Adobe Is Being Sued for Training Its Document AI on Pirated Books. A Photographer Who Fought Back Lost.
Two class-action lawsuits filed in December 2025 and February 2026 allege Adobe trained its SlimLM document-assistance AI on nearly 200,000 pirated books without consent. A March 2026 investigation found that a photographer who tried to stop Adobe from using his images for Firefly training was forced out of arbitration after Adobe demanded $24,000 in fees just to review his inability-to-pay motion.
VaultTools · March 20, 2026
Table of Contents
- The SlimLM lawsuits
- What SlimLM does and why documents are central
- The photographer who fought and lost
- What the arbitration outcome means in practice
- What never uploading a file prevents
- Sources
The SlimLM Lawsuits
On December 16, 2025, a proposed class action (Lyon v. Adobe, N.D. Cal.) was filed against Adobe, alleging the company trained SlimLM, its on-device AI model for document assistance tasks, on nearly 200,000 copyrighted books sourced from the Books3 dataset via the SlimPajama-627B corpus. Books3 is a collection assembled from pirated ebook sources. The plaintiffs allege Adobe used it without licensing agreements, author consent, or compensation.
On February 10, 2026, a second proposed class action was filed in California federal court, broadening the scope of the same allegations. Both suits are ongoing.
What SlimLM Does and Why Documents Are Central
SlimLM is Adobe’s lightweight AI model built specifically for document tasks: summarization, content suggestions, editing assistance, and question-answering over document content. It is deployed across Adobe Acrobat and other Creative Cloud products that handle the kinds of files users routinely process: PDFs, contracts, reports, and creative briefs.
The alleged training data controversy matters specifically in this context because SlimLM’s purpose is to understand and generate document content. The argument in the lawsuit is that Adobe needed large volumes of real human writing to train this capability, and that it sourced that writing from pirated books rather than licensed material.
For users who upload documents to Adobe’s cloud tools, the implication is layered: the files you submit help improve AI systems, whether or not that was disclosed clearly at the point of upload.
The Photographer Who Fought and Lost
In March 2026, PetaPixel published the account of Gerald Carter, a stock photographer who runs Diversity Photos and had contributed extensively to Adobe Stock. Carter spent 18 months attempting to stop Adobe from using his photo library to train Firefly, Adobe’s generative AI image model.
In June 2024, Carter filed for arbitration, the dispute resolution mechanism specified in Adobe’s terms of service. Adobe’s terms prohibit class actions and direct lawsuits, leaving arbitration as the sole recourse for individual contributors. When Carter raised an inability to pay the arbitration costs, Adobe filed a motion requesting $24,000 in fees just to have that inability-to-pay claim reviewed. Facing that cost barrier, Carter withdrew his arbitration claim. Adobe has since cited the outcome as precedent in similar disputes.
What the Arbitration Outcome Means in Practice
Adobe’s terms of service, like those of many large cloud platforms, contain an arbitration clause that routes individual disputes away from courts and into a private forum. The clause is framed as a protection for users against the cost and delay of litigation. The Carter case shows how the mechanism can work in the opposite direction: Adobe was able to impose a five-figure cost on a small contributor simply for seeking review of his financial standing, before any substantive hearing took place.
The outcome is not unique to Adobe. Any cloud platform that processes user-uploaded files and includes similar terms can deploy the same mechanism. The legal right to challenge how your uploaded content is used exists in principle. The practical ability to exercise it is subject to cost structures that most individual users cannot sustain.
What Never Uploading a File Prevents
The Adobe dispute involves two distinct harms: the use of uploaded and licensed content for AI training, and the structural inability to challenge that use effectively. Both harms require the same precondition: that the platform has received and retained your files.
A file processing tool that runs entirely in the browser using WebAssembly never receives the file. Nothing is transmitted to Adobe’s infrastructure, or any other service’s infrastructure. There is no content to index, no training pipeline to feed, and no terms-of-service clause governing what happens to your document after upload, because there is no upload.
The SlimLM lawsuits and the Carter arbitration case describe what the relationship between a user and a cloud platform looks like once the platform holds your files. Browser-based processing is defined by the absence of that relationship.
Sources
- He Tried to Stop Adobe From Training its AI on His Photo Library. He Lost. (PetaPixel)
- Adobe hit with proposed class-action, accused of misusing authors’ work in AI training (TechCrunch)
- Adobe Faces Another Suit Over Alleged AI Training Piracy (Law360)
- Adobe faces class action over SlimLM training (Complete AI Training)
- Photographers Leaving Adobe Behind (The Phoblographer)