Stability AI moves to strike six Getty Images claims in California federal court, arguing garbled AI-generated watermarked images do not constitute trademark infringement or DMCA violation
Stability AI has filed a motion in a California federal court urging the judge to dismiss six claims brought by Getty Images, the stock photography agency, in a lawsuit alleging that Stability AI misused millions of Getty's photographs to train its AI image-generation model, Stable Diffusion. Stability AI's argument targets a specific category of evidence: garbled AI-generated images that reproduce a version of Getty's watermark. The company argues that such images — degraded reproductions that bear the Getty mark in distorted form — do not meet the legal threshold for trademark dilution, trademark infringement, or a violation of the Digital Millennium Copyright Act (DMCA) (US federal legislation that, among other things, prohibits the removal or alteration of copyright management information (CMI) embedded in works). The DMCA claim turns on whether Getty's watermark constitutes CMI and whether Stable Diffusion's reproduction of it in garbled form amounts to an unlawful removal or alteration. The case is one of the most commercially significant AI training data disputes in litigation. A ruling in Getty's favour on any of the six claims — particularly the DMCA CMI count — would significantly increase the legal risk profile of generative AI models trained on scraped image datasets. Stability AI is also facing parallel proceedings in the UK High Court, where Getty has brought an equivalent action under English copyright law, making the US outcome directly relevant to the UK litigation strategy of both parties.
Why this matters
The Stability AI motion to dismiss is a bellwether for the entire generative AI training data litigation landscape. If the California court accepts that garbled watermark reproductions do not constitute DMCA CMI violations, it narrows the evidentiary toolkit available to rights-holders in AI training data cases. Conversely, a ruling that the watermark argument is sufficient to survive dismissal would validate a new litigation theory that can be deployed against any image-generation model trained on scrape-sourced data. For UK lawyers, the parallel UK High Court proceedings mean that the California outcome will influence Getty's English law pleadings and, ultimately, the development of UK copyright law around AI training. This is a live dispute with direct cross-border legal implications, activating IP litigation, technology disputes, and AI governance practice groups simultaneously.
On the Ground
A trainee on the UK High Court parallel proceedings would assist with disclosure review and categorisation — reviewing the Getty image dataset evidence and the AI output examples to assess which category of disclosure each document falls into. They would also help with chronology preparation, tracking the development of Stability AI's training process and the instances of watermark reproduction.
Interview prep
Soundbite
Whether a garbled AI watermark constitutes DMCA copyright management information will set the evidentiary floor for every future AI training data claim.
Question you might get
“What are the key differences between how US and UK copyright law approach AI training data scraping, and how might that affect Getty's litigation strategy in the UK High Court proceedings?”
Full answer
Stability AI has moved to dismiss six of Getty Images' claims in a California federal court, arguing that AI-generated images bearing a distorted Getty watermark do not meet the legal threshold for trademark infringement or DMCA copyright management information violations. This matters because the outcome will define whether garbled AI reproductions of rights-holder marks carry legal liability — a question with direct implications for every generative AI model trained on scraped data. For UK lawyers, the case has a direct UK dimension: Getty has filed parallel proceedings in the High Court, meaning the California ruling will influence UK litigation strategy and potentially English copyright law's development around AI training. The 'why now' is that generative AI models are being commercialised at scale, forcing courts to rule on liability theories that copyright frameworks were not designed to address.
My notes
saved