Love in the Time of LLMs: A Romance Novelist's Experience With the Anthropic Suit

Part 1: Pirate booty?

“Anthropic might owe you money.”

I forget whether I got the email from my then-agent or my more-up-on-current-events friend first. Until then, though, I’d been only casually aware of the major class action lawsuit against Anthropic, which had trained a version of Claude on a vast number of pirated books. AI companies were getting sued all the time, and it wasn’t like I could do anything about the verdict. I think I’d heard that the judge had ruled against Anthropic, but hadn’t paid much attention beyond that.

Suddenly, I had a much more personal interest.

Like many millennials, I have the sort of career best described as three jobs in a trenchcoat, and one of those jobs is writing fantasy romance novels. I’m no Stephen King (or Jessica Fletcher, alas), but by the time Anthropic settled its case, a publisher had brought out eleven of my books. Both my agent and my friend sent me a link to the settlement database, where I found that two of those books had shown up on piracy sites, and Claude had predictably scooped them up.

This doesn’t make me special—or suddenly rich. While Anthropic is paying out $1.5 billion in total, at least 300,000 of us authors are splitting it, as well as dividing the cash with our publishers. My final take will probably come to about $3,000 when the government and the lawyers take their share: enough for a nice vacation, an extremely used car, or a month of rent if I still lived in Boston. The three people who actually brought the case spent much more time and effort on it, and the payoff for each of them isn’t much more than mine.

Obviously, I’m grateful for their efforts, and not just because my bank account benefits. The case sets at least a little much-needed precedent concerning the intersection of internet piracy and AI training. I’ve been thinking lately about both the specific circumstances around the settlement and my general feelings regarding Anthropic, probably the most ethical of the LLM companies.

Part 2: It’s complicated

If you read only the headlines, you could reasonably think that the Anthropic settlement means that generative AI companies can’t scrape content as widely as they did. Maybe you’d assume that any future model training will depend on licensed or out-of-copyright material; licensing has become a major subject of discussion among authors lately, after all.

As with everything else to do with generative AI, the case isn’t quite that decisive.

The judge ruled that Anthropic opened itself up to liability because it used pirated works, and piracy has been grounds for lawsuits since I was in high school and listening to Napster (sorry), long before generative AI hit the scene. Companies can still train their LLMs on books without the authors’ or publishers’ consent. They just have to buy a copy of whatever they use. The recent news that Amazon is destructively scanning rare books brought more public attention to this situation—and the Anthropic case established destructive scanning itself as legal.

As anyone who spent too much time in college listening to the Les Misérables soundtrack knows, however, the law and morality are very different things.

Part 3: Ethical fog

Is it unethical to destructively scan books? Most people who haven’t worked in libraries or bookstores have a visceral reaction to any books being trashed: it brings up images from the Nazis to Fahrenheit 451 to the Satanic Panic. On the other hand, libraries often have to destroy books due to limited space or condition issues, while bookstores return unsold mass-market copies to be remaindered and pulped (a practice that has become much more common since a 1979 decision about tax law, so maybe we can blame Warren Burger for the whole mess and call it a day). On the other other hand, Amazon at least is destructively scanning rare books, which is a much thornier issue, and Anthropic’s been accused of the same thing.

Is it unethical to train LLMs on unlicensed content, even if you legally purchase a copy? Proponents of such training say that all creators use other media in some way, and ask if an LLM trained on a novel is that different from a human author drawing inspiration from what they read. (Our own Arsen Revazov asked major models to weigh in on related issues a few months ago, and you can check out their answers here: A Council of Robots: Five AI Models Debate the Death of the Author — SilentRoom Journal.) As an author myself, I can confirm that there’s really no such thing as completely original media, and I’m fine with that… but, as others point out, transformative work feels different coming from a fellow coffee-and-profanity-fueled creator than it does from a machine that endangers our already fragile income streams.

How similar can phrases, plot structures, or other elements be before crossing a moral line? Is direct competition different from related works—for instance, if an LLM uses my prose to write a line of romance novels, should I respond to that the same way I would if it were creating AI significant others for an app?

I know my reactions. I couldn’t, and wouldn’t presume to, say how other authors should feel.

As if all that isn’t enough, there’s also the question of the specific companies involved.

Part 4: The corporate limbo contest

Let’s get one thing out of the way: Corporations are not your friend. That’s especially true of most tech companies, and even more so of the big AI firms. Concerns about environmental destruction, plagiarism, and a host of other cyberpunk-dystopia-without-the-cool-outfits topics are absolutely valid and worth discussing, and every company that puts out massive LLMs has a track record that would make it a good villain in a 1980s Saturday morning cartoon.

On that scale, where the bar is so low it’s serving drinks in Hell, Anthropic is probably the least worst company around. It was the only major player that didn’t roll over for the Trump administration, it scores generally better (or, really, less bad) on ESG issues, and nobody in its leadership is instantly recognizable in connection with horrible yacht-adjacent stories, conspiracy theory promotion, or whatever Altman’s whole deal is. If any of the big AI corporations have to stick around, Anthropic would definitely be my pick for survivor.

Don’t get me wrong: I support the lawsuit and I’m taking their money. It does strike me, though, that Anthropic actually did settle and pay out. Meta dragged a similar case out to the bitter end and managed to win, Google is now facing a lawsuit about Gemini, and a bunch of authors (including George R.R. Martin) are suing ChatGPT. If any of those models used my work, I doubt I’ll see payment any time soon.

As for Grok, it’s dealing with a… different kind of litigation. Besides, my work encourages women to have both fun and standards: I can’t imagine Elon Musk’s brainchild would pick it up.