Getty Images v Stability AI Part II

In our previous article, we considered the trade mark infringement claims in the recent UK High Court decision in Getty Images v Stability AI. In this article, our Intellectual Property team considers the copyright issues which arose in the same landmark decision.

Ultimately, the UK High Court rejected the copyright infringement claims. It did so on the basis that Stability’s Stable Diffusion models do not contain or store reproductions of the relevant works and therefore they are not “infringing copies” for the purposes of copyright infringement. They are not copies of training data and they do not store or reproduce anything, according to the UK High Court. We analyse the copyright aspects of the decision and its implications.

The secondary infringement claim

Getty argued that secondary infringement of copyright had occurred. This type of infringement involves dealing with infringing articles or a means to produce infringing articles. Getty argued that this had occurred as Stable Diffusion was imported through downloading in the UK. Getty also pointed to distribution via Hugging Face. In determining the secondary infringement claim, the High Court considered the following key questions:

• Is Stable Diffusion capable of being an “article” for the purposes of UK copyright legislation?

Getty submitted that this question should be answered in the affirmative. Stability however, argued that the term “article” under the UK Copyright Designs and Patents Act 1988 (the UK Copyright Act) is limited to tangible objects only. According to Stability, intangible or abstract information, such as the Stable Diffusion Models which are distributed over the internet in tangible form, are not articles.

The UK High Court recognised that in today’s world, electronic copies can be stored in an intangible “cloud” without it taking any physical form. It also took into account the use of the words “any medium” in section 17(2) of the UK Copyright Act. Given the breadth of those words, the UK Court was satisfied that it was “plainly intended” to cover any means of storing electronic copies which would include modern means of storage such as cloud storage. The Court was therefore satisfied that an article, which is an infringing copy, is capable of being an electronic copy stored in intangible form.

• Interpretation of “an infringing copy”

Stability argued that whether an article is tangible or intangible, an infringing copy must be a copy i.e. a reproduction of a copyright work. It further argued that in this case, the model weights of the various Stable Diffusion versions do not store the visual information in the copyright works, and therefore they could not amount to a copy. Getty however argued that under the UK Copyright Act, all that matters is that the making of the article amounted to an infringement. Getty’s position was that there is no requirement for copies of the article to be retained for infringement to arise. On this point, the UK High Court concluded that Getty’s submission had no proper basis. The Court could not accept how an article could amount to an infringing copy if it had never consisted of, stored, or contained a copy.

Specific copyright findings

On the facts of the case, the UK High Court concluded that:

• Stable Diffusion is not capable of being an infringing copy and therefore the UK Copyright Act was not applicable.
• Given that an article may be intangible, downloads of the Model amount to importation for the purposes of the UK Copyright Act and those downloads have taken place within the UK. This is because Stable Diffusion is an open-source model which is available for anyone to download.
• There could never be any act of secondary infringement via the provision of remote software services. The case on secondary infringement therefore failed concerning Dream Studio, and the Developer Platform works.

Additional damages

In their pleadings, Getty claimed additional damages in the case on the following grounds:

1. Stability knew or had reason to believe that Stable Diffusion is an infringing copy of the copyright works.
2. Stability had acted with a “cavalier attitude” to Getty’s rights, “infringing numerous different intellectual property rights on a blatant and widespread scale.”
3. Stability had relied upon copyright and database infringement for the successful operation of its business and Stable Diffusion has been, or has the potential to be, used by end users who would otherwise have used the Getty Images websites.

The Court ultimately dismissed all three arguments. In circumstances where it found that there had been no acts of secondary infringement, it held that there could be no basis for additional damages for copyright infringement. Similarly, although instances of trade mark infringement were found, as discussed in our previous article, the Court could not determine that these were widespread, or that they continued beyond the release of version 2x.

Comment

The UK High Court’s finding that for the purposes of UK copyright law, an “article” can be an intangible item represents an interesting development. It will be welcomed by rights holders as it will apply more broadly in other IP cases going forward. In light of this, it is perhaps unsurprising that Getty Images has already been granted permission to appeal the High Court’s secondary copyright infringement findings, with an appeal likely to be heard in early 2026.

The key takeaways for AI producers and providers however is that providing remote access to an AI model may now be considered to be relatively low risk in the UK from a copyright infringement perspective. This is because AI models are not treated as ‘copies’ under copyright law. However, at the same time, AI providers will not be able to assert that they are merely passive providers of neutral tools as the Court was clear that AI providers exercise “absolute control” over the training data. A fine balance therefore needs to be struck. In the main though, AI developers in the UK can take some comfort from the decision that AI models will not be considered infringing from a copyright perspective, as they do not store or reproduce training data. From a copyright infringement perspective, the infringing act is the copying of training data for the purpose of training, as opposed to the existence of the model or the act of training itself.

As mentioned in our previous article, although this is a UK decision, it does have persuasive authority before Irish Courts. Therefore, Irish businesses and AI developers should carefully monitor the developments in the case.

For further information and expert legal advice on the impact of this decision and how best to protect your intellectual property rights, please contact a member of our award-winning Intellectual Property team.

The content of this article is provided for information purposes only and does not constitute legal or other advice.