What happened at the Anthropic lyrics lawsuit hearing
A federal judge in San Jose spent a full day on Oct. 21 hearing arguments in the Anthropic lyrics lawsuit, a case over whether Claude can be trained on copyrighted songs without a license. She has not ruled. The motions were taken under submission, and no decision date has been set.
Judge Eumi Lee of the U.S. District Court for the Northern District of California heard summary-judgment motions from both sides. Because both parties asked for judgment on the existing record, Lee can decide the case without a jury trial unless she finds disputed facts that require one, according to Mubert. FourWeekMBA described Anthropic as awaiting a fair use ruling, which is where the case stands now.
The case is Concord Music Group v. Anthropic PBC, No. 5:24-cv-03811, per Sigma Law Group. The plaintiffs are Universal Music Group, Concord and ABKCO. They sued in 2023 over lyrics used to train Claude in hundreds of songs by artists including Beyoncé, the Rolling Stones and the Beach Boys, as The News and The420 reported.
Summary-judgment briefing finished in mid-September. Chat GPT Is Eating the World counted the case as one of only three AI copyright cases with completed summary-judgment motions this year, which is why a ruling could arrive without a trial.
Lee has also ruled earlier in the case. She denied the publishers' request for a preliminary injunction, and in that order she wrote that "the threshold question of fair use remains unsettled," according to Yahoo Finance. Anthropic's spokesperson said the company was pleased the court did not grant the publishers' "disruptive and amorphous request." That report was undated in the sources reviewed, so the date of the injunction ruling is unconfirmed.
Why market harm dominated the judge's questions
Lee spent most of her questions on market harm, according to Bloomberg Law as reported by Mubert. Market impact is one of the four factors courts weigh under fair use, and it is where many AI copyright fights get contested. Fair use is the part of copyright law that can allow use of protected work without permission in some cases.
The publishers pressed on volume. Their lawyer told the court that Claude received nearly 120,000 requests to generate "synthetic" lyrics over six months, as Lawyer Monthly reported. Anthropic argued that training analyzes large amounts of material to build capabilities beyond the works used, and that its technology is not a meaningful substitute for the publishers' works, per the same report.
Lee also asked whether the publishers' position would effectively give them a monopoly over lyrics, so that no one could write in an existing style without a license. That question matters beyond this case. A rule that treats style as something to license would reach far past training data, while a rule that ignores the question would leave rights holders with fewer tools when a model imitates their catalog.
Law360, as cited by Mubert, reported that Lee seemed open to Anthropic's fair use defense and said AI "mash-up" lyrics might be fair use. That account is secondhand, and the exact wording of the judge's remarks has not been confirmed against a full transcript. Treat it as a signal about where her thinking may be, not as a holding.
What the money at stake looks like
Under 17 U.S.C. § 504(c), statutory damages for each infringed work run from $750 to $30,000, and can reach $150,000 per work if a court finds the infringement willful. Those figures frame exposure, not a verdict, because no liability finding exists yet. The number of works at issue would turn that range into a headline total, and the sources reviewed do not give one.
The closest benchmark is the book case. In Bartz v. Anthropic, a judge found that training on copyrighted books was fair use but that storing pirated copies was not. The case then settled for $1.5 billion, roughly $3,000 per work, according to Norton Rose Fulbright. That settlement followed a fair-use ruling and does not decide a case like this one, so it offers a price signal rather than a legal template.
How the case fits the courts' split on training data
Courts have not agreed on training data. The Third Circuit affirmed in Thomson Reuters v. Ross on Sept. 29, 2026, holding that Westlaw headnotes are copyrightable and that ROSS's use of them to train a competing tool was not fair use, according to Courthouse News. The three matters point in different directions:
| Case | Status | What the court held on training |
|---|---|---|
| Thomson Reuters v. Ross | Affirmed by the Third Circuit, Sept. 29, 2026 | Using Westlaw headnotes to train a competing tool was not fair use |
| Bartz v. Anthropic | Settled after a fair-use ruling | Training on copyrighted books was fair use; storing pirated copies was not |
| Concord Music v. Anthropic | Argued Oct. 21, 2026; no ruling | Undecided |
The Ninth Circuit added another thread on Sept. 16, 2026, when it affirmed dismissal of a DMCA § 1202(b) claim in Doe v. GitHub. That ruling bears on the publishers' parallel claims, per AI Copyright Lawsuit Updates. On Oct. 12, Lee allowed each side to file a three-page brief on what the decision means for Anthropic's motion on the § 1202(b) claim, and both sides filed on Oct. 16.
The publishers' complaint is one of more than 100 active AI copyright cases, according to AI Lawsuit Tracker. Authors are pressing a parallel fight, and USA 24 has covered nearly 300 authors suing Meta over pirated books used to train AI. A district court ruling does not bind other judges, but parties in those cases will read it closely.
What the outcome means for developers, labels and users
A ruling that training is fair use would weaken licensing-based claims against AI developers. A ruling against Anthropic could push companies toward licensing deals for lyrics and other text. For labels and songwriters, the central question is whether they can charge for training licenses at all. Either way, the decision will shape what AI companies pay for the text they train on.
The sources reviewed report no immediate change to how Claude works. Any effect on consumers would arrive later and indirectly, through product terms, pricing or the kinds of content models are trained on.
What businesses should check now
Most readers need to do nothing today. Businesses that publish AI-generated lyrics, music or marketing copy should review their vendor agreements while the case is pending:
- Check whether the vendor may use your prompts or outputs for training.
- Confirm who owns generated output under the contract.
- Ask whether the vendor offers indemnity for copyright claims, and on what terms.
- Ask how the vendor sourced the training data it relies on.
This is general guidance, not legal advice. Talk to counsel before changing contracts or publishing material based on these questions.
What to watch as Lee weighs the motions
Three questions matter most. The first is timing, since Lee has set no decision date and the motions were taken under submission. The second is scope: whether she rules on fair use alone or also resolves the § 1202(b) claim. The third is trial. Lee reportedly pressed both sides on whether this case could become the first AI copyright case to reach trial.
A fair use win for Anthropic would strengthen the hand of model developers across the industry. A loss would push lyrics and other text toward licensing. The publishers said they "remain very confident in our case against Anthropic more broadly," according to Yahoo Finance, so expect the fight to continue either way. The next filing and the next hearing date will show which direction the case is moving.
