Welcome to the sixth issue of the HLK Newsletter tracking the AI legal landscape. In this issue, we take a look at a recent decision in the UK dealing with copyright issues in Stability AI’s Stable Diffusion and, as an early Christmas present, we report on the EU AI Act – yes, it has been agreed!!
Generative AI in the Courts – Developments in Getty Images v Stability AI
This week, we’re returning to the Getty Images v Stability AI case in the UK, which we first reported in Issue 1 of our AI Observer.
It has been a busy time for the UK case. Stability has tried to have two of Getty’s infringement claims thrown out, while Getty has tried to include yet another infringement claim. The case involves copyright, database rights, trade marks and passing off. For the purposes of this short summary, we are focussing on the copyright issues.
The UK’s High Court has now handed down its judgment, and we strongly suspect that Getty will be the happier of the parties – at least for now. We’ve briefly set out what was decided by the Court, and why it matters, below.
Training and development of Stable Diffusion
The primary issue decided by the Court was whether to throw out Getty’s ‘training and development’ infringement claim – i.e. the claim that Stability’s use of Getty’s images to train and develop Stability’s AI model, Stable Diffusion, amounted to copyright infringement in the UK.
Stability tried to have the claim thrown out based on evidence which, it said, clearly demonstrated that no training or development of Stable Diffusion had occurred in the UK and, as a result, there could be no infringement of UK rights.
The judge, however, rejected Stability’s arguments. Her decision notes that Getty had provided evidence showing that Stability’s evidence regarding where the training and development took place may be contradictory/inconsistent, and this needed further exploration. The judge also felt there were reasonable grounds for believing that evidence relevant to where the training and development had taken place might come to light during the Court’s disclosure process.
As well as maintaining one of Getty’s likely strongest infringement claims, this decision is important because Stability will now be subject to the UK Court’s disclosure process, including disclosing “the precise nature of the development and training of Stable Diffusion”, and managing any confidentiality issues that arise.
This is not, however, a complete victory for Getty as paragraph 59 of the judgment clarifies. The judge made clear that Stability’s evidence, on its face, remains strong evidence that no training or development took place in the UK. So, unless during the proceedings it becomes clear that Stability’s evidence is, in fact, contradictory/inconsistent, or Getty provides other evidence of UK activities, Stability may well come out on top on this issue eventually.
Secondary copyright infringement
The Court also refused to throw out Getty’s ‘secondary infringement’ claims – i.e. that Stability had unlawfully imported, sold etc. Stable Diffusion in the UK.
Stability argued that these claims should not apply to Stable Diffusion due to the specific wording of the UK’s copyright legislation. Stability argued that the claims only apply to infringing copies which are “articles” and, in this context, “articles” means tangible items, rather than intangibles like Stable Diffusion.
Again, the judge refused to throw out Getty’s claim, stating that the issue raised “a novel question, not definitively determined previously, which would be better resolved once all the facts have been ascertained at trial”.
The outcome of this claim has implications for other cases of secondary copyright infringement involving internet downloading. For example, as Getty argued (and referred to at paragraph 94 of the judgment), if Stability’s view is correct, “there would be an infringing copy and secondary infringement if [Stability] had brought a copy of Stable Diffusion into the UK on a memory stick but, because of developments in technology which mean this can now be achieved via a cloud-based service, there would be no infringement”.
New ‘image-to-image’ infringement claim
Finally, the Court allowed Getty to include new infringement claims relating to a feature of Stable Diffusion released in March 2023 (after the original UK infringement claims were filed by Getty). The feature in question “enables Stable Diffusion to generate a synthetic image output in response to an image uploaded by a user” and “[t]he user is able to determine how closely the synthetic image output matches the image prompt by the use of an “image strength” slider”.
The judge was notably unconvinced by Stability’s view that, because it is the user who makes use of the feature of Stable Diffusion, including the ‘image strength’ slider, it is in fact the user who brings about any copying and not Stability.
Again, the outcome of this claim has implications for other cases, particularly testing the boundaries of whether AI developers may be held responsible for the actions of AI users.
We will, of course, continue to monitor this case and other similar lawsuits.
Legislation watch
The big news from a regulatory perspective is that on 8 December, after a marathon 40 hours of negotiations, the EU lawmakers reached agreement on the scope of the EU AI Act. This means that it is looking almost certain that the EU will be the global leaders in regulating AI. However, this isn’t quite the end of the story as the text of the Act still needs to be formally adopted by member states in the coming weeks – so there is still room for plot twists and political drama before the final text of the Act is agreed and made into law. For instance, President Macron of France has already raised a red-flag to the position adopted in relation to foundational AI models criticising the current draft as negatively impacting innovation leaving the EU at a competitive disadvantage. So watch this space.
In other news, META and IBM have recently announced the launch of the AI Alliance in collaboration with over 50 other founding members. The drive behind the alliance appears to be a focus on transparency and openness in the AI space, particularly around open-source. The approach adopted by the AI Alliance is in stark contrast to the likes of Chat GPT’s Open AI and Microsoft who are driving a much more closed approach to the development of AI models.
And finally, following on from our third issue of the AI Observer, the Review Board of the US Copyright Office has this week once again refused to register an artistic work created by using a generative AI tool – finding that it did not have sufficient human input to be protectable. The work was a 2-D computer generated image called “SURYAST” that was created by Ankit Sahni after inputting an original photo into an AI tool called RAGHAV to create a derivative work in the style of Van Gogh’s Starry Night. The Review Board of the US Copyright Office held that Ankit Sahni did not influence sufficient control on the generated output work and, therefore, could not be named as the copyright owner and nor could the AI tool, as only a human can be an owner of copyright in the US. Therefore, the artistic work was not capable of being protected by copyright at all – once again showing the perils of IP ownership for AI generated works.
AI application of the week
Here’s one to think about over the holiday season. Indiana University Bloomington have come up with a proof of concept using AI in which living brain cells learned to recognise the voice of one individual from hundreds of sound clips. Impressive stuff!
This is for general information only and does not constitute legal advice. Should you require advice on this or any other topic then please contact hlk@hlk-ip.com or your usual HLK advisor.