Anthropic Awaits Fair Use Ruling as Perplexity Heads to Jury

US District Judge Eumi Lee heard arguments in San Jose on 21 October 2026 on whether Anthropic’s use of song lyrics to train Claude is protected as fair use, and did not issue a ruling, Reuters reported. The music publishers’ lawyer told the court that Claude received nearly 120,000 requests to generate “synthetic” lyrics over six months.

In New York, the court’s current scheduling order, signed on 15 September 2026, in Dow Jones and the New York Post’s case against Perplexity says: “This case is to be tried to a jury.” We read both dockets and set out where five AI copyright cases stand, what each claims, and the next date on record.

Business Pill · ROYALTIES

A short Business Engineer explainer on royalties: paying each time you use someone else’s work. It teaches the general idea only and says nothing about any company, court or case in this piece.

What Happened in San Jose

The case was filed in 2023 by Universal Music Group, Concord and ABKCO, and alleges that Anthropic infringed copyrights in hundreds of songs by using their lyrics to train Claude without permission, according to the Reuters report as relayed by ET EnterpriseAI. The works include songs by Beyoncé, the Rolling Stones and the Beach Boys.

Fair use, in plain terms, is the US copyright defence that allows some use of protected work without permission. The two questions argued in San Jose, as Reuters describes them, were whether Claude’s use of the lyrics is transformative and whether it harms the market for them.

Anthropic’s lawyer, Sonal Mehta of WilmerHale, argued that training on copyrighted works lets models produce transformative content. Matt Oppenheim of Oppenheim + Zebrak, for the publishers, argued that Anthropic was using their lyrics to generate competing material, and cited the 120,000 requests.

According to Reuters, Lee questioned the publishers’ claim that Claude’s ability to generate song lyrics threatens the market for their copyrighted material. Mehta said there was no evidence that Claude had reduced demand for lyrics, and cited Universal Music Group’s own statement that AI was “overwhelmingly a net positive” for the music industry.

The court docket shows a motion hearing entry on 21 October. Days earlier, on 12 October, Judge Lee let each side file a three-page brief on what a 2026 Ninth Circuit decision, Doe 1 v. GitHub, means for Anthropic’s motion for summary judgment on the publishers’ claims under 17 U.S.C. § 1202(b). Both sides filed on 16 October.

Perplexity’s Path to a Jury

Dow Jones and the New York Post sued Perplexity in Manhattan federal court on 21 October 2024, Startup Fortune reports, and the complaint was later amended to add claims of false designation of origin and trademark dilution. The publishers allege that Perplexity’s answers sometimes invent details and attribute them to the outlets by name. Startup Fortune reports that Perplexity disputed the publishers’ account of their pre-suit letter and called the lawsuit’s framing “misleading”.

The third amended scheduling order, signed by Judge Katherine Polk Failla on 15 September 2026, says all fact discovery “shall be completed no later than December 10, 2026” and expert discovery no later than March 16, 2027. It gives the parties’ “present best estimate of the length of trial” as “10 to 14 days”. No trial date appears on the docket, and a conference is set for 14 December 2026.

Some press accounts give earlier discovery dates; the dates here are the ones in the court’s own order. By our count, the conference is 53 days from the date of this piece.

The AI Copyright Docket

The five cases below are the ones our coverage and the two dockets let us place. The selection is ours and the list is not exhaustive; every row comes from the source named in it.

What a Copied Work Can Cost

Under 17 U.S.C. 504(c), US statutory damages run from $750 to $30,000 per work, and up to $150,000 per work if a court finds willful infringement; these are limits set by law, not awards. The one settled number on record is the Bartz v. Anthropic settlement: $1.5 billion, or approximately $3,000 per work across an estimated 500,000 works, as our earlier pieces reported.

As we read the table, the cases ask two different questions. The music and news cases ask whether training and answers are fair use. The shadow-library and Sony-Warner cases ask how the training data was obtained, and who decided to obtain it.

The Structural Read

As we read it, the music case turns on market harm: the publishers point to nearly 120,000 requests for synthetic lyrics in six months, and Anthropic points to Universal Music Group’s own statement that AI was “overwhelmingly a net positive” for the music industry.

As we read it, the Perplexity case adds a trademark theory on top of copying: false designation of origin and trademark dilution, over answers the publishers say attribute invented details to them, as Startup Fortune reports.

As we read it, the price range on the table is wide: US law sets $750 to $150,000 per work, and the one settled reference, the Bartz v. Anthropic settlement, works out at approximately $3,000 per work.

Three Implications

FOR AI LABS Reuters reports that Judge Lee questioned whether Claude’s ability to generate lyrics threatens the publishers’ market, the market-harm question at the centre of fair use.

FOR PUBLISHERS The publishers’ lawyer put a usage count on competing use: nearly 120,000 requests for synthetic lyrics over six months, Reuters reports.

FOR AI SEARCH The court’s order puts Perplexity on a jury track, with fact discovery to 10 December 2026 and expert discovery to 16 March 2027, and no trial date yet.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Four Intelligence Moats.

The framework lists four moats, one of them this: “The Corpus Moat — built by pretraining, eroding as the public web is exhausted and synthetic data closes gaps.”

As we read it, each case in the table is a dispute over the cost of building that corpus: whether training on it is fair use, or how it was obtained and who approved it. The framework was published in July 2026 and does not discuss these cases.

What Is Not Established

No court in these cases has ruled on fair use, and every infringement claim in the table is an allegation, not a finding. Judge Lee’s questions at the hearing are not a ruling. The Perplexity schedule is the court’s current order, not a trial date.

We read Reuters’ report through ET EnterpriseAI’s account of it, not the original article, and the Perplexity claims summary comes from Startup Fortune. The 53-day count is our arithmetic. The statutory figures are ranges in the law, not estimates of what any party owes.

Connecting the Coverage

On 20 October we reported that a magistrate judge ordered Meta to name the staff who approved shadow-library downloads in the Kadrey cases. On 19 October we covered Canadian publishers suing Anthropic for up to $20,000 per work, and Seoul Economic Daily’s report that Google is buying Korean books for AI training, where the roughly $3,000 Bartz figure is being read as a price tag.

On 30 August we reported that Sony Music Publishing and Warner Chappell sued Anthropic over how its training data was acquired. What is new is the 21 October hearing on the fair use question itself; the Perplexity order, signed on 15 September 2026, is the current schedule for a news publishers’ case against an AI search engine.

The Bottom Line

Judge Lee heard Anthropic’s fair use argument on 21 October 2026 and has not ruled, while the Perplexity case has a court order that says it is to be tried to a jury, with discovery running to March 2027. As we read the five cases, the AI copyright fight has moved from complaints to the court calendar.