Can AI Compose a Hit Song? Copyright Will Decide
Last Updated: September 10, 2026 | By Mihail Sebastian | AI Experiments
AI already composes chart-plausible songs. The fake Drake track and the label lawsuits against Suno and Udio show the real fight is legal, not musical.
In April 2023, a song with Drake and The Weeknd on vocals went viral. Neither artist had recorded a note of it.
That track settled one half of this article’s title question. Machines, or people armed with them, already produce music that listeners accept as the real thing. What nobody has settled is the other half: whether that music is legal, and who gets paid when it is.
The hit song question turned out to be a copyright question. Here is how it got there, and where the fight stands.
How Far the Technology Came
In 2016, researchers at Sony’s Computer Science Laboratory released “Daddy’s Car”, a song composed by AI in the style of The Beatles. Humans wrote the lyrics, arranged the parts, and produced the recording. The melody was the machine’s, and the result was a curiosity: impressive, clearly experimental, nobody’s playlist staple.
Two years later, Taryn Southern released “I AM AI”, an album built with AI composition tools. She proved an artist could ship a full commercial record with a machine as co-writer. The workflow was still slow and hands-on, closer to programming than songwriting.
Then generative AI collapsed that workflow into a sentence. Systems like Suno and Udio take a text prompt, “a melancholy synth-pop song about leaving home”, and return a finished, produced track: vocals, lyrics, mix and all, in about a minute.
The output is not reliably brilliant, but it no longer sounds like an experiment. It sounds like music.
The gap between “Daddy’s Car” and a Suno track is eight years. Whatever the current systems cannot do, betting on that gap staying open is a bad trade.
The Fake Drake Moment
“Heart on My Sleeve” was the track with the cloned Drake and The Weeknd vocals. An anonymous creator posted it, it racked up millions of plays within days, and the streaming platforms pulled it after Universal Music Group objected. The takedown worked; the demonstration could not be taken back.
Voice cloning is the sharpest edge of the whole field because a voice sits outside the categories copyright was built for. A song’s composition and its recording are protected works. The sound of a specific human being is something else: closer to identity than to property, and covered unevenly, in the US mostly by state-level publicity rights rather than federal copyright.
Lawmakers have started patching that gap. Tennessee’s ELVIS Act, signed in 2024, made a performer’s voice an explicitly protected property right, aimed squarely at unauthorized AI clones.
Grimes offered the counter-model within weeks of the fake Drake track. She invited anyone to make music with an AI version of her voice, in exchange for half the royalties on anything successful. Consent plus a revenue split, agreed before release instead of litigated after, is the opt-in template the industry keeps circling back to.
The Lawsuits That Decide the Field
The decisive fight is not about any single song. It is about the training data underneath all of them.
In June 2024, the major record labels, coordinated by the RIAA, sued Suno and Udio in US federal court. The complaints allege that both companies trained their models on copyrighted recordings without a license, and they seek statutory damages of up to $150,000 per infringed work. Across catalogs the size of the majors’, that arithmetic is existential.
The legal core is one question: is training a model on copyrighted recordings fair use, or infringement at industrial scale? The AI companies argue that training is “transformative” in the fair-use sense, that the model learns patterns rather than storing copies. The labels argue that a business built on their catalog, competing with their catalog, is the opposite of fair use.
Notably, neither company denies what the models learned from. In their court filings, both acknowledged that their training data includes copyrighted recordings and rested the defense on fair use itself. That narrows the cases to the question that matters, with no factual escape hatch.
The cases are unresolved, and the same question is being fought over text, images, and code in parallel suits across the industry. However the music cases land, they will set the economics.
Fair use means the training data was free and the value belongs to whoever built the model. Infringement means every serious music model needs a license, and the labels become the gatekeepers of the next generation of tools.
The Disclosure Layer
Underneath the ownership fight runs a quieter obligation: telling listeners what they are hearing.
The EU AI Act includes transparency duties for generated content, under which AI-generated media must be identifiable as such. Applied to music, that points toward labeling AI tracks rather than banning them. The Act regulates disclosure and leaves the copyright fight to copyright law.
Platforms and charts face the same question with no statute to lean on. Streaming services must decide whether AI tracks are tagged, demoted, or treated like any other upload, and chart compilers must decide whether a song with no human performer competes with ones that have them. Those policies are still being written, and they will shape whether an AI hit can officially exist at all.
So Can AI Compose a Hit?
Technically, the honest answer is: closer than most people think. The distance from “Daddy’s Car” to full produced tracks on demand took under a decade, and the fake Drake episode showed that listeners, given a familiar voice, do not reliably tell the difference.
Legally, nothing is settled. Whether training was theft, whether a cloned voice is protected, whether an AI track can hold a copyright at all: every one of those questions is open, in court or in front of regulators.
And the biggest question is not whether a machine can write a hit. It is who gets paid for the millions of human songs the machine learned from. The winner of that fight, more than any advance in the models, decides what AI music becomes: a licensed instrument inside the industry, or a parallel industry that replaces it.
