Adobe lawsuit over AI training: what working writers need to know
Adobe is facing a proposed class action in California federal court claiming it trained its SlimLM models on copyrighted books without permission. The suit, filed on December 17, 2025 by author Elizabeth Lyon, says Adobe used pirated copies of her instructional books and others to train SlimLM, a small language model used for document tasks on mobile devices. This is Adobe's first case of this type, amid similar claims against OpenAI and Anthropic. Monetary damages are being sought, and the outcome could set direction for how training data is sourced.
Why this matters to writers
The complaint points to Adobe's training data lineage: SlimLM was trained on SlimPajama-627B, which the filing says was made by copying and altering the RedPajama dataset. RedPajama has been linked to Books3, a 191,000-title collection at the center of copyright disputes, and has been cited in lawsuits against Apple and Salesforce. That chain raises a core issue: can companies that rely on "open" datasets be held liable if the compiler included unauthorized books? Courts will be weighing who bears responsibility along the data supply chain.
The bigger picture
Adobe has publicly promoted the use of licensed, public-domain, and Adobe Stock content for safer commercial outputs. But even one contaminated source in the training mix can drag a model-and its users-into risk. The market is shifting toward fully licensed datasets, stronger provenance, and clear opt-outs for authors. Expect more scrutiny on dataset origin, contracts, and audit trails.
Practical steps to protect your work
- Register your books and major works. Registration strengthens your position for statutory damages if infringement is found. Start here: U.S. Copyright Office.
- Check and update your contracts. Add clear clauses that forbid licensing your work for AI training without explicit written permission. Require attribution, audit rights, and notification if your content is sublicensed.
- Use opt-outs where available. Some AI companies offer forms to exclude works from future training. Keep confirmations for your records.
- Monitor datasets linked to Books3 and its derivatives. Search public dataset indexes and mirrors for your name, titles, or unique phrases. Keep screenshots and timestamps of anything suspicious.
- Keep clean records. Maintain dated drafts, ISBNs, submission emails, and publication proofs to establish authorship and timelines.
- Join collective efforts. Groups coordinating policy and defense can amplify your voice and share updates: Authors Guild: AI and Authors.
For editors and publishers working with writers
- Insist on licensed training data if you deploy AI tools. Avoid datasets with Books3 links unless clear permissions exist.
- Add warranties and indemnities to vendor contracts. Require data lineage documentation, author opt-out support, and takedown processes.
- Consider licensed data providers and provenance tools. Vendors like SyndiGate (noted in the complaint's context) and data-lineage trackers are gaining importance as compliance expectations rise.
Opportunities to watch
- Licensed dataset vendors. Demand is growing for multilingual, rights-cleared corpora with auditable provenance.
- Provenance and compliance software. Tools that trace data origin, verify licenses, and track opt-outs will become standard for enterprise AI.
Bottom line for writers
This case won't be the last. The training-data supply chain is under the microscope, and liability may extend beyond the original compilers. Protect your catalog now, tighten your contracts, and keep an eye on how courts assign responsibility. Staying informed today can save you hassle-and money-later.
Level up your AI literacy without risking your rights
If you want structured ways to work with AI while staying compliant, explore curated paths here: Complete AI Training: Courses by Job.
Your membership also unlocks:
