AI Copyright Splits: Music Majors Sue Anthropic, DOJ Backs OpenAI

AI copyright law is now splitting into two fronts at once: whether training a model on protected work counts as fair use, and how the training data was obtained in the first place.

How the AI Copyright Fight Splits in Two

In one six-week window across late August and early September 2026, the American AI copyright reckoning pulled hard in two opposite directions. On one side, three major music publishers escalated against Anthropic over song lyrics. On the other, the US Department of Justice told a federal judge that training AI on copyrighted text is generally fair use, siding with OpenAI. Same window. Opposite signals.

The pattern underneath is what matters. Courts and the executive branch are drifting toward treating the act of training as lawful. The fight is migrating to a narrower question: was the underlying data licensed, or was it scraped and torrented? Fair-use-for-training is trending AI-friendly. How the data was acquired is where the money now lives.

Front One: The Music Majors Close Ranks on Anthropic

On 28 August 2026, Sony Music Publishing and Warner Chappell Music sued Anthropic in the US District Court for the Northern District of California, according to Music Business Worldwide. The complaint names Anthropic CEO Dario Amodei and co-founder Benjamin Mann as individual defendants, over tens of thousands of song lyrics allegedly copied to train Claude.

With Sony Music Publishing and Warner Chappell joining, all three major music publishers are now litigating against Anthropic over lyrics. The filing follows Round Hill Music's suit on 17 August 2026, so the window alone produced two fresh music-publisher actions against a single AI lab.

The named works read like a jukebox: "All I Want for Christmas Is You," "Eye of the Tiger," "Livin' On a Prayer," "September," "Hallelujah," and "Uptown Funk." The plaintiffs seek up to $150,000 per work for wilful infringement, plus up to $25,000 per removal of copyright-management information. Statutory damages at that scale, multiplied across tens of thousands of compositions, put the exposure in the multi-billion-dollar range.

One of the largest and most blatant ongoing thefts of intellectual property in history.

— — Sony / Warner Chappell complaint, via Music Business Worldwide

Anthropic pushed back. The company's statement, reported by TechCrunch and Axios, was blunt: "We disagree with the publishers' claims and we intend to defend ourselves robustly in court." The complaint frames the conduct as a campaign of illegally torrenting, scraping, and downloading copyrighted works.

This is not the first music-publisher action against Anthropic. Concord, Universal, and ABKCO sued over roughly 500 songs in October 2023, then amended in January 2026 to more than 20,000 songs while seeking $3 billion, and BMG filed over 493 compositions in March 2026. An amended lyrics complaint landed around the same moment Anthropic's separate authors settlement won approval, as Music Business Worldwide reported.

Front Two: The DOJ Backs OpenAI on Fair Use

Days later, the balance tilted the other way. On 1 September 2026, the US Department of Justice filed a statement of interest urging Judge Sidney Stein in the Southern District of New York to find that training AI on copyrighted text is generally fair use. The DOJ called the practice "extraordinarily transformative" and tied it to economic competitiveness and national security, according to the Washington Post.

On 4 September 2026, OpenAI and Microsoft moved for summary judgment in the New York Times copyright case, with the Times and a group of authors filing dueling cross-briefs, as Axios reported. A statement of interest is non-binding. The four-factor fair-use test still governs whatever Judge Stein decides. Even so, the executive branch putting its weight behind OpenAI's reading of fair use is a signal rights-holders cannot ignore.

Read the two fronts together and the shape is clear. The DOJ backs OpenAI on the lawfulness of training. The music publishers hammer Anthropic not on training as such, but on the alleged theft that fed the model. Evidence from the window suggests the courts are quietly separating the two ideas.

The Line That Decides Everything: Licensed vs Scraped

Here is the tell. In July 2026, Anthropic secured final approval of a $1.5 billion class settlement with authors, the largest copyright recovery in US history, covering roughly 500,000 pirated works at a benchmark near $3,000 per work, per the Authors Guild and JURIST. That release covered past acquisition and copying only, not the outputs of the model and not future conduct.

So Anthropic already conceded liability for how the data was obtained, then wrote a ten-figure cheque. Weeks later the music majors arrived arguing the same theory for lyrics. Analysis of the sequence shows why: once a lab pays for piracy-acquisition, every rights-holder with a scraped catalogue can see a number to aim at. The licensed-versus-scraped line is not a footnote. It is the whole game.

The market is already sorting along that line. Warner Music Group settled with Suno and Udio in November 2025 and pivoted toward licensed AI music platforms, while Universal and Sony kept litigating Suno, according to Music Business Worldwide. In the United Kingdom, the Getty Images v Stability AI case showed how narrow these rulings can land, with the High Court largely rejecting Getty's copyright claims in late 2025 before granting permission to appeal a secondary-infringement point.

Training may be ruled fair. How you got the training data is where the bill comes due.

Why This Matters for African Catalogues

This is where I read the moment through Emergent Intelligence (EI) — the dignity-first frame I use for what the world calls AI. If the settled rule becomes "training is fair use, but acquisition must be lawful," then the single most valuable thing a creator owns is a licence trail. For a Global-South creator economy, the licensed-versus-scraped distinction decides whether African catalogues get paid or simply get ingested.

A Zambian songwriter, a Johannesburg studio, a Nairobi label: none of them can outspend Sony or Warner Chappell in a Northern California courtroom. What they can do is register, license, and document, so their work sits on the paid side of the line the American courts are now drawing. Dignity, in AI copyright, is turning into a paperwork discipline. The catalogue that can prove its provenance is the catalogue that survives the scrape.

Frequently Asked Questions

These are the questions people are asking about the AI copyright fights of 2026. Short answers follow, drawn from Music Business Worldwide, the Washington Post, Axios, and the Authors Guild.

What is the AI copyright split of 2026?

In short, the AI copyright split is the divergence, inside one six-week window, between courts and the DOJ leaning toward training as fair use and music publishers escalating against Anthropic over how its training data was obtained. Reporting from Music Business Worldwide and the Washington Post shows the two fronts moving in opposite directions.

How does the DOJ statement of interest affect OpenAI?

Simply put, the DOJ statement of interest is a non-binding brief urging Judge Sidney Stein to treat AI training as fair use, according to Washington Post analysis. It cannot decide the New York Times case, because the four-factor fair-use test still governs, but the DOJ backing OpenAI signals where federal policy stands.

Why is the Anthropic music-publisher lawsuit significant?

The key is the plaintiffs. Data from Music Business Worldwide reveals that Sony Music Publishing and Warner Chappell joining makes all three major music publishers litigants against Anthropic, with statutory damages of up to $150,000 per work across tens of thousands of compositions.

Who is exposed by the licensed-versus-scraped line?

In other words, any creator whose work was scraped rather than licensed. Evidence from Anthropic's $1.5 billion authors settlement shows liability attaching to unlawful acquisition, which is exactly the theory the music majors now press for lyrics.

What are the risks for Global-South creators?

The answer is exclusion by default. Analysis of the current rulings suggests African and other Global-South catalogues without a documented licence trail risk being ingested without payment, since the emerging rule rewards provable provenance over raw ownership.

Sources:

Music Business Worldwide — Sony / Warner Chappell sue Anthropic · TechCrunch — Sony, Warner sue Anthropic · Axios — Anthropic response · MBW — amended lyrics complaint · Washington Post — DOJ backs OpenAI · Axios — summary-judgment motions · Authors Guild — $1.5B settlement approval · JURIST — record settlement · Blackstone Chambers — Getty v Stability AI · MBW — Warner settles with Suno · Related on this site: The earlier NYT v OpenAI phase · The Anthropic Stand · AI copyright and the Google Gemini books suit