RIAA, NMPA and 6 music bodies file an amicus brief in the Anthropic copyright case, arguing AI lyric generation infringes copyright

Key highlights

• Eight organizations including RIAA, NMPA, and SoundExchange filed in support of publishers’ partial summary judgment motion against Anthropic on April 2, 2026
• The central legal argument: AI lyric generation substitutes for human-authored lyrics in the market, failing the fourth prong of the fair use test
• A ruling in publishers’ favor would require AI companies to license training data proactively, not just pay damages after the fact

Eight music bodies unite behind a single legal theory

On April 2, 2026, eight US music industry organizations filed an amicus brief supporting music publishers in their copyright infringement lawsuit against Anthropic. The filing targets a motion for partial summary judgment the publishers had already put before the court in March 2026.

The eight filers: RIAA, NMPA, A2IM, SONA, Black Music Action Coalition, Music Artists Coalition, Artist Rights Alliance, and SoundExchange. That coalition spans rights holders, indie label representatives, and artist advocacy groups — a broader alignment than you’d typically see in a single filing.

The underlying case: UMG, Concord, and ABKCO sued Anthropic, alleging the company used copyrighted song lyrics to train the Claude AI model without licenses or compensation. Coverage of the original music publishers lawsuit gives the full case background.

The market substitution argument, explained

The brief doesn’t argue Anthropic copied specific songs in a way the output reproduces. It argues something broader: that AI lyric generation harms the market for human-authored lyrics at scale, which fails the fourth prong of the fair use doctrine.

The key quote from the brief: “Claude and other AI systems can generate lyrics for thousands of songs in the span of time that it takes a human author to write the lyrics to just one song.” The implication is that the substitutional harm to the licensing market for copyrighted lyrics is “very real, and it is severe.”

This sidesteps the transformative use debate entirely. Even if Claude’s output doesn’t reproduce training data verbatim, if it displaces demand for licensed human-written lyrics, fair use fails. Music Business Worldwide reports the brief also argues a functioning AI licensing market already exists — meaning Anthropic had a path to license and chose not to.

Why this legal theory has historical weight

The RIAA used market harm arguments successfully in Napster (2001) and Grokster (2005). The same fourth prong logic that shut down file sharing networks is now being applied to AI training. The parallel isn’t perfect — Anthropic is a licensed commercial product, not a peer-to-peer piracy platform — but the legal framework is identical.

The takeaway: Market substitution doesn’t require proving that AI output copies a specific song. It requires proving that AI-generated lyrics compete with and displace licensed human-authored ones. That’s a harder empirical question, but also a harder argument to dismiss on motion to dismiss or summary judgment grounds.

The contrarian read: Anthropic trained on song lyrics as text, not audio recordings. Whether AI lyric generation actually harms the licensing market for specific copyrighted lyrics is empirically unproven. The industry doesn’t have clean substitution data. Courts will have to decide what counts as market harm in a category that barely existed three years ago.

This connects to broader legislative battles. The Trump AI framework explicitly defends AI fair use with no licensing mandate or opt-out requirement. The UK copyright opt-out collapsed after 11,500 responses pushed back. The GEMA ruling against Suno gives the music industry a win in Germany. The Anthropic case is the US test of the same argument at the training data level. AI and copyright guidance from the US Copyright Office is still evolving.

What this means for builders and creators

If the publishers win partial summary judgment on market substitution grounds, the implications go well past Anthropic. Every major AI company trains on text that includes song lyrics. A ruling that training without a license constitutes market harm would require retroactive licensing conversations across the industry.

For working musicians and producers, the Patreon CEO framing on AI training as a creator rights issue captures why this coalition came together: consent, credit, and compensation. The amicus brief puts those principles into a legal theory that doesn’t require proving track-by-track copying. If you are building any AI tool that generates or processes lyrics, the fourth prong of UK AI training copyright debate and the Anthropic case are the two watchpoints right now. A ruling is expected in 3 to 6 months.