Genre Grapevine on the Anthropic AI Settlement

Reminder: If you like my journalism or my fiction writing, please consider becoming a paid subscriber to my Substack.

The Friday news dump is a time-honored tradition in politics and business, where bad news is released at the end of the week because it gets less scrutiny. So I honestly wasn’t surprised that the settlement terms for the class action lawsuit brought by authors against Anthropic for pirating millions of books was released late on a Friday.

At first glance the settlement sounds amazing, with the company paying $1.5 billion for having violated copyright on an estimated 500,000 books. That equals $3,000 per book before the payment of legal fees. The settlement, which can be downloaded here, notes this amount is “four times the minimum statutory damages amount and fifteen times the minimum statutory damages amount for innocent infringement.”

However, that amount’s a far cry from the maximum penalty for violating copyright, which is $150,000 for each work infringed. It’s almost like the Anthropic settlement is saying that if you violate the copyright of authors, be sure to do it in bulk so you can get the most bang for your buck.

In addition, the settlement of $3,000 per work is for 500,000 works, not for 500,000 authors (despite what this New York Times article states). That may sound like a lot of books. However, the Federal judge overseeing the class action ruled earlier this year that Anthropic pirated more than 7 million digitized books to train the company’s AI. While the settlement will require Anthropic to delete its entire pirated databases of books, most of those books won’t be included in the settlement payments.

In fact, the settlement payments barely cover 7% of the books Anthropic pirated. And Anthropic gets to keep the generative AI system the company trained on those stolen books.

The reason more authors aren't in the settlement is because, per the settlement language, "works that satisfy criteria for inclusion in Class" must have a "title, author(s), publisher, an ISBN and/or ASIN, and United States copyright registration number." Those final words are critical. Basically, only works with an official copyright registration with the U.S. Copyright Office are included.

Remember all those publishers who didn't register the copyrights for their authors? That's going to bite those authors in the ass. And yes, I'm included in this screwing because many of my books weren't officially registered.

In some good potential news, the settlement states "If the Works List ultimately exceeds 500,000 works, then Anthropic will pay an additional $3,000 per work that Anthropic adds to the Works List above 500,000 works." So 500,000 works isn't an arbitrary cut-off. This means it will be vital that authors who have copyright registrations make sure they are included.

In other good news, it appears the settlement doesn’t prevent the authors of works who aren’t included from filing a new class action lawsuit. Per the settlement language, “the Settlement only releases claims for ‘works on the Works List.’ Id. So if a Class Member owns multiple works, some of which are on the Works List and some of which are not, the release covers only those works on the Works List and nothing else.”

Also worth noting, as the Science Fiction and Fantasy Writers Association explained in an email to members, that the final list of works covered by the settlement “will differ substantially from the list on The Atlantic website” and that this final list “will not be finalized and released until early October.” This is an important note because the list of pirated works published by The Atlantic is the one most writers have been searching to see if their works were pirated.

Is Short Fiction Included?

Sadly, it is unclear if short story writers with copyright registrations will be included in this settlement. I wrote a thread last week exploring how difficult it can be for short fiction authors to know if their stories were pirated, along with how to discover if your stories are in the pirated database. In short, authors with works in anthologies and multi-author book collections may not be easily discoverable in the pirated databases.

According to the settlement language, the class action covers “Any work possessing an ISBN or ASIN which was registered with the United States Copyright Office within five years of the work’s first publication and which was registered with the United States Copyright Office before being downloaded by Anthropic or within three months of publication.”

Short fiction and other shorter works published within books with multiple authors might not be covered because those shorter works don’t have an individual ISBN or ASIN, which would have instead been granted to the overall book. Also, what if the book that published or reprinted your short fiction didn’t get registered with the copyright office? Are you covered even if your individual story in the book received its own copyright registration, or was copyrighted within another book such as a short story collection?

The settlement also doesn’t cover works published in magazines, which instead of an ISBN or ASIN number receive an ISSN number. Many short fiction and other magazines have received copyright registration over the years, but the way the settlement was written they appear to also not qualify.

All of this is of extreme interest to me because my short story "The Eight-Thousanders" has been reprinted in anthologies that were pirated and included in the book databases such as The Best Horror of the Year Volume Thirteen edited by Ellen Datlow, The Longlist Anthology, Volume 7 edited by David Steffen, and Nebula Awards Showcase 56. I have a copyright registration for the story but the way the settlement is written it appears I may not be included.

Let me repeat that: Even with a copyright registration, it looks questionable on if I’ll be included. I suspect other authors will be in the same boat as I am.

Of course I'll push like hell to be included in the settlement, but we’ll see how that goes.

What Does The Settlement Mean?

Mary Rasenberger, CEO of the Authors Guild, called the settlement “an excellent result for authors, publishers, and rightsholders generally, sending a strong message to the AI industry that there are serious consequences when they pirate authors’ works to train their AI, robbing those least able to afford it.” The Human Artistry Campaign said the settlement “is a huge victory not only for the authors involved, but for all writers, artists, and creators who know that their work has value and their rights should be respected when it comes to AI uses.”

And they are correct – the settlement is a warning that tech companies can’t simply pirate books without legal consequences. But I also feel like the Authors Guild and the plaintiffs in this lawsuit took the easy way out with their settlement. Instead of working to include as many writers as possible in the settlement, they instead agreed on arbitrary and minimal definitions of which books would be included. It’s almost like they wanted a quick win – any win – even if they screwed over most authors.

In addition, the Anthropic settlement isn't a true punishment for a company that stole seven million copyrighted works. While $1.5 billion sounds like a lot, Anthropic is worth $183 billion. $1.5 billion is barely a slap on the wrist for a company worth that much.

And only $3,000 per work for 500,000 pirated books? Compare that to the massive penalties inflicted on individuals in previous decades for pirating copyrighted music, such as the $675,000 penalty against a Boston University student who downloaded 31 pirated songs as a teenager or the Minnesota woman who paid $220,000 for 24 illegally downloaded songs. Does that sound fair or proportional?

Essentially Anthropic is writing a check so they can keep doing what they want to do. I wouldn't be surprised if other tech companies that pirated copyrighted writings and art look at this settlement and decide to follow suit. They'll see it as a relatively cheap way to remove a big chunk of their legal liability while continuing to build their generative AI systems.

Again, the settlement does set expectations that any company that pirated copyright works must pay authors and artists. But an expectation based on a settlement isn’t actual legal precedent, as a ruling by the Federal judge in this case would have been. The groups pushing for this settlement went for the quick win and in the process screwed over the vast majority of authors.

But the settlement is also all authors have at this moment. Lieff Cabraser, the law firm managing the lawsuit, has set up a new website for authors to submit their information if you believe you might be covered by the class action.

Sadly, the intake form doesn’t ask authors if they’re holding their nose while submitting their information.