US judge blocks Anthropic’s $1.5b book piracy settlement

A United States federal judge has paused Anthropic’s proposed US$1.5 billion book piracy settlement, citing concerns about the agreement’s terms and transparency.

Anthropic, an AI company, had agreed to settle a class action lawsuit from US authors claiming its models were trained on hundreds of thousands of copyrighted books.

Judge William Alsup questioned whether class action lawyers were creating a deal without enough input from affected authors.

Alsup said he needs more details on the claims process and the exact number of works covered, asking for clarity to prevent further lawsuits against Anthropic.

The proposed settlement would pay authors and publishers about US$3,000 per covered work, with an estimated 465,000 books affected.

🔗 Source: The Verge

🧠 Food for thought

Implications, context, and why it matters.

Settlement rejection reveals judicial concerns about protecting authors in AI copyright cases

Judge Alsup’s rejection demonstrates heightened scrutiny over class action settlements that could disadvantage content creators in AI disputes.
The judge specifically worried about lawyers creating deals “behind closed doors” that they would force “down the throats of authors,” showing courts are prioritizing author agency over quick resolution 1.
His demand for a “drop-dead list” of pirated titles reflects judicial awareness that unclear settlement terms could expose companies to future litigation 1.
This contrasts with the typical class action approach where broad settlements are favored for efficiency, suggesting AI copyright cases may face stricter judicial review.
The settlement would have covered approximately 500,000 titles with payouts of around $3,000 per work, making the scope and individual compensation amounts significant factors in the court’s evaluation 2.

Anthropic case establishes piracy liability separate from fair use training debates

This settlement addresses alleged criminal-level piracy from sites like Library Genesis rather than the fair use training issues that dominate AI copyright discussions 2.
The distinction matters because courts have previously supported AI companies in fair use cases, such as when Judge Alsup himself ruled that Anthropic training on purchased books counts as fair use 1.
The $1.5 billion settlement specifically resolves “past liability related to piracy” without establishing new legal precedents for legitimate AI training practices 3.
This creates a clearer legal framework where companies face liability for using illegally downloaded content, while purchased or licensed content for training remains more legally defensible.