from the can’t-risk-it dept
In what may be the least surprising news in the world of copyright and the internet, Anthropic just agreed to settle the copyright lawsuit that everyone’s been watching, but not for the reasons most people think. This isn’t about AI training being found to infringe copyright—in fact, Anthropic won on that issue. Instead, it’s about how copyright’s broken statutory damages system can turn a narrow legal loss into a company-ending threat, forcing settlements even when the core dispute goes your way.
Anthropic had done something remarkably stupid beyond just training: they downloaded unauthorized copies of works and stored them in an internal “pirate library” for future reference. Judge Alsup was crystal clear that while the training itself was fair use, building and maintaining this library of unauthorized copies was straightforward infringement. This wasn’t some edge case—it was basic copyright violation that Anthropic should have known better than to engage in.
And while there were some defenses to this, it would likely be tough to succeed at trial with the position Judge Alsup had put them in.
The question then was about liability. Because of copyright’s absolutely ridiculous statutory damages (up to $150k per work if the infringement was found to be “willful”), which need not bear any relationship to the actual damages, Anthropic could have been on the hook for trillions of dollars in damages just in this one case. That’s not something any company is going to roll the dice on, and I’m sure that the conversation was more or less: if you win and we get hit with statutory damages, the company will shut down and you will get nothing. Instead, let’s come to some sort of deal and get the lawyers (and the named author plaintiffs) paid.
While the amount of the settlement hasn’t been revealed yet, the amount authors get paid is going to come out eventually, and… I guarantee that it will not be much. The expected “class” in this case included about 7 million works. If you went absolutely crazy and said that the settlement was $1 billion (which I highly doubt it would be), and the lawyers take half of that, that means authors would receive… about $70 per work? And if I had to guess, the amount will be significantly below that.
This pathetic payout perfectly illustrates why the whole “we need licensing for AI training” argument is economically illiterate. Even setting aside that Judge Alsup already said training is fair use, the math just doesn’t work. When you’re dealing with millions of works, each individual piece contributes essentially nothing to the final model. The transaction costs alone would eat whatever pennies authors might receive. The only winners in any licensing scheme would be the collection societies and lawyers skimming off the top.
Instead what will happen—what always happens with these collective licensing deals—is that a few of the bigger names will get wealthy, but mainly the middleman will get wealthy. These kinds of schemes only tend to enrich the middlemen (often leading to corruption).
So this result is hardly surprising. Anthropic had to settle rather than face shutting down. But my guess is that authors are going to be incredibly disappointed by how much they end up getting from the settlement. Judge Alsup still has to approve the settlement, and some people may protest it, but it would be a much bigger surprise if he somehow rejects it.
Filed Under: ai, authors, copyright, fair use, licensing, settlement, statutory damages, william alsup
Companies: anthropic
