Anthropic’s historic $1.5 billion settlement with authors, announced Sept. 5, might seem like a John Grisham legal thriller, establishing a possible benchmark to compensate copyright holders for using their work to train AI models.
But for me and many other authors, it’s more like an episode of Antiques Roadshow — pulling a forgotten artifact off the shelf and having it reappraised in a new light.
If only someone could tell us whether our dusty old books are worth anything.
The settlement, which must still be approved by a federal judge, is designed to resolve a massive class action lawsuit involving Anthropic’s alleged use of pirated copies of about 500,000 copyrighted books to train the AI model behind its Claude AI application.
The ultimate outcome could have big implications for the other giants of the emerging AI economy, including Microsoft, OpenAI, and Amazon, which has invested billions into Anthropic as part of its broader push into generative AI.
According to the Authors Guild, the $1.5 billion award will be split among the rightsholders (including both authors and publishers) of “all of the books included in the class after administration fees, lawyers’ fees and expenses are paid.”
On average, that boils down to about $3,000 per book.
This is where my Antiques Roadshow analogy comes in. I wrote, co-authored, or contributed to 15 books between 1980 and 2000. Getting a settlement of $3,000 per book for even a few of them would be a very welcome surprise for some old assets I didn’t think would ever yield anything further.
As an author, you can get an idea of whether one of your books might be part of the settlement by using a search tool published by The Atlantic earlier this year. None of my books showed up in the search results from that tool, but several books by my sister — historian, author, and academic Dr. Julie Wheelwright — did appear.
The Atlantic’s search tool will only tell you if your book is listed on Library Genesis (“LibGen”). It is one of the two pirate sites — the other being Pirate Library Mirror (“PiLiMi”) — named in the settlement.
I did my own additional investigation to see if any were potentially on the second site, PiLiMi. That involved using a browser in private browsing mode and a VPN to go to a site called “Anna’s Archive.” The Authors Guild says PiLiMi is a mirror of Anna’s Archive. And one of my books was there: a slim volume I wrote in 1987 called “A Personal Guide to Personal Computing.”
None of this, however, will tell you for certain if your book was allegedly downloaded and used by Anthropic for AI training — or if it meets the other criteria listed by the Authors Guild for inclusion in the settlement.
What’s next in the case
Figuring out which books merit payouts from the settlement — and how to accurately and properly distribute the payout — is a huge job, which is still ongoing (and one of the reasons the settlement is not yet approved).
U.S. District Judge William Haskell Alsup, who is overseeing the case, made clear how uncertain things would have been for all parties if it had gone to trial — describing outcomes ranging from a total bust to as much as $150,000 per copyrighted work.
So where does that leave authors? The preliminary approval hearing will not happen until Sept. 25, so we’re in the dark for now. There’s no easy way, at this point, to tell if a given book is one of the 500,000 that will be part of the settlement, because the lists were filed under seal to keep author and publisher information private.
Authors Guild CEO Mary Rasenberger explained in an interview that the legal team for the authors is rushing to complete all the work and answer all the questions from the judge in time for the Sept. 25 hearing.
“The turnaround time on this is kinda nuts,” she said. “The lawyers have been making Herculean efforts. Anthropic gave us a huge list and we had to connect to each title with each copyright registration number.”
As it works through all of this, the Authors Guild has also is directing authors to a dedicated settlement web site to collect names and contact details of any authors who are not members of the organization. It urges any author who suspects their work may be part of this settlement to provide their information.
The larger meaning of the deal
The impact of this settlement could extend well beyond an unexpected bonus for authors.
Regina S. Penti, technology and IP transactions partner at the law firm Ropes & Gray, says that while the Anthropic settlement quantifies the risk of using allegedly pirated data, it’s just one chapter in a much bigger story. The legal landscape is still evolving.
While the settlement is a major milestone, it’s not binding precedent for other courts or cases, said Penti, who co-leads of the firm’s Artificial Intelligence Industry Group.
“Still, it’s a strong signal that courts are taking the source of training data seriously,” she said. “Other AI companies and rights holders will be watching this outcome closely as they negotiate, litigate, or set policies.”
For authors like my sister, the AI training debate is about recognizing the years of work required for a single groundbreaking idea. She makes the case for why books aren’t just data to be consumed as AI training fodder without consultation, consent or compensation.
In her research on the Mata Hari, for example, she consulted archives in seven countries, including the United States, Britain, France, and Russia. To understand what happened to the Dutch double-agent during the First World War, she collaborated with colleagues to analyze and piece together a complicated set of often contradictory statements.
But for all this effort, she often sees her research appearing online without attribution. What often goes unappreciated, she says, is “how much sheer effort has gone into bringing about that single groundbreaking idea.”
And that, in the end, is what this settlement is truly about.