A landmark payout reaches Dublin
On 20 July 2026, Judge Araceli Martínez-Olguín of the US District Court for the Northern District of California granted final approval to a $1.5 billion (€1.31 billion) settlement in Bartz v. Anthropic — the largest copyright class-action settlement in US history. Anthropic will pay roughly $3,000 (€2,630) per qualifying book, covering an estimated 482,000 works that authors say were downloaded from the pirate libraries LibGen and PiLiMi to build a training dataset for Claude.
The payout traces back to Judge William Alsup's June 2025 ruling in the same case: training an AI model on lawfully acquired text was "exceedingly transformative" fair use, but assembling a permanent library from pirated copies was not, and that question went to trial. Anthropic settled rather than litigate the piracy claim further.
Hundreds of Irish writers are among the claimants. The Irish Writers Union negotiated its own guidance for members and is actively helping authors file, and best-selling names — Sally Rooney, Marian Keyes, Colm Tóibín — are confirmed beneficiaries, alongside less prominent authors and publishers. Conor Kostick, the Union's disputes officer, stands to receive up to €7,500 for three books. Children's author Oisín McGann qualifies for up to €10,000 across four registered titles. O'Brien Press expects over €10,000 for roughly a dozen titles found on the claims list.
The steelman: this is what piracy accountability looks like
The strongest case for the settlement is straightforward. Authors whose books were scraped wholesale from shadow libraries, without licence, payment, or even notice, have a clean claim: their property was taken and monetised by a well-funded commercial actor. A US class action delivered fast, court-supervised, aggregate compensation without forcing thousands of individual writers into years of litigation they couldn't afford. And because the ruling is public, it puts every AI lab — not just Anthropic — on notice that scraping pirate libraries carries a real, quantifiable cost. As Kostick put it: "The precedent is really, really important because it has since emerged that it wasn't just Anthropic that trained its models using pirate databases." That deterrent effect is a genuine public good, and proportionate-regulation advocates should welcome it rather than wave it away.
Where the compensation runs out
But the settlement class has a hard boundary that has nothing to do with whether a book was actually pirated: it only covers works with a US Copyright Office registration filed within statutory windows tied to publication and Anthropic's download dates, per the Authors Guild's own settlement FAQ. Registration is not required to hold copyright anywhere, including in the US — it's a formality tied mainly to eligibility for US statutory damages. Most Irish-published books were never registered with a foreign government office simply to sell in Dublin or London.
The result, laid out in The Irish Times's reporting on the settlement, is stark: McGann's four registered books qualify, but twelve more of his titles were pirated and excluded outright. O'Brien Press says only a dozen of the "couple of hundred" titles Anthropic swept up made the claims list. The books were equally scraped; only the ones with the right American paperwork get paid.
The lesson isn't the registration gap — it's the fair-use core
It would be a mistake for Irish or EU policymakers to read this settlement as evidence that AI training itself needs tighter statutory restriction. The payout exists because Anthropic pirated books, not because it trained models on text it lawfully held — Judge Alsup's fair-use finding on the training question itself stands untouched by the settlement. That distinction matters enormously for Ireland, which has no equivalent settled doctrine. Ireland transposed the EU's Copyright DSM Directive text-and-data-mining exception via S.I. No. 567/2021, but whether that exception actually covers large-scale AI training — and how rightsholders can effectively exercise the machine-readable opt-out — remains legally unresolved across the bloc. Conflating a piracy judgment with a referendum on TDM would push Ireland toward exactly the outcome this settlement shows is broken: a compensation system gated by procedural formality rather than by whether infringement occurred.
The more useful response is narrower. First, Irish and EU enforcement should keep treating unlicensed shadow-library scraping as straightforward infringement under existing law — no new AI-specific statute is needed to punish what Anthropic did. Second, the registration exclusion is a solvable access problem, not an argument for a broader liability regime: collective licensing bodies and reciprocal recognition between US and EU rights registries would let Irish authors claim against future settlements without needing to have filed paperwork in a jurisdiction most never intended to litigate in. Punishing piracy and clarifying lawful TDM use are two different policy problems; Ireland should solve the first with enforcement it already has, and the second with clearer opt-out infrastructure — not by importing America's registration formality as if it were a model to copy.