US judge approves Anthropic’s $1.5bn copyright settlement

TL;DR:

• US District Judge Araceli Martinez-Olguin granted final approval to Anthropic’s $1.5bn settlement with authors who accused it of misusing their books to train Claude.
• The judge overruled objections that the sum was too small, calling them “not grounded in a realistic assessment of the overall risks and rewards of a trial”.
• Anthropic says more than 91% of covered authors and publishers have claimed their share; some opted out and are pursuing separate cases.

The first major US training-data case to settle now has a court-approved number attached to it, which makes it the reference point every subsequent claim will be argued against. Attorneys were awarded just over $101m of the $187.5m in fees they had requested.

The underlying dispute produced an unusual split. Judge William Alsup, now retired, ruled last June that Anthropic made fair use of the authors’ work to train Claude — a finding Anthropic deputy general counsel Aparna Sridhar noted “remains the law today” — but found the company violated authors’ rights by retaining more than seven million pirated books in a central library not necessarily used for training. A trial to determine damages for that retention had been scheduled for last December, with exposure running into the hundreds of billions of dollars.

That asymmetry is what the settlement price reflects. Training was permitted; acquisition was not.

Why the number travels

Lead plaintiffs’ attorney Justin Nelson called it “the largest known copyright recovery in history”. For UK rightsholders the figure matters more than the jurisdiction. It establishes what a defendant with genuine trial exposure will pay rather than test a piracy finding in front of a jury, and it does so while the fair use question sits unresolved in Britain — where there is no US-style fair use doctrine, and where the government’s text and data mining proposals remain contested by publishers and the creative industries.

UK publishers pressing their own claims against Google now have a settled comparator where previously they had only pending litigation.

Looking forward

The dozens of remaining cases brought by authors and news organisations are unaffected by this approval, and the authors and publishers who opted out are still litigating against Anthropic directly. What has changed is the negotiating baseline. Any AI developer that acquired training material through channels it cannot document now knows roughly what the market rate for that exposure looks like — and that the provenance of the data, rather than the act of training on it, is where the liability has landed.