US judge pauses approval of Anthropic’s $1.5bn authors’ settlement

TL;DR:

• US District Judge Araceli Martinez-Olguin has held back final approval of Anthropic’s proposed $1.5bn settlement with authors over the alleged use of pirated books to train Claude, asking for more detail on lawyers’ fees and payments to lead plaintiffs.
• The deal is the largest known US copyright settlement and the first major US AI-training case to settle, covering more than 480,000 works with claims already filed by holders of over 92% of those works.
• A group of more than 25 writers who opted out — including Dave Eggers and Vendela Vida — filed a separate California complaint against Anthropic on Wednesday, signalling that the litigation tail extends well beyond the headline settlement.

Now-retired Judge William Alsup initially greenlit the agreement last September. The case stems from a 2024 lawsuit accusing Anthropic — which is backed by Amazon and Alphabet — of using pirated versions of authors’ books without permission to teach Claude. Alsup ruled last June that Anthropic made fair use of the works for training but found it violated authors’ rights by saving more than 7 million pirated books to a central library not necessarily used for training.

The procedural state

Martinez-Olguin did not grant final approval at the San Francisco hearing, instead asking for written supplementation on fees and lead-plaintiff payments. A trial had been scheduled for December to determine damages for the alleged piracy, with potential exposure running into the hundreds of billions of dollars — which is why the settlement, even at $1.5bn, looked like the cheaper outcome for Anthropic and its investors.

UK angle: precedent ripple into UK creative-industries debate

UK rights-holders and the publishing industry are watching closely. The Mumsnet litigation, the recent UK government consultation responses on text-and-data mining, and ongoing tensions between rights-holders and frontier labs over training-data licensing all turn on the same legal axis: whether AI training constitutes fair use, and what compensation framework applies to past piracy. The Anthropic deal sets the floor — both as a payable quantum (claims on 480,000 works) and as a procedural template for future class-style copyright settlements involving AI training corpora.

Looking forward

Two near-term questions for UK readers. First, whether parallel UK-jurisdiction claims against frontier labs accelerate now that a US settlement quantum exists as benchmark. Second, whether the DSIT consultation outcome on AI and copyright reflects this new floor — UK rights-holders will read a $1.5bn settlement as proof that voluntary licensing frameworks are workable, while frontier labs will read it as proof that settled past liability is preferable to litigation.