The ChatGPT Moment for Music

Venture Capitalists are out way over their skis with music copyright. Almost every Tier-1 investor I've polled thinks the lawsuits against Suno and Udio are an easy settlement, compares it to copying a blog post, and doesn't understand how music labels play the long game. One investor I talked to actually invested in a founder who claimed he could design an algorithm to clean-room the training. That's laughable to anyone who knows the field. The experts know that licensing is all you need.

I know a product manager who went back into music after AI because she could see that the music industry is a decade ahead of everyone else in the sophistication of its intellectual property regime. Music grew less quickly than movies and games in the last generation, but it's already AI-ready — planned, litigated, and responsive to technological change in ways that other content industries are only beginning to appreciate. The only advisors most VCs have access to are lawyers who bill by the hour (incentives obviously misaligned) and the infringing founders themselves (also obviously misaligned). They need someone who has navigated these rights successfully and isn't burned out by the industry. They don't have that person. It shows.

The music industry itself sees the long-term future clearly and is largely unified in what it's asking for, as this rant by Benji on the Unlimited Supply podcast makes plain. Universal used its deal with Udio to demonstrate that a separate, controlled market with fair licensing is possible, which undercuts the only argument the anti-copyright crowd has left. That's checkmate for anyone still pushing for the destruction of music copyright. The history here matters, and VCs don't know it. So let's start with the graveyard.

• Napster (2001): 9th Cir. affirmed injunction; contributory/vicarious infringement. Napster never recovered; assets sold in bankruptcy.
• Aimster (2003): 7th Cir. found contributory infringement; encryption didn't save them.
• Grokster / Morpheus (2005): U.S. Supreme Court: inducement liability; service shuttered, paid to settle.
• Kazaa (2006): Paid $115M global settlement with labels; promised content filters.
• MP3.com (2000): Summary judgment for labels over "Beam-It"; paid large settlement, business gutted.
• LimeWire (2010): Liability on summary judgment; $105M settlement; shut down.
• Turntable (2011): My own Turntable got restricted almost immediately in 2011 due to lack of licensing.
• Grooveshark (2015): Consent judgment; site closed, domains/IP handed to labels.
• ReDigi (2018): "Used MP3 resale" loses on reproduction right; first-sale didn't apply.

The Tier-1 VCs I've polled don't recall this graveyard. They see YouTube — a special case — and TikTok — an even more special case that only survived this long because the entire weight of the Chinese state was behind it, which is itself causing it to get divested at 90% off. They don't think about the personal exposure either. The LimeWire founder was found personally liable. AI founders pushing infringement face the same risk.

Music is different from text, photos, and video. It has a much denser set of rights built up over more than a century, and any gray zone that's ever appeared has been filled in by Congress — as recently as 2018, during a gridlocked Congress. The music labels' main business is ensuring their rights continue, and they are very, very effective at lobbying. Every new layer reduces the gray-zone arguments the infringers depend on.

An engineer was arguing with me once that IP is dead, and after we went through the specifics of music, he ended up writing a ten-thousand-word essay proposing yet another new kind of intellectual property right to layer on top of the existing ones. His other move was to say China doesn't respect copyright and that AI music is already a hit there. But China has zero global cultural relevance in music, and the US continues to increase its dominance. That soft power and the artists who create it go hand in hand.

Why is Music different from Photos, Videos, and Text?

• Concentrated rights & turnkey licensing. Three majors control most commercially valuable masters and publishing admin — and there are efficient collectives (ASCAP/BMI/MLC/SoundExchange). Courts care about licensing practicability; music has it.
• Long, consistent enforcement history. From Napster to Grooveshark, courts and settlements have rewarded the rights-holders' theory of harm. Judges have muscle memory here.
• New statutory hooks. The 2018 MMA standardized digital mechanicals and federalized pre-'72 recordings; DMCA §1201 gives labels a powerful lever even before you reach fair use. State NIL laws now speak directly to voice clones.
• Market evidence for training licenses. Once the majors ink AI training and AI-creation deals (UMG–Udio), the "we couldn't license it" argument evaporates.

The rights lattice for music gets denser over time, never simpler: compositions → performances → mechanicals → recordings → digital performances → anti-circumvention → neighboring rights → modern blanket licensing. The attribution and remuneration the industry is asking for from AI isn't much different from what it asked for during the sampling explosion, which has long since become settled business.

US History

• 1831 – Musical compositions explicitly added to federal copyright.
• 1897 – Congress creates a public performance right for musical works, which is the seed of ASCAP/BMI.
• 1909 – The mechanical right and compulsory license schematic (player-piano era) become law, the foundation for "mechanicals."
• 1971/1972 – Congress creates federal copyright for sound recordings (the master) effective Feb 15, 1972. Until then, no federal protection for recordings.
• 1995 – Digital Performance Right in Sound Recordings Act adds a digital public-performance right for sound recordings (webcast/stream); SoundExchange later administers the statutory license. Terrestrial radio stays exempt.
• 1998 – DMCA §1201: anti-circumvention. You can't bypass access controls (e.g., YouTube's rolling cipher) even if you swear the downstream use is "fair." It's a separate cause of action which is itself a super fast slam-dunk case.
• 2018 – Music Modernization Act: builds the MLC for digital mechanicals; pulls pre-1972 masters into federal protection (CLASSICS Act) and harmonizes remedies.

International layer

• Rome Convention (1961) and WIPO WPPT (1996): "neighboring rights" for performers and phonogram producers — remuneration for broadcast/communication to the public. (U.S. is idiosyncratic — no terrestrial radio performance right — but the international default is performer/producer remuneration.)
• EU DSM Directive (2019) Art. 4: TDM opt-out: rightholders can reserve rights to block text/data mining via machine-readable means (metadata/T&Cs). That matters for AI training.
• EU AI Act (2024/2025 rollout): transparency and training-data summary duties for foundation/GPAI models; "copyright compliance" is explicitly on the checklist. The EU is building a paper trail requirement the "scrape now, apologize later" crowd can't meet.

The stacks of rights, just scratching the surface

If one doesn't understand these and why it's different from text, why assert so strongly that you understand where this is going?

A. Musical work (song / publishing)

• Public performance (ASCAP/BMI/SESAC/GMR blanket licenses)
• Mechanical (reproduction/distribution) for phonorecords & streams (now centrally administered for digital by the MLC)
• Synchronization (sync) for A/V uses (always negotiated; no compulsory)
• Print (sheet, lyrics)

B. Sound recording (master / label)

• Reproduction/distribution of the master
• Digital public performance (webcasting/interactive distinctions; SoundExchange for statutory non-interactive)
• Master sync (paired with the composer's sync)
• Neighboring rights outside the U.S. (and U.S. digital via §114)

C. Identity rights

• Right of publicity / name-image-likeness (NIL) state law:
  • California §3344/3344.1 (living + 70-year post-mortem) and New York §50-f (including "digital replicas")
  • Case law: Midler v. Ford and Waits v. Frito-Lay: you can't imitate a well-known voice to sell things — identity can be in the sound itself. White v. Samsung and No Doubt v. Activision expand the contour of "identity" beyond name/photo to robot stand-ins and avatars. That's squarely relevant to voice clones and sound-alikes.
  • 2024–25 wave: Tennessee's ELVIS Act adds explicit voice-clone protections with civil/criminal teeth; California updated post-mortem law for "digital replicas." The trend line is hostile to unlicensed voice emulation.

Things are changing week to week. This is the ChatGPT moment for music. With the combination of all of these rights stacked on top of each other, licensing is the only path to a sustainable business — unless you want to be the next bankruptcy on the list, regardless of which administration is in power.