Tidal is labelling AI tracks and cutting their royalties. That's the real shift.

For about a year the AI-music conversation on streaming platforms has been about labels — a tag on a track saying a machine was involved. Useful for listeners, largely cosmetic for economics.

Tidal changed that. From 15 July 2026 it began labelling wholly AI-generated tracks and making them ineligible for royalty attribution (Variety).

That second half is the actual news. A label is information. Cutting royalty eligibility is a business decision about who gets paid.

Why a label alone was never going to be enough

The scale problem is the reason. On Deezer, AI-generated tracks account for roughly 39% of daily uploads — around sixty thousand tracks every day — with more than 13.4 million detected and tagged during 2025 (iMusician).

At that volume, the concern stops being aesthetic. Streaming royalties come from a shared pool. Every fraudulent stream on a mass-uploaded track is money that would otherwise have reached someone else. A significant share of the streams these uploads generate appears to be fraudulent, which quietly reduces what human artists are paid.

Labelling tells a listener what they are hearing. It does nothing about the pool.

Everyone is arriving, by different routes

Tidal is not alone, but the platforms have not converged on one approach.

• Apple Music has confirmed AI transparency tags are coming, with the responsibility for accurate tagging placed on labels and distributors (RouteNote).
• Deezer has been tagging publicly the longest and has the clearest data on the scale of the problem.
• Suno, on the generation side, is adding audio watermarks and a labelling system so tracks stay identifiable after leaving the platform (Engadget).

Three of those are detection. Tidal’s is enforcement.

The distinction that will matter most

Every one of these policies depends on a line between AI-generated and AI-assisted, and that line is where the difficulty lives.

Tidal’s policy targets wholly AI-generated tracks. Reasonable in principle. Extremely difficult at the edges. Where does a track sit when the composition is human and the vocal is generated? When the lyrics are written and directed by a person, and the production is not? When a model was used the way a producer uses any other tool?

Most serious AI-native work sits somewhere on that spectrum rather than at either end. A policy that treats “a machine was involved” as a single binary state will misclassify a great deal of it — in both directions.

What this means if you release music made with AI

Three practical things, none of them speculative:

Tag your own releases. Apple Music is putting the obligation on labels and distributors. Being accurately tagged by you is a materially better outcome than being auto-flagged later, and it is the difference between disclosure and detection.

Expect the definition to move. These policies are new and the generated/assisted boundary will be redrawn more than once. Anyone whose distribution strategy depends on landing on a particular side of it is building on sand.

Understand what is being priced. The industry is not really pricing whether a machine was involved. It is pricing volume without intent — mass-uploaded tracks gaming a shared pool. That is a real problem and it deserved a response.

The uncomfortable part

The cleanest defence against these policies is to have been open from the start. Nothing to detect, nothing to reclassify, no moment where a listener feels misled.

That is a genuinely easier position — and it is worth saying plainly that it is easier for a project with two records than for anyone with a catalogue built under different assumptions. Retrofitting transparency onto existing work is a harder problem than starting with it, and the platforms are not being particularly gentle about the transition.

What is coming is not a debate about whether AI music is allowed. It is a slow sorting of the difference between people making things and people flooding a pool, and the tooling to tell them apart is getting better quickly.