Is AI cover songs legal?

The Three-Question Legal Maze Facing AI Cover Creators

Ask three different lawyers whether AI cover songs are legal, and you'll likely get three different answers — not because they're bad lawyers, but because the question itself is legally broken into pieces no court has resolved. What looks like one activity — generating a cover of a song with AI — actually raises three distinct legal questions, each governed by different law and each currently unsettled.

The first question concerns training inputs: did the AI model learn from copyrighted recordings without permission? This is the issue at the heart of the June 24, 2024 lawsuits filed by Sony, Universal, and Warner against AI music generators Suno and Udio, which Reuters reported could turn on a fair-use defense no court has yet ruled on in the AI music context. The RIAA's position is blunt: the suits document "shameless copying of troves of recordings," while Udio insists it is "completely uninterested in reproducing content in our training set."

The second question concerns outputs: does your AI cover actually copy protected material? Here the law is clearer, and less forgiving. A legal-technical analysis concludes that reproducing copyrighted lyrics is prima facie infringement even with a new melody, while sound-alike vocals without actual sampling fall outside federal copyright under 17 U.S.C. §114(b), which excludes "merely imitated" sounds.

The third question — voice imitation — is mostly not copyright law at all. It lives in a patchwork of state right-of-publicity statutes, including Tennessee's 2024 ELVIS Act and California's Midler and Waits precedents.

Why the confusion persists despite all this activity:

• The U.S. Copyright Office's multi-part AI report found high-volume training copying "goes beyond established fair use boundaries," but directs courts to analyze cases individually rather than issuing categorical rules.
• Courts haven't ruled on music AI fair use; a book-training case (Bartz v. Anthropic) found training "exceedingly transformative," but that reasoning hasn't been tested for music.
• The stakes are severe: willful infringement carries statutory damages up to $150,000 per work.

For creators, the practical takeaway is that "is it legal?" has no single answer — it depends on which of the three questions your specific cover triggers. At SkillzBench, where we evaluate AI tooling and workflows, we see the same pattern in practice: the risk lives in the details of how a tool was built and used, not in a blanket verdict. The emerging safe path, signaled by Universal Music's October 2025 settlement and licensing deal with Udio, runs through licensed frameworks — not through fair-use gambles.

What the Law Actually Protects: Lyrics, Melody, and the Voice Loophole

Here's the puzzle that trips up almost everyone asking whether AI cover songs are legal: federal copyright law protects lyrics and melodies fiercely, yet it barely touches the thing everyone's actually worried about — the cloned voice singing them.

The compositional side is clear-cut. Federal law robustly protects lyrics and melody, and according to a legal-technical analysis of AI music, unauthorized reproduction of lyrics is prima facie infringement even when paired with a brand-new melody. Willful infringement carries statutory damages of up to $150,000 per work plus profits under 17 U.S.C. §504, assuming timely registration.

The vocal side is where the law has a gap. Under 17 U.S.C. §114(b), "merely imitated" sounds are explicitly excluded from copyright protection, so a sound-alike vocal that never samples the original recording generally falls outside federal copyright. As Harvard Law lecturer Louis Tompros put it after the viral "Heart on My Sleeve" Drake fake, the strongest argument against AI voice covers "is not a copyright argument at all, but a right of publicity argument."

That liability has migrated to state law, creating a patchwork:

• Tennessee's 2024 ELVIS Act prohibits unauthorized commercial use of a cloned voice and extends liability to the tech providers that enable it.
• New York added "voice" to its statutory right of publicity.
• California's common law — Midler v. Ford (1988) and Waits v. Frito-Lay (1992) — imposes liability when a distinctive, widely known voice is deliberately imitated for commercial use.

There's a third wrinkle: the AI output itself may own nothing. Under the human-authorship requirement, "works that are created by AI without human intervention or human involvement cannot be copyrighted," Tompros explains in Harvard Law Today — though hybrid works with genuine human creative choices in selection and arrangement can still qualify.

For creators and teams evaluating AI music tools, this fragmented landscape is exactly the kind of thing SkillzBench is building verification infrastructure around — pinning agent skills to specific versions and license-checking them, so the provenance of what your AI stack does isn't a mystery. The law draws sharp lines around lyrics and melody while leaving the voice to state courts; knowing which rule applies to which element is the first step in staying on the right side of it.

From Lawsuits to Licenses: How the Industry Is Resolving the Fight

The lawsuits that once defined the AI music fight are quietly giving way to something else: deals. In October 2025, Universal Music Group settled with Udio and signed licensing agreements covering its recorded music and publishing catalogs — a move legal observers describe as a watershed moment and a blueprint for licensing catalogs for AI-generated content.

The numbers behind this shift explain why. Anthropic reached a $1.5 billion preliminary settlement with a class of authors in September 2025, according to the D.C. Bar's Washington Lawyer — a figure that makes betting on fair-use defenses look increasingly expensive. And abroad, the Munich Regional Court sided with GEMA against OpenAI in November 2025, finding that unlicensed use of protected musical works for training violated German copyright law, in what is regarded as the first major European ruling on the issue.

Meanwhile, platforms are enforcing their own standards. Spotify removed approximately 75 million AI-generated "spammy tracks" over a twelve-month period announced in September 2025, and major labels are deploying neural fingerprinting technology to detect AI-derived infringement before it spreads.

Why licensing is winning over litigation:

• No U.S. court has ruled definitively on fair use for AI music training, leaving both sides exposed to unpredictable outcomes, as Reuters reporting notes.
• The Copyright Office's AI report found that high-volume copying for training "goes beyond established fair use boundaries," but left the final call to case-by-case court analysis.
• Music AI faces a harder fair-use path than text AI because generated songs compete in the same commercial purpose as the original artist.
• Settlements deliver certainty faster than years of appeals — the UMG–Udio deal gives both parties a working framework immediately.

For creators, the practical takeaway is clear: licensed frameworks are becoming the safe route, while unlicensed AI covers remain a legal gamble. RIAA CEO Mitch Glazier framed the industry position plainly — AI has "great promise," but only on a "sound, responsible, licensed footing."

The same verification mindset applies to the broader AI tool ecosystem, where SkillzBench is building independent benchmarking and security scanning so users can see which agent skills actually deliver lift rather than risk. In music, as in AI tooling, the market is moving from "trust me" to "show me the license — and the evidence."

Your Practical Playbook: Releasing AI Covers Without Legal Exposure

The difference between a legal AI cover and a lawsuit often comes down to a handful of decisions made before you hit publish. Statutory damages for willful infringement run up to $150,000 per work under 17 U.S.C. §504, so a few minutes of diligence is worth more than any apology afterward.

Start with the composition itself. The safest path is to build AI covers on original material or properly licensed compositions — never copy protected lyrics or melodies. Unauthorized lyric reproduction is prima facie infringement even when the melody is entirely new, and legal analysis of AI music finds federal law "robustly protects lyrics and melody."

Voice is the second landmine. Although copyright doesn't protect a performer's voice, state law does. Tennessee's 2024 ELVIS Act prohibits unauthorized commercial use of cloned voices and extends liability to the tech providers themselves, while precedents like Midler v. Ford and Waits v. Frito-Lay impose liability when a distinctive, widely known voice is deliberately imitated for commercial gain. If your AI cover sounds like a famous artist, don't market it that way.

Disclosure is your cheapest insurance. Reed Smith attorneys note that clear AI disclaimers may defeat UK "passing off" claims and US false-endorsement claims, since both require misrepresentation that the real artist is involved. Label the track as AI-generated in the title, description, and metadata.

For creators releasing AI-assisted music, the practical playbook looks like this:

• Use original or licensed compositions — never reproduce protected lyrics or melodies in your prompts or output.
• Avoid cloning recognizable voices for commercial release, even with disclosure.
• Disclose AI use prominently to undercut passing-off and false-endorsement claims.
• Track the licensing model — Universal Music's October 2025 settlement with Udio, called a "watershed moment" by industry observers, signals that licensed AI frameworks are becoming the viable path.

If you're on the other side — an artist whose work or voice is being exploited — the guidance from the D.C. Bar's Washington Lawyer is equally concrete: register your copyrights before posting anything, audit your contracts for AI clauses that may grant sweeping rights, and document economic harm as it occurs.

One caveat applies to everything above: the fair-use question for AI music training remains unresolved in court, and platforms are enforcing aggressively — Spotify removed roughly 75 million AI-generated "spammy tracks" over twelve months. That's why verification matters. At SkillzBench, our mission is checking what's trustworthy before you rely on it, and in music as in AI tools, the cost of skipping that step keeps rising.