Andersen v. Stability AI class-action copyright lawsuit (2023–ongoing)

On 13 January 2023 the Joseph Saveri Law Firm, the typographer-turned-lawyer Matthew Butterick, and Lockridge Grindal Nauen P.L.L.P. filed a putative class-action complaint in the United States District Court for the Northern District of California, on behalf of three visual artists — cartoonist Sarah Andersen, illustrator Kelly McKernan, and concept artist Karla Ortiz — and a proposed class of "thousands, possibly millions" of other artists whose copyrighted works had been used to train generative text-to-image AI systems without consent, credit, or compensation. The original defendants were Stability AI Ltd., Stability AI Inc., Midjourney Inc., and DeviantArt Inc.; the case, Andersen et al. v. Stability AI Ltd. et al., was assigned case number 3:23-cv-00201 and drawn to US District Judge William H. Orrick III. Andersen v. Stability AI is, on the public record this corpus tracks, the field-defining US strategic-litigation campaign by creative-industry workers against generative-AI training-data practices — the first major American copyright class action against text-to-image AI developers, the case in which a federal court first allowed artist-side direct and induced-infringement claims to survive a motion to dismiss, and the sectoral test case that adjacent professional-community organising — the Concept Art Association's federal-policy advocacy, the Authors Guild's parallel publishing-side suits, and the WGA and SAG-AFTRA AI tracks in the 2023 Hollywood strikes — has invoked as proof of concept that worker-side litigation can move the AI training-data frontier.

Origins: artist organising in late 2022

The case crystallised over the autumn and winter of 2022 out of a wave of artist discovery, online organising, and public protest about generative-AI training-data practices. Three convergent developments did most of the work. First, Spawning's "Have I Been Trained" tool made the LAION-5B training set searchable by image and by URL, letting individual artists confirm in concrete terms that their copyrighted works had been ingested into the Stable Diffusion training corpus — and Kelly McKernan was among the first to make that discovery viral on social media. Second, in early-to-mid 2022 Karla Ortiz had begun observing AI systems reproduce her distinctive style, describing the experience as "invasive in a way that I have never experienced", and through late 2022 had organised artist town halls and connected with machine-learning researchers in preparation for a possible legal action; she was at the same time a board member of the Concept Art Association, which in December 2022 launched its own GoFundMe-funded federal-lobbying push on generative AI. Third, Sarah Andersen's 31 December 2022 New York Times op-ed, "The Alt-Right Manipulated My Comic. Then A.I. Claimed It.", framed for a mainstream-press audience both the alt-right harassment-via-style-imitation and the Stable-Diffusion-via-style-imitation problems that motivated the lawsuit. Ortiz saw McKernan's viral tweet and Andersen's op-ed and reached out to both, and the three agreed to be named plaintiffs in a class action that Saveri and Butterick were already preparing in parallel with their LLM-side litigation against GitHub Copilot.

Alongside these specific organising threads, the December 2022 ArtStation member protest — which forced the Epic Games-owned portfolio platform to add a "NoAI" tag — and the controversy over the Colorado State Fair's 2022 award to an AI-generated artwork had supplied the public-mood backdrop against which the suit was filed.

The January 2023 complaint

The original complaint named three classes of claims. The first and central set was copyright: direct and vicarious copyright infringement against Stability AI for training Stable Diffusion on five billion LAION-scraped images including the plaintiffs' works, and against Midjourney and DeviantArt for distributing models that the complaint alleged were either trained on, or built atop, that same training corpus. The second set was statutory: violations of the Digital Millennium Copyright Act for stripped or altered copyright-management information; violations of plaintiffs' right of publicity under California Civil Code § 3344 and at common law; and breach of DeviantArt's Terms of Service for the platform's launch of its DreamUp text-to-image product on data its users had uploaded under the prior, non-AI TOS. The third was California unfair-competition law.

Counsel framed the case publicly from the date of filing as a campaign against unconsented training-data practices generally, not only against the named defendants. Butterick called Stable Diffusion "a parasite that, if allowed to proliferate, will cause irreparable harm to artists"; Saveri said the case was "a larger fight for preserving ownership rights for all artists and other creators" and argued that "what these artists object to is Stable Diffusion settling on a business strategy of massive copyright infringement from the outset". On the artist side, the campaign's rhetorical core became the "three Cs" — consent, credit, compensation — and that framing has been carried forward unchanged into every subsequent stage of the litigation and into the parallel sectoral organising of the Concept Art Association, the Authors Guild, and the Graphic Artists Guild.

The 30 October 2023 motion-to-dismiss ruling

On 30 October 2023 Judge Orrick issued his first major ruling — a 31-page order on the defendants' motions to dismiss the original complaint that largely went against the plaintiffs. Orrick dismissed with prejudice the copyright claims of Karla Ortiz and Kelly McKernan, holding that neither had registered the works they sued on with the US Copyright Office before filing, as 17 U.S.C. § 411(a) requires; the California unfair-competition claims fell with prejudice as preempted by federal copyright law. Most of the remaining claims — direct infringement against DeviantArt and Midjourney, all vicarious-infringement claims, the DMCA Section 1202 CMI claims, and the right-of-publicity claims — were dismissed without prejudice with leave to amend, with the court demanding "greater levels of specificity" about each defendant's particular reproductive conduct. The only claim to survive without amendment was Sarah Andersen's direct-copyright-infringement claim against Stability AI, based on its alleged copying of her LAION-ingested works into Stable Diffusion's training pipeline.

Substantively, Orrick's October 2023 reasoning made two moves that have shaped every later AI-copyright case. He treated the LAION dataset as a factual matter — accepting that LAION had scraped billions of images at Stability AI's behest — but held that the act of scraping was not yet, on the complaint's pleadings, a sufficient allegation of infringement without more. And he expressed explicit skepticism about the plaintiffs' "compressed copies" theory — the assertion that Stable Diffusion stores compressed copies of its training images in its model weights — and demanded clarification of how the model allegedly preserved them. On derivative works, he flagged the awkwardness that plaintiffs themselves had pleaded outputs were "unlikely to be a close match for any specific image in the training data", which made the substantial-similarity inquiry difficult on the pleadings.

The November 2023 amended complaint and the August 2024 ruling

On 8 November 2023 Saveri and Butterick filed a First Amended Complaint expanding the named-plaintiff roster from three to twelve artists — the original three plus Hawke Southworth, Grzegorz Rutkowski, Gregory Manchess, Gerald Brom, Jingna Zhang, Julia Kaye, Adam Ellis, Hope Larson, and Jess Fink — and added Runway AI as a defendant. The amended pleadings re-pleaded the dismissed claims with the per-defendant specificity Orrick had demanded; introduced two alternative theories of direct copyright infringement that came to be known as the Model Theory and the Distribution Theory (the first that Stable Diffusion itself retains internal representations of training images and reproduces them on inference, the second that the distribution of model weights is itself an unauthorised reproduction); and added a new Lanham Act count alleging that Midjourney's public 4,700-artist "Names List" and the company's Discord-based "showcase" pages constituted false endorsement and vicarious trade-dress infringement against the named artists.

On 12 August 2024 Judge Orrick issued the second major ruling — the order on the defendants' motions to dismiss the amended complaint that has dominated AI-copyright commentary ever since. The court denied the motions on the central copyright and Lanham Act claims and granted them only on a tightly bounded set of statutory claims.

On direct copyright infringement, Orrick held that the plaintiffs' allegations made it plausible that Stable Diffusion retains training-image content in a form that enables reproduction, and that the Model and Distribution Theories could proceed in tandem into discovery. On induced copyright infringement, he held that Stability AI's marketing of Stable Diffusion — including public statements by then-CEO Emad Mostaque indicating the model could "recreate" training images — together with academic and contemporaneous press evidence of replicated outputs was enough to defeat a Sony "substantial non-infringing uses" defence at the pleading stage and to permit the inducement claim against Stability, Midjourney, DeviantArt, and Runway to proceed. On the Lanham Act, Orrick allowed the false-endorsement and vicarious-trade-dress claims against Midjourney to proceed, citing the Names List, the showcase imagery, the CLIP-model-as-trade-dress-database allegation, and the "mental recognition" of Midjourney's calling out of plaintiffs by name.

On the dismissal side, Orrick dismissed with prejudice the DMCA Section 1202 CMI claims — both the § 1202(a) provision-of-false-CMI claim and the § 1202(b) removal-of-CMI claim — holding that the plaintiffs had not satisfied the "double scienter" requirement that the defendants knowingly provided false CMI with the intent to induce or enable infringement. He also dismissed with prejudice the breach-of-contract claims against DeviantArt and the breach of the implied covenant of good faith against DeviantArt. The right-of-publicity claims under California § 3344 were dismissed; the unjust-enrichment claim was dismissed without prejudice with leave to amend.

The August 2024 ruling has been the most-quoted single decision in the early US AI-copyright case-law. Loeb & Loeb's contemporaneous client alert summarised the holding as the first US federal recognition that artists may pursue both direct infringers (Stability, Runway) and downstream distributors (Midjourney, DeviantArt) in an "AI supply chain" theory, and that AI products reproducing "the distinctive look and feel" of artists' work raise valid trademark concerns. The decision sent the surviving claims into discovery.

Counsel, plaintiffs, and the artist-organising spine of the case

The litigation has been run from filing onward by the Joseph Saveri Law Firm — a San-Francisco-based class-action firm that, in parallel with this case, has run the GitHub Copilot–Microsoft–OpenAI software-side litigation, the OpenAI / Anthropic LLM-side authors' suits, and the April 2024 Google Imagen complaint — together with Matthew Butterick, whose digital-publishing background (typography, online legal-writing books) gave the campaign its public-facing communications voice, and with Lockridge Grindal Nauen P.L.L.P. as co-counsel.

The named-plaintiff roster is structurally a sectoral coalition rather than a single discipline. Sarah Andersen is a syndicated cartoonist whose Sarah's Scribbles strip had been the most viral target of style-imitation; Kelly McKernan is a Nashville-based watercolour and acrylic illustrator working in a fantasy-and-portraiture register; Karla Ortiz is a San-Francisco concept artist who has done work for major Marvel and Universal productions and who sits on the Concept Art Association board. The November 2023 expansion drew in further name-recognition cases — Grzegorz Rutkowski (a Polish digital painter whose name had been the single most-prompted artist name in Stable Diffusion's early viral period and on Midjourney's Names List), Gerald Brom and Gregory Manchess (career fantasy illustrators), Hawke Southworth and Hope Larson and Adam Ellis (comics), Jingna Zhang (photographer), Julia Kaye (cartoonist), and Jess Fink — and turned the proposed class from a three-discipline cluster into a cross-creative-industry one. Ortiz testified before the US Senate Judiciary Subcommittee on Intellectual Property in July 2023 on the case and on the wider artist-side argument: "I have never been asked…never been credited…never been compensated."

The lawsuit is not formally a Concept Art Association action — the named parties are individual artists, not the organisation — but the organising lineage is publicly intertwined. CAA's federal-lobbying advocacy programme on generative AI, launched in December 2022 with a GoFundMe campaign that Ortiz publicly fronted, ran in parallel with the lawsuit and is described by the organisation itself as part of the same sectoral wave of professional-community organising that the lawsuit consolidated. CAA participated in the FTC's October 2023 Creative Economy and Generative AI roundtable with Ortiz on the panel — the same week, in effect, that Judge Orrick was preparing his first motion-to-dismiss ruling.

Status and forward arc

The case is currently scheduled for trial on 8 September 2026 before Judge Orrick in San Francisco, with a February 2026 third amended complaint as the live pleading and active discovery proceeding throughout 2025–2026. A separate but coordinated complaint filed by the same counsel against Google over Imagen in April 2024 proceeds on its own track. The proposed class — US copyright owners whose works trained the defendants' image-generation systems — has not yet been certified; class-certification proceedings are anticipated alongside or after the close of merits discovery.

Significance for the broader AI-good movement

For the corpus this team is mapping, Andersen v. Stability AI marks four enduring contributions. First, it is the first US federal case in which artist-side copyright claims against generative-AI developers have survived a motion to dismiss on a training-data-ingestion theory — establishing, on the public record, that the question of whether unconsented AI training infringes copyright is a question the federal courts will hear on the merits rather than dispose of at the pleadings. Second, it is the first US federal case in which an artist-style prompting feature — Midjourney's Names List and showcase — has been allowed to proceed as actionable Lanham Act false endorsement and vicarious trade-dress infringement, opening a trademark-side route to constraining how generative-AI products market themselves around named artists. Third, the case has functioned as a sectoral organising spine for adjacent US creative-industry advocacy: the Concept Art Association's federal-lobbying programme, the Authors Guild's publishing-side suits against OpenAI and Anthropic, and the worker-side bargaining frame of the 2023 WGA and SAG-AFTRA AI provisions all cite Andersen v. Stability AI as the proof-of-concept that the legal frontier on AI training data could be moved by creative-industry workers themselves rather than by Congress or the Copyright Office acting alone. Fourth — and connecting back to the corpus's wider thesis — the case is the field-defining instance of grassroots / professional-community organising as the upstream input to high-stakes US AI litigation: the lawsuit could not have been filed without the late-2022 wave of artist organising on ArtStation, Spawning, and CAA, and the legal claims it has put before a federal court are the worked-out form of the grievances those artist communities were already articulating in public for several months before the complaint was filed.