What's at Stake—and What Isn't—as Bartz v. Anthropic Settlement Heads Toward Approval

Yesterday, Judge Araceli Martínez-Olguín of the Northern District of California held an important hearing in Bartz v. Anthropic to determine the fairness of the proposed $1.5 billion settlement. The hearing is one of the last major procedural steps before the court decides whether to grant final approval. Judge Martínez-Olguín took the matter under submission rather than ruling from the bench, and indicated that short orders may follow as soon as today.

The scope of this settlement has been widely misunderstood in press coverage, and it is worth restating before turning to what happened at the hearing: Bartz is not really a settlement about AI training. In June 2025, Judge William Alsup ruled that Anthropic's use of books to train its large language models was fair use, and he also concluded that Anthropic was entitled to scan books it had lawfully acquired. Nothing about this settlement disturbs those rulings.

What Does the Settlement Achieve?

What the court left unresolved was Anthropic's potential liability for downloading and storage of files from two shadow libraries—Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi).

In his June 2025 order, Judge Alsup found that the piracy issue was not fair use and that the case could proceed to trial on that issue, prompting the settlement. A short time later, he certified a class that included any copyright holders whose reproduction rights were implicated in the downloaded datasets and met certain copyright registration and identification criteria.

As compared to what the plaintiffs had originally proposed, Judge Alsup both narrowed the class (only copyright-registered works with ISBNS or similar were included) but also broadened it, by including not just authors but anyone with a legal or beneficial copyright interest in those works. Most importantly, this brought publishers into the suit, and some large publishers now stand to reap a substantial windfall. In some cases, the recoveries will be shared, sometimes contentiously, between authors and publishers.

For Anthropic, the settlement releases them from liability for past conduct in exchange for a $1.5 billion payment and destruction of the pirated copies. It does not create any forward-looking licensing scheme, and it does not bind any other AI developer or any future case.

What a Fairness Hearing Is

A fairness hearing is the procedural moment in any class action settlement at which the court decides whether to approve the deal the parties have negotiated. Because a class settlement binds absent class members—that is, people who did not personally agree to the terms and may not even know the lawsuit exists—the court has a duty to assess whether the proposed resolution is fair to them.

That duty is set out in Federal Rule of Civil Procedure 23(e)(2), which provides that a court may approve a class settlement "only after a hearing and only on finding that it is fair, reasonable, and adequate."

The rule directs the court to consider, among other things, whether class representatives and class counsel adequately represented the class, whether the deal was negotiated at arm's length, whether the relief is adequate in light of the costs and risks of continued litigation and the proposed attorneys' fees, and whether the settlement treats class members equitably relative to each other.

In practice, the fairness hearing is also the moment at which objectors—class members who have filed written objections to the settlement—have the opportunity to address the court directly. The court is not bound to follow either Class Counsel's recommendations on how to handle objections or the objectors' arguments; it makes its own determination of whether the settlement satisfies Rule 23.

What Happened at the Hearing?

Two impressions from the hearing are worth highlighting.

The first is how empty the proceeding was. The courtroom itself had only a handful of people in attendance, and the Zoom feed—which was the principal way for class members and the public to follow the hearing—was sparsely populated as well.

The objectors who appeared did so almost entirely by Zoom. Only one attorney appeared in the courtroom representing a small group of four objectors. Given the size of the class (rightsholders for nearly half a million books!) and the dollar value of the settlement, the modest turnout was striking.

During the hearing, Class Counsel reported a claims rate of 92.77% and 350 valid opt-outs covering 1,802 works, which suggests that the great majority of class members are content to participate in the settlement as proposed.

The composition of those claims is also notable: as of the March 19 motion for final approval, the Settlement Administrator had received more than 99,000 claims representing 273,331 works — 90,476 from authors, 2,379 from publishers, and the remainder from literary estates, trusts, and loan-out companies. The author-to-publisher imbalance in the claimant count is a useful reminder that a relatively small number of publisher claimants may end up controlling the rights to a substantial share of the books in the class.

The handful of objectors who appeared raised (in the mere two minutes each was given to speak) a familiar set of concerns—most prominently, that the Works List excludes works that should have been included (for example, because of group registration practices, pseudonyms, or other registration formalities), and that the per-work payment is small relative to the statutory damages ceiling.

The second impression is that the bulk of the Judge's attention was directed not at the merits of the objections but at the structure of attorneys' fees and expense accounting. Most of the Judge's questions focused on protecting class members from their own lawyers: scrutinizing the proposed fee multiplier, asking how the $15 million cost reserve would be administered and what would happen to unspent funds, and questioning an accounting line that appeared to move money between Class Counsel and non-appointed counsel.

At one point during the hearing, the Judge asked counsel to identify three leading cases that support the size of the fee request. Class Counsel's responses leaned on the result for the class (a $1.5 billion settlement, the largest copyright settlement award ever, and the high claims rate) and on the difficulty of litigating an untested area of law.

Whether the Court will accept those justifications in full, in part, or with modifications, is one of the things final approval will tell us.

What's Next in Bartz and in other Book Class Action Cases?

If Judge Martínez-Olguín grants approval, this will be one of the last major procedural hurdles for the settlement to resolve Bartz and bind all class members who did not opt out by the deadline. Anthropic will be required to destroy the pirated copies and to pay the $1.5 billion into the settlement fund, from which administrative costs, attorneys' fees, and per-work payments will be distributed.

But for many authors implicated in Bartz, the litigation around AI training and book piracy is not ending; it is shifting to other cases.

There are dozens of other AI copyright suits at varying stages—including consolidated actions against OpenAI and Microsoft, and active putative class actions against Meta, Apple, Salesforce, Adobe, and others.

Several of those cases appear to be adopting the strategy that proved effective in Bartz: rather than challenging AI training as such, they focus on the underlying piracy and on infringing copies. And separately, some authors are now being represented—without having asked to be—in suits brought by major publishers asserting claims in the authors' works, most recently in the Elsevier-led suit against Meta filed earlier this month.

For many authors whose books appear in the Bartz Works List, a claim filed in this settlement will not be the last decision to make about how, and by whom, those works will be used by AI.