The takeaway: On Friday, eight music-industry bodies — the global and US record-label groups IFPI and RIAA, the independent-label bodies A2IM, WIN and IMPALA, the Recording Academy, the performers' union SAG-AFTRA and the Human Artistry Campaign — introduced a voluntary labelling programme for AI in recorded music: an "AI-Generated" mark for tracks that are mostly or entirely machine-made, and an "AI-Assisted" mark for human recordings that used AI along the way. It's the first industry-wide answer to a flood that Deezer measures at 44% of all new uploads. The catch is in who's missing: the labels would be applied by artists, labels and distributors themselves — and not one streaming platform has committed to displaying them. Spotify declined to comment.

Two tiers, one honest distinction

The scheme, announced by the RIAA on 10 July, draws the line most listeners actually care about. A track gets the AI-Generated mark — upper-case "AI" in a black tile — when "Generative AI was used to generate the entirety or the primary portion of the creative elements of the recording": prompt-to-song output, or tracks where the lead vocal or main instrumental parts came from a model. The AI-Assisted mark — lower-case "ai" in a white tile — covers the other case: "The recording was created substantially by humans and expresses human creativity; however, generative AI was used for some expressive elements."

That second tier matters, because AI-as-tool is already everywhere in production — stem separation, mixing, vocal tuning — and a scheme that stamped a scarlet letter on all of it would have been ignored. The joint statement from IFPI chief executive Victoria Oakley and RIAA chairman and CEO Mitch Glazier puts the rationale simply: "Fans want to know whether and how generative AI has been used in the music to which they listen."

The scope is deliberately narrow for now: the programme covers the recording only — generative AI used in lyrics, composition, cover art or music videos isn't labelled "at this point". There are no dates, either; the bodies say the marks "will be available for use in the near future" while they work with streaming services, distributors and standards bodies on implementation.

The numbers behind it

The scale of what's being labelled is no longer in dispute. Deezer — the first platform to detect and tag fully AI-generated tracks at platform level — reported in April that "AI-generated tracks now represent 44% of all new uploaded music", nearly 75,000 tracks a day. The same report carries the two numbers that explain the industry's motive: that flood earns "between 1-3% of the total streams" — and 85% of those streams are detected as fraudulent and demonetised. AI music, at today's volumes, is less a creative wave than a royalty-skimming one. The RIAA's own release cites Apple Music saying more than a third of tracks uploaded to its platform are "100% AI". A CISAC-commissioned study projects 24% of music creators' revenues will be at risk by 2028 — a cumulative €10 billion loss for music.

And listeners genuinely can't tell: in Deezer's survey with Ipsos last November, 97% of respondents couldn't reliably distinguish AI-generated tracks from human ones. That's what made last summer's Velvet Sundown episode — an AI "band" that racked up over a million Spotify monthly listeners before being exposed — a preview rather than a one-off. Deezer's detector tagged that project as fully AI-generated; Spotify never flagged it.

The catch: the half that displays the label hasn't signed

Read the signatory list again: it is entirely the rights-holder side of the industry. The scheme is voluntary and self-applied — "flagged voluntarily by artists, record labels and distributors", as the Wall Street Journal's report, cited by Music Business Worldwide, puts it — with no verification or detection mechanism attached. The parties who would actually put the tile in front of listeners are conspicuously uncommitted: AFP reports that Spotify declined to comment and Apple Music didn't respond. DiMA, the streaming services' trade body, offered a cautious welcome without adopting anything. (To be fair to the platforms, both run disclosure systems of their own — Spotify adopted the DDEX metadata standard for AI credits last September, and Apple Music introduced declaration-based transparency tags in March — so "uncommitted" applies to these marks specifically.) Only Deezer leaned in, calling it "encouraging to see steps being taken towards a unified approach to generative AI in music" and saying it's ready to support an industry-wide framework — which costs Deezer little, since it built the detection infrastructure more than a year ago.

The self-reporting design has an obvious hole, and critics found it within hours: the people flooding platforms with 75,000 tracks a day to skim royalties are precisely the people who won't volunteer a label. And an artist quietly using AI for a lead vocal has a stigma-shaped incentive to stay quiet too. What the scheme really does is create a standard vocabulary — like the "Explicit" tag, which also depends on platforms choosing to display it — and put polite pressure on the streaming services to adopt it.

The regulatory clock

The timing isn't accidental. On 2 August — three weeks after this announcement — the EU AI Act's Article 50 transparency obligations become applicable, requiring machine-readable marking of AI-generated content, with the European Commission's voluntary Code of Practice on labelling already published. In the United States there is no equivalent: Tennessee's ELVIS Act and the federal NO FAKES Act (which cleared the Senate Judiciary Committee unanimously in June) protect voices and likenesses from cloning, but nothing in US law requires anyone to disclose that a song is synthetic. The industry proposal, in other words, is the music business volunteering worldwide for roughly what Brussels is about to mandate regionally — and getting ahead of whatever comes next.

For listeners, the test of whether any of this matters is simple and short-term: watch whether Spotify, Apple Music, YouTube Music or Amazon commit to displaying the marks. A label that exists only in a press release tells you how a song was made exactly as well as no label at all.