Tech Industry Backs Anthropic in Music Publishers Lyrics Case, Citing Fair Use

When music publishers Universal Music Group, Concord, and ABKCO sued Anthropic in 2023 for allegedly training Claude on copyrighted song lyrics, the case quickly became one of the most-watched battlegrounds in the broader AI copyright war. This week, tech industry organizations escalated the fight by filing amicus briefs urging the court to side firmly with Anthropic, arguing that using lyrics to train large language models is a textbook example of "fair use" under U.S. copyright law.

The filings arrived as both sides press for summary judgment, a ruling that could short-circuit a trial and set a precedent reverberating across every active AI copyright case. With music publishers framing AI training as the "largest copyright infringement in history" and tech groups warning that an adverse ruling would "kneecap" the entire generative AI industry, the stakes have rarely been clearer.

What the tech industry brief argues

The amicus brief, filed in support of Anthropic's pending summary judgment motion in the U.S. District Court for the Northern District of California, makes three core arguments familiar to anyone tracking the AI copyright docket.

First, the brief frames LLM training as a "transformative use" — the foundational test under the Supreme Court's Warhol v. Goldsmith and Google v. Oracle decisions. Anthropic and its supporters argue that ingesting copyrighted lyrics to teach a model statistical relationships between words is fundamentally different from reproducing or distributing those lyrics. The model is not a jukebox; it is a probability engine that learns language patterns.

Second, the brief leans heavily on Judge William Alsup's June 2025 ruling in Bartz v. Anthropic, where the same court held that Anthropic's training on lawfully acquired books was "exceedingly transformative" and qualified as fair use. Tech industry groups argue the same logic should extend to lyrics, particularly where Anthropic took technical measures to prevent Claude from reproducing copyrighted text on demand.

Third, the brief warns of catastrophic policy consequences. If using copyrighted text for AI training is not fair use, the argument goes, virtually every commercial LLM in existence would face existential liability, foreign competitors operating outside U.S. copyright jurisdiction would gain a decisive edge, and the U.S. would effectively cede AI leadership at the moment it matters most.

How publishers see it

Music publishers and their amici see things very differently. Universal Music Group, Concord, ABKCO, and a coalition of supporting organizations including the RIAA, NMPA, and major book and newspaper publishers have argued that Anthropic crossed a line that book training did not.

Their core counterargument is that song lyrics are a uniquely creative, compact form of expression. Unlike a 300-page novel, a four-minute song is concentrated copyrighted material, and Claude has been documented producing near-verbatim reproductions of lyrics from artists including Beyoncé, the Rolling Stones, and Don McLean. Publishers argue that this output, combined with Anthropic's alleged failure to license the lyrics, places the case squarely outside any reasonable fair use defense.

They also push hard on the fourth fair use factor — market harm. A licensing market for AI training data already exists, the publishers argue, and Anthropic deliberately bypassed it. If courts bless that conduct as fair use, the entire economic foundation of music publishing collapses.

Why this case matters more than the settlement

The Anthropic case has multiple fronts moving at once, and it is easy to lose track. The $1.5 billion proposed settlement with book authors in Bartz v. Anthropic — currently delayed by Judge Araceli Martinez-Olguín pending more detail on payouts and attorney fees — addresses past use of pirated books. It does not resolve the lyrics question.

The Concord case is different in three crucial ways:

• It tests fair use on its merits. Unlike the Bartz settlement, which sidesteps the legal question by paying out, the lyrics case is heading toward a substantive ruling on whether AI training on creative works is fair use.
• It involves a different category of work. Songs and lyrics carry distinct copyright treatment under U.S. law, and the music industry has historically been among the most aggressive enforcers of its rights.
• It will produce binding case law. A summary judgment ruling — for either side — will be cited in every subsequent AI copyright dispute, including the Meta cases, the OpenAI cases, and the Cruz v. Anthropic class action filed last month.

The Bartz ripple effect

Tech industry groups are explicitly leveraging Judge Alsup's Bartz ruling, and for good reason. That decision was the first major federal court ruling to find AI training on copyrighted text to be fair use, and it has become the cornerstone of every defense filing across the industry.

But Bartz came with a significant caveat: Alsup found that Anthropic's separate act of downloading pirated copies of books from shadow libraries was not fair use, exposing the company to potentially massive damages — the issue that drove the $1.5 billion settlement. The ruling effectively split AI training into two questions: how the training itself functions legally, and whether the training data was acquired lawfully.

That split is now front and center in the lyrics case. Anthropic argues that even if some training inputs were obtained through questionable means, the training process itself remains transformative. Music publishers argue the opposite: that Anthropic's conduct in obtaining lyrics, combined with Claude's documented reproduction of them, places the case outside Bartz entirely.

What to watch next

Several developments will shape the case in the coming weeks. The court is expected to set a hearing schedule on the dueling summary judgment motions, with arguments likely later this summer. A ruling could come as early as fall 2026.

The amicus brief landscape will continue to grow. With the RIAA, NMPA, and publishers already on the music industry's side, and tech industry groups aligning behind Anthropic, this case has become a proxy for the broader fight over whether U.S. copyright law accommodates the realities of generative AI — or whether AI companies must build a licensing-first ecosystem from scratch.

Watch for filings from the U.S. Copyright Office, which has been active throughout 2025 and 2026 on AI policy questions, and from federal courts in adjacent cases. Any ruling here will be tested almost immediately in the Meta lawsuit filed by Scott Turow and five major publishers earlier this month, and in the parallel proceedings against OpenAI and Microsoft pending in the Southern District of New York.

The bigger picture

The Concord case is more than a music industry dispute. It is the first major opportunity for a federal court to rule directly on whether using copyrighted creative works to train commercial AI is fair use, with a full evidentiary record and aggressive briefing on both sides.

Whatever happens, the ruling will not be the last word. Appeals are inevitable, and the Supreme Court has shown growing interest in copyright questions involving transformative use. But it will be the most authoritative answer yet to a question every creator, AI company, and copyright lawyer has been asking since ChatGPT launched: where does the law actually draw the line?

For now, the briefs are stacking up, the judges are reading, and an industry that has bet hundreds of billions of dollars on the assumption that AI training is fair use is about to find out whether the courts agree.