case-law

Andersen v. Stability AI: The Artists' Class Action, Explained

By Nora Whitfield · August 8, 2026

Andersen v. Stability AI was the first lawsuit of its kind. Artists suing an AI company as a group, arguing their work had been ingested by the millions to build a commercial product that now competes with them. Filed in January 2023, it predates almost every other major AI copyright case discussed anywhere on this site. More than three years later, it's still going, and 2026 is the year it finally starts moving toward resolution.

The case and the artists behind it

Sarah Andersen, Kelly McKernan, and Karla Ortiz, working visual artists, filed suit in the Northern District of California against Stability AI, Midjourney, and DeviantArt, docket number 3:23-cv-00201. Their claims center on Stable Diffusion, the Midjourney product, DreamStudio, and DreamUp, the AI image tools built and distributed by the three defendants, alleging the platforms were trained on the artists' copyrighted work without authorization, and now generate images capable of substituting for, and competing directly with, the artists' own commercial output. Copyright infringement is the core claim, alongside right of publicity and other related theories.

What made this case significant from the start wasn't necessarily its legal novelty. It was that it was the first time individual creators organized to bring this kind of claim against the AI image-generation industry as a group, rather than a single company or estate pursuing an isolated dispute. Every subsequent artist-side and creator-side suit against an AI image generator has, in some sense, been built on the template Andersen established.

Where the case actually stands in 2026

For a long stretch, Andersen moved slowly through motions to dismiss, amended complaints, and preliminary skirmishes over which claims and which defendants would survive. That phase is largely over. As of 2026, the case has reached a real procedural milestone: a third amended complaint was filed on February 27, 2026, with defendants' answers following on March 13. The case is now heading toward the two events that actually decide whether claims like this succeed at scale.

Class certification and summary judgment motions are scheduled to be heard together on November 4, 2026. That's a significant date to watch, because it will determine two separate things at once: whether the case proceeds as a genuine class action representing a broader group of artists beyond the three named plaintiffs, and whether any of the core legal theories survive as a matter of law before ever reaching a jury. A trial date has been set for April 5, 2027, contingent on what survives the November hearing.

That timeline matters for understanding where this case sits relative to the rest of the AI copyright landscape. Several other major cases, Bartz v. Anthropic among them, have already reached settlement or a fair-use ruling on the merits. Andersen, despite being the earliest-filed case in this entire area, is still working through foundational procedural questions. Visual art and image-generation claims have moved more slowly and faced tougher sledding than the book-publishing cases, partly because the underlying technical and evidentiary questions, how training data maps to specific outputs, whether a given generated image is substantially similar to any specific training image, are harder to establish with the same clarity a text-reproduction claim can offer.

Why this case is harder to win than it looks

The core difficulty in Andersen, and in image-generation copyright claims generally, is a technical one as much as a legal one. It's relatively straightforward to show a language model reproduced a specific paragraph of a specific book, as the New York Times case demonstrates. It's much harder to show that a diffusion model's output of, say, a fantasy landscape in a painterly style is substantially similar to any one of the millions of images it was trained on, as opposed to simply having learned a general style from exposure to many similar works. Style itself isn't protectable under copyright, only specific expression is, which means the artists have had to build their case around theories that can survive that distinction, arguing not just "my style was learned" but that specific protected expression was copied or that the resulting product substitutes for and harms their specific commercial market.

That's a genuinely harder case to build than the piracy-focused claims in Bartz or Kadrey, where the underlying question, was this specific book obtained through an authorized channel or a pirated one, is much more binary and easier to prove with documentary evidence.

What this means for AI authors

If you're a visual artist watching the AI copyright landscape and wondering why cases like Andersen seem to move more slowly than the headline-grabbing settlements in publishing and music, the honest answer is that image-based claims face a harder evidentiary road, not that the artists' underlying concerns are less legitimate. Proving substantial similarity or protectable copying in a visual, style-heavy medium is simply a more technical and more contested exercise than proving a chatbot reproduced a paragraph of text.

The practical lesson for your own work, whatever the eventual outcome in November 2026 and beyond, is the same one that runs through nearly every case in this pillar: the stronger your own documentation of your creative process and your original work's distinctiveness, the better positioned you are, both to participate in claims like this one if you're affected, and to defend your own AI-assisted visual work if anyone ever challenges its originality or its relationship to material you were trained on or influenced by. A clear record of your own creative decisions is valuable regardless of which direction the underlying industry-wide legal fights eventually resolve.

Copyrightable is not a law firm and doesn't provide legal advice. This case remains active with major rulings expected in late 2026. Confirm current status at the official docket (3:23-cv-00201, N.D. Cal.) before relying on this summary.