Getty Images alleges in London and California that Stability AI copied millions of its photographs to train Stable Diffusion and infringed its trademarks. The High Court rejected Getty’s UK secondary copyright claim in November 2025 but found limited trademark infringement, and Getty’s appeal to the Court of Appeal is pending. In the U.S. case, fact discovery closes on November 20, 2026, and jury selection is set for January 18, 2028.
Latest development Aug 18, 2026
On August 18, 2026, Judge Trina L. Thompson of the U.S. District Court for the Northern District of California granted in part Getty’s motion to modify the schedule. Fact discovery now closes on November 20, 2026, and dispositive motions are due on May 7, 2027, for a hearing on August 31, 2027. Jury selection remains set for January 18, 2028.
Why it matters
The UK appeal will decide whether an AI model that stores no copies of its training images can be an “infringing copy” under English law. The U.S. case tests whether the training itself infringes. Both affect AI developers and rights holders.
Background
In January 2023, Getty Images (U.S.), Inc. and Getty Images International UC announced proceedings against Stability AI Ltd in the High Court of England and Wales (Chancery, IP List). Getty alleged that Stability AI scraped more than 12 million Getty photographs to train its Stable Diffusion model. It claimed infringement of copyright, database rights, and trademarks, and the case tested whether training generative models on scraped content is unlawful. In February 2023, Getty filed a parallel suit in the U.S. District Court for the District of Delaware.
The UK trial opened on June 9, 2025. Getty could not show that the training took place in the UK. It therefore dropped its primary training-infringement and database-right claims mid-trial and pressed secondary infringement and trademark claims. In August 2025, Getty voluntarily dismissed the Delaware case and refiled it in the U.S. District Court for the Northern District of California. The refiled complaint reaches the training conduct. It keeps copyright, DMCA, and trademark theories and adds a market-dilution theory of harm.
On November 4, 2025, the High Court rejected the secondary copyright infringement claim. It held that an AI model that stores no copy of the works is not an “infringing copy” under the CDPA. The court gave Getty only a limited trademark win where outputs reproduced Getty watermarks and dismissed the section 10(3) claim. In December 2025, Getty obtained permission to appeal the copyright ruling to the Court of Appeal. Stability was refused permission to appeal the trademark findings.
On April 23, 2026, the California court largely denied Stability’s motion to dismiss and dismissed only the DMCA false-CMI claim, with leave to amend. On August 18, 2026, it revised the schedule, and jury selection remains set for January 18, 2028. The UK appeal remains pending with no hearing date confirmed. Dozens of other AI training cases continue worldwide.
Timeline 10 events
-
OrderJudge Thompson revises schedule and extends fact discovery
Fact discovery now closes on November 20, 2026. Dispositive motions are due on May 7, 2027, for an August 31, 2027, hearing, and jury selection remains set for January 18, 2028.
-
OrderN.D. Cal. largely denies Stability’s motion to dismiss
Trademark infringement, false designation, dilution, and unfair competition claims proceed, and Getty’s copyright claims remain. The court dismissed only the DMCA false copyright management information claim, with leave to amend.
-
AppealGetty wins permission to appeal to the Court of Appeal
Permission was granted in full on the secondary infringement (“infringing copy”) point as novel and important. Stability was refused permission to appeal the trademark findings.
-
OrderN.D. Cal. scheduling order sets jury trial for January 2028
The docket in No. 3:25-cv-06891 shows jury selection and trial set for January 18 to February 9, 2028, before Judge Trina L. Thompson. The estimated trial length is 14 days.
Getty Images (US), Inc. v. Stability AI, Ltd., 3:25-cv-06891 – CourtListener.com
-
JudgmentHigh Court rejects secondary copyright claim, gives narrow trademark win
In [2025] EWHC 2863 (Ch), the court held that model weights that store no copies are not “infringing copies” under the CDPA. It made only limited trademark findings on watermark outputs and dismissed the section 10(3) claim.
-
OrderGetty refiles U.S. case in the Northern District of California
Getty voluntarily dismissed the Delaware case. The refiled case, No. 3:25-cv-06891, was assigned to Judge Trina L. Thompson and keeps copyright, DMCA, and trademark theories while adding a market-dilution theory of harm.
-
OrderGetty drops primary training-infringement and database-right claims mid-trial
Getty accepted it could not show the training of Stable Diffusion occurred in the UK, leaving secondary infringement and trademark claims for judgment.
-
HearingUK trial opens before Mrs Justice Joanna Smith
-
FilingGetty files parallel U.S. suit in the District of Delaware
Getty Images (U.S.), Inc. v. Stability AI, Inc., No. 1:23-cv-00135 (D. Del.).
-
FilingGetty announces High Court proceedings against Stability AI in London
Experimental AI forecast
How it is madeHow will the Court of Appeal of England and Wales resolve Getty’s appeal on the “infringing copy” point by December 31, 2027?
- Court of Appeal dismisses Getty’s appeal46%
- Still pending on December 31, 202730%
- Court of Appeal allows Getty’s appeal17%
- Appeal settled or withdrawn7%
- Resolves by
- Dec 31, 2027in 14 months
- Record confidence
- Medium
The most likely result is that the Court of Appeal dismisses Getty’s appeal by December 31, 2027 (46%). The High Court’s reading of “infringing copy” rests on the finding that the model stores no copies of Getty’s works, a factual base that is hard to disturb on appeal. A decision could still slip past the window.
How it resolves. Resolves on the Court of Appeal’s handed-down judgment on Getty’s appeal from [2025] EWHC 2863 (Ch). “Appeal dismissed” if the court upholds the High Court’s holding that Stable Diffusion model weights are not “infringing copies” under the CDPA. “Appeal allowed” if the court holds, in whole or in material part, that the model can be an infringing copy, including on remittal. “Settled or withdrawn” if the appeal is withdrawn or discontinued, or a settlement is publicly reported before judgment. “Still pending” if none of these has occurred by December 31, 2027.
Reasoning
Starting point. Civil appeals heard by the Court of Appeal of England and Wales after permission is granted: Roughly 30% to 40% of civil appeals that reach a full Court of Appeal hearing are allowed in whole or part.
Civil appeals heard by the Court of Appeal succeed roughly 30% to 40% of the time, and that base rate is the starting point. The novel point earned full permission, which slightly raises the chance of reversal, but the appeal turns on a statutory phrase applied to a trial finding that the weights hold no copies. Appellate courts rarely revisit such technical factual findings, so dismissal remains the leading merits outcome. Timing is the main uncertainty because no hearing date was confirmed nine months after permission. A hearing in the first half of 2027 with a reserved judgment a few months later would fall inside the window, but listing delays could push it into 2028. Settlement is unlikely before judgment because the U.S. case, which reaches the training conduct and survived dismissal in April 2026, is the larger exposure for both sides and heads to trial in January 2028. Any global deal would more likely follow U.S. summary judgment rulings after August 2027. The leading alternative to dismissal is that the appeal remains undecided at year-end 2027.
Trial finding that model weights store no copies of works · No Court of Appeal hearing date confirmed yet · Full permission granted on a novel and important point · U.S. trial track reduces pressure to settle the UK appeal
Generated automatically by Claude Opus 5.5 on Sep 24, 2026 from this case’s record on this site, and published without editorial review. The probabilities are the model’s judgment, not measured frequencies, and they are scored publicly as cases resolve. Not legal or investment advice. How the forecasts are made
Disclosure: Anthropic, which makes the model that wrote this forecast, competes with or has commercial ties to a party to this case.