial re-utilisation in the UK. 

On to safe harbours tomorrow! There really is about five cases wrapped up in this one.