An AI Meets Copyright Update: Settlement on the Horizon
Once I get started on a topic, I have to update new developments. So, here we go on the AI and copyright situation.
Background
In two earlier posts, we looked at the Anthropic copyright case. In a first ruling, the court noted that training an AI model on lawfully obtained books can be transformative and possibly fair use, thus avoiding copyright liability. That was good news for AI.
Then came the bad news for Anthropic. The court found Anthropic’s massive downloads from pirate libraries like LibGen and Books3, involving millions of titles, were most certainly not lawful. So, training was probably okay, but pirating was copyright infringement. And that exposed the company to infringement liability on a scale that could have reached the billions at statutory damages of $150,000 per book with millions of books.
The settlement
Now for the update. Anthropic has agreed to a 1.5-billion-dollar settlement. That’s a record-setting copyright recovery, but it provides a way out of the astronomical statutory damages the company might otherwise have faced.
Still unresolved
The case isn’t over yet. The deal still needs court approval, and the court has already pressed for more transparency in the claims process. Settlement of class action claims is always a complicated process, and the judge is skeptical that the proposed deal will be fair to all authors.
And those interested in how copyright law impacts AI are left hanging. The trial court’s observations that training AI with purchased books can be fair use was of great interest, but it was not a final decision. And we don’t know what an appellate court would have done with that issue. If the settlement goes through, we’ll have to wait for the next case to find out.
Where we are now
It looks like many authors will see compensation, and Anthropic will avoid the risk of open-ended liability if the Court approves the class action settlement. But clear boundaries for fair use in the AI era have yet to be drawn. Still, we can be be pretty sure that no court is likely to tolerate wholesale copying of pirated works.