Apple Music will label AI songs, and the tags behind them are now mandatory

Apple Music will start showing listeners which songs were made with AI. The disclosure behind those labels also stops being optional.

Apple emailed music industry partners on Thursday to announce Made With AI Labels, Billboard reported after obtaining the message. Any track a content provider tags as “materially generated using AI” will carry a visible label.

Apple Music is the last of the major services to put a mark like this in front of listeners.

The labels arrive “later this year”. Apple has not given a date.

What actually changed

Apple introduced AI Transparency Tags in March. Record labels and distributors could use them to disclose AI in artwork, a track, a composition or a music video.

Those tags were voluntary. Apple said at the time it would defer to content providers on what counted as AI.

Thursday’s email removes the choice. “Content providers will be required to include AI Transparency Tags in any instance where AI was used to create a material portion of the content, including tracks that are AI platform generated,” it reads.

Apple also set out what it means by that. “We define AI-platform-generated content as anything that is primarily derived from a generative AI service.”

The company told creators the goal is to “provide listeners with as much transparency as possible”, according to 9to5Mac, and pointed them to its Music Style Guide.

Apple is asking labels to mark their own homework

The email puts the burden on the people uploading. “As we implement Made With AI Labels, content providers will be important partners,” it says. “They are best positioned to know how their content was created.”

Apple can do this itself. Oliver Schusser, who runs Apple Music, told Billboard in April that the company built detection in-house.

“We have developed, and we’ve never talked about this, but we’ve developed technology in-house that would allow us to exactly see what music people are delivering us, what AI model it is and all that,” he said.

Asked why Apple lets providers self-disclose instead of using that system, Schusser gave a reason that explains the scale of the problem.

“Well, first of all, you should know that every label in the world is delivering AI, they might not know about it,” he said. “They themselves also need to develop tools to understand the scope of AI use.”

Nobody has said how this gets enforced

Apple’s email does not explain what happens to a provider that fails to tag a track. Billboard, The Hollywood Reporter and Digital Music News all noted the omission.

Apple has also not set a threshold. The email does not define “a material portion” beyond the AI-platform-generated case.

Nor has Apple said whether it will check tagged tracks against its own detection, or only take providers at their word.

The two numbers Apple has published

Schusser gave Billboard the scale in April. More than a third of what Apple Music receives each month is “100% AI” music.

That is uploads, not catalogue. The figure counts what arrives, not what sits on the service.

Almost nobody plays it. “The usage of the AI music on Apple Music is really tiny,” he said. “I’m rounding but it’s below 0.5% of usage.”

Apple has been acting on the fraud side for longer. It redistributed royalties from around 2 billion manipulated streams in 2025 into its payout pool for artists and labels, AppleInsider reported, and it doubled its penalty fees for streaming fraud in January.

Every other platform moved first

Apple is last among the major services to make its AI disclosure visible.

Spotify said this month it would badge AI artist profiles and keep them out of editorial and algorithmic recommendations. Those badges arrive in mid-September. Spotify labels the profile rather than the song, and artists can appeal.

TIDAL went further in June, saying it would strip royalties from AI-generated music and tag every track it catches.

Deezer has been running detection for over a year and catches around 60,000 AI tracks a day. In July it said fully AI-generated tracks made up more than half its daily uploads, and that most of them were fraud. Its detector already scans Apple Music playlists for listeners who want to check.

Others have skipped labelling and banned it. Bandcamp and Qobuz refuse fully AI-generated music outright, and Beatport said days ago it would do the same while tagging AI-assisted tracks, Stereogum reported. Amazon Music has not announced a policy.

The record industry asked for this

A coalition led by the RIAA and IFPI proposed a labelling system for streaming services weeks ago. It borrowed the way services already flag explicit tracks.

That proposal had two tags rather than one. “AI-generated” would cover songs where AI produced key instrumentals or lead vocals, or where the whole track came from a prompt. “AI-assisted” would cover songs using AI in places while humans still performed the lead vocals and instrumentals.

Apple’s system does not make that distinction. It has one label, which will sit on anything a provider tags.

The Digital Media Association, which represents the streaming services, said at the time that platforms were open to supporting the proposal.

Where the music itself comes from

The tracks Apple wants tagged largely originate on generative services such as Suno and Udio.

Suno moved on disclosure from the other end this month. It added watermarking and fingerprinting to its output while facing lawsuits.

What is not settled

No launch date exists for the labels, and Apple has not said whether the requirement starts before they appear.

Apple has not published a penalty for non-compliance, a definition of “material”, or any plan to apply its own detection to tracks that arrive untagged.

Schusser’s figures date from April. Apple has not updated them since, and has not said whether the share of AI uploads has moved.

Apple has also not said what a listener will be able to do with the label once it appears. Spotify and Deezer both keep fully AI tracks out of recommendations. Apple has announced a label and nothing else.