Ex-WSJ reporter who exposed Theranos sues AI giants over alleged book piracy - what writers need to know
John Carreyrou, the Pulitzer-winning journalist whose reporting helped bring down Theranos and Elizabeth Holmes, has filed a federal lawsuit in California against six major AI companies. The complaint alleges that Google, xAI, OpenAI, Meta, Anthropic, and Perplexity used copyrighted books to train large language models without permission or payment.
Carreyrou, author of "Bad Blood," filed a single joint complaint alongside five other writers. It's the first copyright case to name xAI as a defendant and adds to a growing wave of lawsuits claiming that AI systems were trained on pirated books.
A spokesperson for Perplexity said the company "doesn't index books." The other defendants did not immediately respond to requests for comment, according to Reuters.
Why this case matters for writers
- The core claim hits home: books allegedly ingested to build lucrative AI products, with no consent or compensation.
- The suit targets the economic engine of modern AI - training data. If the court agrees that ingestion infringes, payouts and practices could shift.
- This is not a class action. The authors argue class cases let defendants settle cheaply. The complaint says, "LLM companies should not be able to so easily extinguish thousands upon thousands of high-value claims at bargain-basement rates."
What's different about this lawsuit
- First to name xAI as a defendant, widening the legal front.
- Deliberate choice to avoid class status. The plaintiffs say class actions dilute individual claims and cap exposure.
- Direct criticism of a recent Anthropic settlement. The filing argues class members would receive "a tiny fraction (just 2%) of the Copyright Act's statutory ceiling of $150,000" per infringed work - and the authors opted out.
Key details at a glance
- Defendants: Google, xAI, OpenAI, Meta, Anthropic, and Perplexity.
- Core allegation: books were pirated and used for training LLMs without permission or compensation.
- Potential stakes: statutory damages can reach up to $150,000 per work in some cases. See 17 U.S.C. § 504 for how damages are assessed (source).
- Context: Carreyrou is now at the New York Times. The case was filed by attorneys at Freedman Normand Friedland, including Kyle Roche.
The broader fight over training data
Authors and publishers have been filing suits that challenge how AI companies collect and use text, images, and other works. This complaint keeps the focus tight: books as high-value assets allegedly ingested at scale.
The filing frames the industry's growth as built on "stolen intellectual property" - a claim the defendants are expected to contest. Expect heated arguments over fair use, data sourcing, and what "training on" a work actually means in legal terms.
Practical steps for authors right now
- Register your works. It's the strongest way to enforce rights and seek statutory damages if needed (U.S. Copyright Office).
- Audit your contracts. Check what your publisher or agent allows regarding digital copies, licensing, and dataset exposure.
- Preserve evidence. Keep drafts, timestamps, ISBNs, and publication records. If you see AI outputs that closely mirror your text, document them with dates and screenshots.
- Watch class action notices. If you receive one, evaluate whether opting in or opting out serves your interests. The payout math and control over your claim matter.
- Police unauthorized copies. Use DMCA takedowns for illicit mirrors of your books. Reducing pirated copies can limit dataset exposure.
- Set alerts. Track your titles and name across the web and in public dataset discussions. Early signals help.
- Ask your platforms about training policies. Some services offer opt-outs for web content. It won't fix past ingestion or book piracy, but it can limit future use.
Legal and industry signals to monitor
- Motions to dismiss. Courts may weigh in on whether training constitutes infringement at the pleadings stage.
- Discovery. If allowed, it could expose sourcing practices, dataset composition, and internal knowledge about book ingestion.
- Settlement structures. Are offers per-work, per-author, or pooled? Do they reflect potential statutory damages or a small fraction?
- Court commentary. In a related proceeding, a judge criticized efforts to steer authors away from a class settlement toward "a sweeter deal." Expect ongoing scrutiny of how lawyers organize authors.
Bottom line for writers
This case pushes on the core question: who gets paid when books become inputs for AI products. If courts side with authors, compensation frameworks and licensing markets could emerge. If not, authors will need stronger self-help strategies and tighter contracts.
Either way, treat your catalog like a business. Register your works, track their use, and keep your options open as the rulings come in.
Further reading and resources
- Statutory damages overview: 17 U.S.C. § 504
- Skills and tools for writers working with AI (ethics, prompts, and workflows): Complete AI Training - Courses by Job
Your membership also unlocks:
