If AI Can Make Unlimited Music, Who Decides What You Hear?
Last updated September 28, 2026. Lawsuits mentioned here are ongoing; allegations are not findings.
When music costs almost nothing to make, attention becomes the scarce resource. Whoever filters, ranks, and sequences the flood decides what you hear: streaming algorithms, labels and distributors, human curators, or the moment you're listening in. Making music is no longer the bottleneck. Choosing it is.
What happened this month
Three stories from the last few weeks point at the same shift.
Suno shipped a model built on licensed music
On September 9, Suno released v6, trained with licensed data from Warner Music Group, BMG, and Believe, and said it plans to retire its older models. Warner settled its lawsuit with Suno last year; BMG signed a deal in August. Sony and Universal are still suing. The labels that took AI companies to court are now also licensing them.
Universal sued DistroKid over AI uploads
On September 15, Universal Music Group filed suit against DistroKid in federal court in Delaware, alleging copyright infringement and deceptive trade practices and accusing the distributor of "flooding platforms with AI-generated 'slop.'" The complaint identifies 1,000 recordings and seeks statutory damages. These are allegations. DistroKid had not publicly responded when the suit was reported, and no court has ruled.
The upload numbers keep climbing
Deezer reported in July that fully AI-generated tracks passed 50% of its daily uploads at their June peak, roughly 90,000 tracks a day. Those tracks account for only 1 to 3% of streams, and Deezer says up to 85% of streams on fully AI-generated tracks in 2025 were flagged as fraudulent. Its response: AI-detected tracks are removed from algorithmic recommendations and editorial playlists, and are now tagged for listeners.
The supply curve broke. Attention didn't.
Put those stories together and you get a simple chain:
- AI drops the cost of making a track toward zero.
- Supply explodes. Half of new uploads on one major service, in about a year.
- Listening time stays fixed. Nobody got more hours in the day.
- The filter becomes the product. Whatever decides what reaches your ears matters more than what gets made.
The most telling number is the gap between half the uploads and a few percent of the streams. The flood is real on the supply side and mostly invisible on the listening side. That gap isn't an accident. It's the filter working.
Three places the filter can live
1. The platform
This is where most filtering happens today. Deezer keeps detected AI tracks out of recommendations. Spotify launched a spam filter aimed at mass uploads and duplicate tracks, banned unauthorized voice clones, and backed a DDEX standard for disclosing AI's role in music credits. Platforms work at enormous scale. The tradeoff is that you inherit their rules and their incentives, which are built around engagement across hundreds of millions of people, not around you.
2. The gatekeepers upstream
Labels and distributors decide what enters the system. Licensing deals like Suno's and lawsuits like Universal's are fights over that gate: who gets paid, what counts as legitimate, what gets distributed at all. That matters a lot for artists. For listeners it's indirect. It shapes what's available, not what plays next.
3. The moment you're in
The third option starts from a different question. Instead of "what's popular" or "what's allowed," it asks what fits right now: focus, a workout, winding down. A catalog of a hundred million tracks can't answer that by itself. Context narrows it to something a person can hear.
Licensing answers a different question
It's tempting to read the Suno deals as the resolution. They settle something real: whether a model was built on music its makers had the right to use. That's a legitimacy question, and it's the right one to argue about.
But a fully licensed model that can produce endless tracks still leaves you with endless tracks. Legitimacy isn't relevance. Even in the cleanest version of the AI-music future, someone or something still has to decide what's worth your next twenty minutes.
Where Noema fits
We built Noema around the third filter. There's no catalog to search. You pick a station (Focus, Relax, Workout, Sleep, Party, or Commute) and music starts. The decision about what plays is scoped to one moment instead of an infinite library. On Pro, Your Frequency reorders a station around what you finish versus what you skip, so the filter becomes yours over time. We wrote about how that skip-learning works in more detail.
To be straight about our side of this: we don't train the generation model. It runs on a third-party provider, and what we control is the app layer on top of it. Noema also doesn't solve discovering human artists. If you want to find a new band, a streaming service with good editors is still the better tool. Our honest comparison of AI music apps covers where each one fits.
What we're watching
- The Sony and Universal cases against Suno. Whether they end in licenses, like Warner's, or in rulings.
- DistroKid's response. The first real test of whether a distributor is responsible for what it distributes.
- Labeling. Deezer's tags and the DDEX credits standard decide whether listeners can see what's AI and choose for themselves.
- Recommendation policy. Whether other platforms follow Deezer in keeping AI tracks out of algorithmic picks, or go the other way.
Sources
- Suno replaces its AI models with a new one trained on licensed music as copyright suits pile up — TechCrunch, Sept 9, 2026
- Universal Music Group sues DistroKid, accusing it of "unlawful practices" and "flooding platforms with AI-generated slop" — Music Business Worldwide, Sept 15, 2026
- AI music tops 50% of daily uploads on Deezer — Deezer Newsroom, July 2026
- Spotify has deleted 75m+ "spammy" tracks as it unveils new AI music policies — Music Business Worldwide, Sept 25, 2025