There have been few more closely-watched cases in the last decade than Getty Images v Stability AI.

As Stevens & Bolton’s Emma Thompson explained here the trial began on 9 June. It finished on 30 June and has proved to be something of a rollercoaster, so what happened and what might the judgment – when it comes – now mean?

Getty’s claims

Getty began the trial asserting four claims.

1. Copyright infringement – Getty claimed that:

• copyright in its images was infringed both by using them to train Stability AI’s platform “Stable Diffusion” and by the reproduction of them in the platform’s output; and
• Stability AI committed secondary infringement on the basis that Stable Diffusion is an (intangible) “article” knowingly imported into the UK by Stability AI without Getty’s authorisation.

2. Database right infringement – In addition to copyright infringement, Getty argued that its images constituted a database protected by sui generis database right (which has its origins in the EU Database Directive and now has a UK equivalent following Brexit).

3. Trade mark infringement – Getty claimed that the inclusion of Getty’s watermark in output images of Stable Diffusion constituted registered trade mark infringement.

4. Passing off – Similarly, Getty argued that the inclusion of its watermark was a misrepresentation to the public that the images belonged to or were authorised or endorsed by Getty, when they were not.

Stability AI’s defence(s)

Stability AI’s defence was detailed and multi-layered, but in very broad terms its key arguments on each claim were:

1. Copyright infringement:

• The training and development of Stable Diffusion took place outside the UK, so no infringement occurred in the jurisdiction.
• Stable Diffusion did not reproduce a substantial part of Getty’s images.
• For the purposes of secondary infringement, only tangible articles are afforded protection and as Stable Diffusion did not hold a copy of Getty’s images within it, it could not infringe.

2. Database right infringement:

• There had not been sufficient investment in obtaining, verifying and presenting the contents of the alleged dataset by the Getty entity that purported to own the database right (based in Ireland).
• All relevant rights (including potential database rights) had been assigned to the First Claimant, which as it was a company incorporated in the USA did not qualify for UK database right protection. 

3. Trade mark infringement:

• It was the prompts used by Getty to create the output images relied upon that had caused its watermark to appear.
• Getty’s trade marks were not used “in the course of trade” by Stability AI and there was, therefore, no infringing act.

4. Passing off:

• For similar reasons there had not been any misrepresentation by Stability AI as to the origin of the output images, nor had there been any damage to Getty’s goodwill.

What happened at trial?

The trial took place over three weeks before Mrs Justice Joanna Smith. During the final days of the trial, in its closing submissions, Getty dropped its primary copyright infringement claim and its database right infringement claims. This was seemingly because the evidence adduced at trial supported the argument that Stable Diffusion had been trained in the USA, and also that it was difficult to demonstrate definitively that a substantial part of Getty’s works were reproduced in Stability AI’s outputs. 

As a result, the court will now not need to make any findings on what were arguably the thorniest and most interesting issues before it, namely whether the use of copyright works to train a commercial AI model constitutes an infringement of copyright in the UK, and the extent to which the output of such an AI model that incorporates those copyright works also infringes. 

Where does it leave us?

Judgment is expected to be handed down in the autumn. The court may express some obiter views on the claims that were ultimately dropped, but the decision will not now give clarity on some of the key issues that the industry and observers were hoping would be resolved.

Nevertheless, the court will still have to decide:

• Whether Stable Diffusion is an “article” for the purposes of secondary copyright infringement and if Stabilty AI is liable on that basis (which could have implications for other AI platforms that have used similar training and distribution arrangements).
• If Stable Diffusion’s output images infringed Getty’s trade marks (which would potentially set a precedent for claims based on other trade marks appearing in AI outputs without the owners’ permission).
• Whether the appearance of Getty’s name in output images constitutes passing off. 

There are already different approaches being taken around the world to the legality of training. Both Singapore and Japan have broad text and data mining copyright exemptions which have seen many AI companies moving their data scraping operations there. In the USA the discretionary concept of “fair use” gives scope for arguments either way. The EU has taken a different approach, adopting an “opt out” system, however, with the technical and legal mechanisms not yet in place it remains to be seen how this will work in practice.

There are other AI cases in the pipeline, both in the UK and abroad, but all of this means the spotlight will now swings back to the UK government and the progress of the AI legislative agenda in the UK.

The government referred to the case in its recent Copyright and Artificial Intelligence consultation, and a cynic might say that the government was hoping that the court would make some decisions for it. However, the consultation also acknowledged direct intervention through legislation is likely to be necessary to establish a fair balance in law.

The reaction to the recent consultation and the amendments sought during the passage of the Data (Use and Access) Act 2025 (DUAA) through Parliament shows just how strong views are on both sides, and that a quick resolution is unlikely. 

Under the DUAA the Secretary of State must publish a full assessment of the economic impact of the various options set out in the Copyright and AI consultation by March 2026 together with the government’s proposals. An interim report is also required by the end of 2025, so for now we return to a waiting game….