Big numbers often get thrown around in the aftermath of legal battles, as judges hand down judgements—or attorneys arrange settlement amounts—in the tens, or hundreds, of millions of dollars. Still, even jaded legal observers can occasionally run into a genuinely daunting number while parsing this stuff. Like, say, the $1.5 billion settlement that AI company Anthropic has agreed to pay in the ongoing class-action suit against it, launched by authors who said the company infringed on their copyrighted works by feeding them as training data to its “AI assistant” Claude. Sure, parts of that sum (calculated at $3,000 per work for a staggering number of works, and with its first $300 million installment due just five days after the settlement is approved) might potentially vanish in a puff of future bankruptcy. But it’s still the “largest publicly reported copyright recovery in history,” according to legal documents from the authors’ attorneys.
That being said, the win here on the wider AI front is quite a bit less clear than “hand our clients the annual estimated GDP of Grenada” might suggest. Yes, U.S. District Judge William Alsup set the stage for Anthropic to eat that massive price tag by ruling that the company clearly violated copyright agreements via how it acquired the books it fed into its own personal woodchipper. (I.e., downloading pirated datasets of millions of books that had been floating around the internet.) And, yes, the settlement will require Anthropic to destroy those “shadow library” datasets in its possession. (But notably, with no actual changes to the Claude large language model itself.) Most critically, though, back in June, Alsup also ruled that “reproducing purchased-and-scanned books to train AI” falls under fair use, calling the case “exceedingly transformative” as a justification for the designation.
As such, both sides in the fight issued statements claiming a form of victory today, with the authors’ side focusing mostly on the massive size of the settlement amount. Anthropic, meanwhile—which has been backed in the past with more than $6 billion in contributions from Amazon and Google—focused its statements on the legal precedent it achieved in the case: “In June, the District Court issued a landmark ruling on AI development and copyright law, finding that Anthropic’s approach to training AI models constitutes fair use. Today’s settlement, if approved, will resolve the plaintiffs’ remaining legacy claims.” What this likely means is that AI companies aren’t going to slow down—especially with, say, a $1.5 billion mortgage suddenly hanging over their heads—but simply become a lot more choosy about how they get their training data.
[via Deadline]
