Anthropic's Pirate Bill Came Due at $1.5 Billion

The largest copyright settlement in American memory is a bill for shoplifting. The reading got a softer hearing. On Monday, U.S. District Judge Araceli Martínez-Olguín approved Anthropic's $1.5 billion class deal with authors whose books fed Claude. Plaintiffs' counsel call it the largest known copyright recovery. Authors and publishers stand to collect about $3,000 per work across a class covering roughly 500,000 titles.

The claim numbers matter. Martínez-Olguín found that about 95 percent of the class received notice and roughly 91 percent of affected authors and publishers have already filed claims. Only 350 class members opted out. Dozens of objections failed. Late opt-outs mostly failed too. The judge cut the lawyers' fee request and the named plaintiffs' service awards, then still called the deal meaningful relief and granted final approval.

5 min read

The intellectual spine of the case belongs to retired Judge William Alsup. Last summer he split the fight in two. Training a model on copyrighted books, he held, can qualify as fair use. Building that library by vacuuming pirate sites such as Library Genesis and Pirate Library Mirror did not. Purchased and scanned books were fine. Napster-style mass downloads were a separate wrong that could have gone to a jury. Anthropic settled rather than try that theory in front of one.

Anthropic general counsel Aparna Sridhar told Reuters the lab is pleased that more than 91 percent of covered authors and publishers have claimed their share. The company still points to Alsup's fair-use holding as industry-shaping. That is the clever part of the public message. The check clears the piracy count. The opinion keeps the training story alive.

Authors who wanted a bigger moral win are right to feel half-served. Three thousand dollars per book is real money and a thin price for a permanent statistical ghost of a career. Alsup's fair-use ruling remains a single district-court decision. Because Anthropic settled, no appeals court will harden or kill that holding in this case. Other suits continue, including parties who say the per-book figure is too low.

Still, the settlement draws a bright operational line. How you acquire the corpus can bankrupt you even when what you do with the text later looks transformative in court. Labs that treated shadow libraries as free infrastructure now have a public invoice. Labs that buy, license, or scan with receipts have a safer path. The philosophy of machine learning did not get settled Monday. The logistics of theft did.

Call it a culture industry bargain with machine learning. Authors get paid for the raid. Anthropic keeps the weights. Copyright law, for one afternoon in San Francisco, remembered that a library built from stolen copies is still a stolen library, even when the reader is a model.