Copyright and AI law

Getty Images v Stability AI: a plain-English timeline

Steam rises from a street opening beneath blurred night-time lights.
Steam rising after dark.Photograph by Robert Harper. Robert Harper Photograph Library, Archive.London. From the archive, shown for illustration.

The best-known AI copyright case in Britain never decided whether training on photographs is lawful. Here is what it did decide, what was abandoned, and what the Court of Appeal has just heard.

Getty Images sued Stability AI in London in January 2023, alleging that millions of its photographs had been used to train the Stable Diffusion image generator. Nearly four years later, no English court has said whether that training was lawful. The claim that would have answered the question was withdrawn in the last days of the trial, and the appeal heard on 11 and 12 November 2026 concerns something narrower.

This article is general information, not legal advice. Checked on 18 November 2026.

The timeline at a glance

Date What happened Status
January 2023 Getty issues its High Court claim: copyright, database right, trade mark infringement and passing off Decided at first instance
9 to 30 June 2025 Trial before Mrs Justice Joanna Smith. Late in the trial Getty drops its primary copyright and database right claims Abandoned, later formally dismissed
14 August 2025 Getty files a fresh US claim in the Northern District of California, moved from Delaware Pending
4 November 2025 High Court judgment, [2025] EWHC 2863 (Ch). Secondary copyright claim dismissed; limited trade mark infringement found Decided
December 2025 Getty granted permission to appeal on secondary infringement, [2025] EWHC 3343 (Ch) Decided
23 April 2026 US court rules on Stability's motion to dismiss: trade mark, dilution and unfair competition claims survive; one claim dismissed with leave to amend Decided; case continues
11 and 12 November 2026 Court of Appeal hearing, case CA-2026-000007. BPI intervenes in support of Getty Heard
Not yet announced Court of Appeal judgment Pending

What did Getty claim, and what did it abandon?

The original claim had four parts. First, that Stability infringed copyright by copying Getty's photographs to train and develop the model. Second, that images generated by Stable Diffusion reproduced Getty's works. Third, that the model itself was an infringing article brought into the UK, which is secondary infringement. Fourth, that Getty's watermarks appearing in generated images infringed its trade marks and amounted to passing off.

Two of those did not reach judgment. Getty dropped the training claim during the trial because there was no evidence that the copying took place in the UK. UK copyright law applies to acts done in the UK, and the training was not shown to have happened here. Getty dropped the output claim after Stability blocked the prompts said to produce infringing images, which according to Mishcon de Reya's analysis gave Getty the practical remedy it was seeking. An order sealed in January 2026 records that those claims were dismissed, having been "fought to closing submissions and then abandoned", as CMS reports it.

So when the judgment came, the question most people associate with the case was no longer in it.

What did the High Court decide on 4 November 2025?

Three things matter.

An "article" can be intangible. Secondary infringement under sections 22 and 23 of the Copyright, Designs and Patents Act 1988 concerns importing, possessing or dealing with an "article" that is an infringing copy. Stability argued an article had to be a physical object. The judge disagreed: statutes are read as "always speaking", and an electronic copy held in cloud storage can qualify.

But the model is not an "infringing copy". Under section 27, an infringing copy has to be a copy of the work. The judge accepted that Stable Diffusion's model weights do not store or reproduce Getty's photographs. A model that was made using copies, but does not contain them, was held not to be an infringing copy. The secondary infringement claim failed on that ground.

Trade marks: a narrow finding. The court found what it called limited and historic infringement under sections 10(1) and 10(2) of the Trade Marks Act 1994, arising from watermarks appearing in outputs of some early versions of the model. The reputation-based claim under section 10(3) failed, and the judge chose not to rule on passing off. Law-firm summaries differ slightly on which marks and model versions the successful findings covered, so the judgment itself is the safer source for that detail.

What is the appeal about?

In December 2025 the trial judge gave Getty permission to appeal on the secondary infringement point. She refused Stability permission to appeal the trade mark findings.

The appeal turns on a question of statutory meaning. Getty's argument, as summarised by Burges Salmon on the second day of the hearing, is that "making" a model includes training it, so an imported model is an infringing copy even if no copy of any photograph sits inside it. Stability's answer is that whether something is a copy depends on what it is, not how it was produced.

Stability also asks the Court of Appeal to overturn two findings that went against it: that an intangible model can be an "article" at all, and that Stability itself imported or distributed the model in the UK. Getty disputes both.

The British recorded music industry body, the BPI, was allowed to intervene in support of Getty. Its concern is the one that makes this appeal important well beyond stock photography: on the High Court's reasoning, a developer can train a model outside the UK on British works and then offer it here without secondary copyright liability.

Pinsent Masons, writing two days before the hearing, made the necessary caveat. The primary infringement claims were withdrawn at trial, so the appeal cannot address them. Whatever the Court of Appeal decides, it will not be a ruling on whether training in the UK infringes.

When will there be a judgment, and what could change?

The hearing ended on 12 November 2026. No hand-down date has been published and we have not been able to confirm from a primary source which judges sat. One press report anticipated judgment in late 2026 or early 2027.

There are two possible directions, and it would be wrong to guess between them. If the appeal is dismissed, the present position continues: the model is not an infringing copy, and rights holders must look to the place where training happened. If the appeal is allowed, offering a model in the UK could itself attract liability where it was trained on UK works without permission, wherever that training took place. Either result could be appealed further to the Supreme Court with permission.

What about the American case?

Getty's US claim against Stability AI was refiled in the Northern District of California on 14 August 2025 (3:25-cv-06891). On 23 April 2026 the court allowed the trade mark, dilution and unfair competition claims to go forward and dismissed a claim about false copyright management information with leave to amend. Stability did not challenge the copyright claim at that stage, so the training question is live there. The case is at the pre-trial stage. A scheduling order of November 2025 set a jury trial for January 2028, according to the docket-based site ChatGPT Is Eating the World; we have not confirmed whether that timetable still stands, and US schedules often move.

That matters because the American court can reach what the English court could not. If the copying took place in the United States, US law governs it, and the answer will depend on fair use. No US court has yet ruled on fair use for an image model, as our status table of AI copyright cases shows. The only appellate ruling so far, on a non-generative text tool, is covered in our piece on fair use after Thomson Reuters v Ross.

What the case means for photographers and agencies

For now, three general points hold. They are observations, not advice on any particular dispute.

  1. No UK exception is written for commercial training without permission. The only text and data mining exception covers non-commercial research, and the case did not change that. The wider statutory picture is set out in our article on UK copyright and AI after the 2026 report.
  2. Where the copying happens matters. A claimant needs evidence of acts in the UK, or a claim in the country where training took place.
  3. Watermarks and metadata have evidential value. Getty's trade mark success, limited as it was, came from watermarks surfacing in outputs. Rights information embedded in files serves a similar purpose in showing where an image came from; you can check what a file carries with our EXIF and metadata viewer.

A different kind of image source

Getty's case is about a library aggregated from many contributors and published at scale online. The Robert Harper Photograph Library is the opposite shape: 95,487 unique photographs by one photographer, by the archive's own inventory count, owned by 825 Ventures Ltd, the owner and rights holder. More than half of the archive is still on film and has yet to be scanned, and only a small selection has been published online by Archive.London, including decades of studio work in the fashion collections.

That does not answer every rights question, which is why licensing is negotiated directly and is subject to rights review. If you would sooner license image data than litigate about it, the AI training data licensing page sets out how to begin.

Frequently asked questions

Who won Getty v Stability AI?

At first instance, mostly Stability AI. On 4 November 2025 the High Court dismissed Getty's secondary copyright infringement claim and found only limited, historic trade mark infringement from watermarks in some early outputs. Getty has appealed the copyright point. The Court of Appeal heard the appeal on 11 and 12 November 2026 and has not yet given judgment.

No. Getty withdrew its claim about training and development during the trial because it could not show the copying happened in the UK, and UK copyright is territorial. The judge decided a different question: whether the finished model, made available in the UK, counted as an infringing copy. She held it did not.

When will the Court of Appeal give judgment in Getty v Stability AI?

No date has been announced. The hearing finished on 12 November 2026. Law-firm reports say only that judgment will follow; one press report expected it in late 2026 or early 2027. Anyone relying on the case should check the Judiciary website and the National Archives case-law service for the current position.

What is the appeal actually about?

Whether an AI model can be an infringing copy under section 27(3) of the 1988 Act when it does not contain the works it was trained on. Getty says making a model includes training it. Stability AI disagrees, and also challenges the findings that an intangible model can be an 'article' and that it imported or distributed the model in the UK.

Sources

This article is general information, not legal advice. It reflects the position as we understood it on the date shown and may be out of date. Published by Archive.London. Corrections are welcome under our editorial policy.

Licensing the archive for AI and data projects

95,487 photographs from one photographer and one rights holder. Read the dataset card, search the 503-folder inventory, discuss a sample set under NDA, or arrange a first look as scanning proceeds.

See AI and data licensingWrite to the archive