Udio admits it pulled training data with yt-dlp, then argues that's fair use

Udio has answered Sony Music’s second copyright lawsuit, and it names the tool it used to collect training data. Udio acknowledges acquiring some of that data with yt-dlp, an open-source program that downloads audio straight off YouTube.

Udio conceded the YouTube scraping itself back in April, answering Sony’s amended complaint in the original case. What’s new is the specificity, and the venue: this is the second lawsuit, with 30,117 recordings attached, and Udio is now naming the mechanism on the record.

Its position throughout has been that the act was legal.

The answer responds to the complaint Sony filed in late July, after a judge blocked its attempt to fold those recordings into the original 2024 case. Sony is seeking damages for training-related infringement, DMCA violations and more.

Why naming yt-dlp matters in a case that already has a live DMCA claim

Sony’s complaints include claims under the DMCA’s anti-circumvention rules. That’s the part of US copyright law making it illegal to break the technical locks on a protected work.

This theory has already been tested once, and it held. In April 2026 the New York court refused to dismiss the DMCA claim in the original case, ruling it could proceed. How an AI company obtained its data carries its own liability, separate from whether training is fair use.

So Udio is repeating an admission into a claim that a judge has already let stand.

The remaining legal question is narrow and genuinely unsettled. The DMCA treats two kinds of protection differently: controls that stop you copying a work, and controls that stop you accessing it in the first place. Bypassing an access control is the clearer violation.

Nobody has definitively ruled which kind YouTube uses. AI developers have argued consistently that YouTube’s measures are copy controls, which would put stream ripping on safer ground.

Udio’s fair use argument rests on what the listener never sees

On training itself, Udio leans on fair use, the rule letting you use copyrighted work without permission in limited cases. Its framing is that the copying is invisible.

The platform says its AI tool uses a back-end technological process, invisible to the public, in the service of creating an ultimately non-infringing new product, and that this process is quintessential fair use.

Read that carefully, because the emphasis is doing work. Udio isn’t arguing the training copies were licensed. It’s arguing they don’t count, because what reaches a listener is something new.

“

A back-end technological process, invisible to the public, in the service of creating an ultimately non-infringing new product.

Udio’s copyright misuse counter-argument targets Sony’s licensing position

Udio’s answer goes further than defense. Among its affirmative defenses is copyright misuse, a doctrine that can make a copyright unenforceable when the owner uses it to reach beyond what copyright actually grants.

Udio accuses Sony of anticompetitive activities that extend an unlawful monopoly over the production and commercialization of music.

The argument has an obvious factual hook. Udio now holds licenses with Universal Music, Warner Music, Merlin, Kobalt, Believe and the NMPA. Sony is the last major standing outside, and it’s the one suing. Udio wants the court to look at that pattern and ask why.

Whether a judge finds that persuasive is a different matter. Copyright misuse is a hard defense to win on, and refusing to license your own catalog is ordinarily just a business decision.

What Udio’s answer changes for artists and producers

If you release music, the practical stake here is the DMCA question, not the fair use headline.

A ruling that YouTube’s protections are access controls would make stream ripping for training data plainly illegal, regardless of what the model outputs. That’s a cleaner hook for rightsholders than the fair use fight, which has been grinding through courts for two years without a settled answer.

It would also apply to every developer that built a catalog the same way, which is most of them.

Nothing here resolves the case. Udio still says training is fair use, Sony still says it isn’t, and the court has years of runway. The part that has actually moved is the record: the how of Udio’s data collection is now documented in its own filings, twice, and the claim built on it has already survived one attempt to kill it.