📰 Key Takeaways

Anthropic reached a $1.5 billion class-action settlement last year, stemming from a court ruling that training AI models on copyrighted content is fair use, but obtaining that content through piracy is not. The settlement got final approval this July, and payments are now rolling out. Under the terms, authors of nearly 500,000 books will receive $3,000 for each work that was pirated. If a book is still in print with its original publisher and actively being sold, the payout is split 50/50 between author and publisher. If it’s self-published, or the publisher has already reverted the rights back to the author (book out of print), the author is entitled to the full amount. But lately a lot of authors have been speaking up on social media, claiming publishers appear to be claiming more than their fair share. Suspense/thriller author April Henry called out HarperCollins after the publisher filed a claim for one of her books whose rights had been reverted to her at least 17 years ago — and on the same day, she got a credit alert flagging the publisher as her “employer,” even though she’s never worked for the company. Victoria Strauss of the author advocacy blog Writers Beware says the complaints she’s hearing from authors mostly fall into two buckets: publishers claiming money for books whose rights were already reverted, and publishers claiming the full 100% when they’re only entitled to 50%. Strauss says she’s not ready to chalk this up to bad faith — she thinks it’s more likely sloppy record-keeping, and some publishers have already admitted the errors and asked Anthropic to correct them. Mary Rasenberger, CEO of the Authors Guild, told the New York Times she doesn’t think publishers are deliberately trying to grab extra money — she sees it as a predictable outcome of a messy settlement process combined with poor recordkeeping. That said, Strauss notes she’s seen an unusually high volume of complaints over the past two days, with the same errors showing up again and again — suggesting this might not just be a minor systemic glitch, but a broader structural problem. On top of that, some literary agencies have also been caught filing for a cut, which Strauss finds surprising since agents aren’t the actual copyright holders of the books.

💬 JudyAI Lab Take

Anthropic’s infringement settlement is now caught up in a payout dispute: authors are accusing publishers of over-claiming settlement funds, in some cases even filing for books whose rights had long since reverted back to the author — and it’s drawing plenty of attention online.

This whole situation points to something easy to overlook — copyright disputes over AI training data don’t just wrap up once the settlement money starts flowing. If anything, the “execution layer” is where more structural problems tend to surface. When a payout process involves a big web of stakeholders (authors, publishers, agents), unclear rights records and fuzzy revenue-split rules can easily spark a whole new round of disputes. For AI builders, this is a reminder that the licensing chain behind training data needs a much more transparent tracking system — signing off on fair use or a settlement agreement isn’t where the responsibility ends. The rights status of a data source should be treated as something that needs ongoing verification, not a static fact you check off once.

Actionable takeaway: if your project involves licensed training data, it’s worth setting up a clear record of who owns what rights and a way to track it over time — it’ll save you from disputes down the road.

📅 Source Info

🔗 Further Reading