[et_pb_section fb_built=”1″ _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_row _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”]<\/p>\n
From mid-September 2026, Spotify listeners will start seeing a marker on certain artist profiles: an\u00a0<\/span>“AI Persona” badge announced on 11 August<\/a>. It signals that an artist’s public identity \u2014 their name, their face, their cover art \u2014 is generated by artificial intelligence and points to no real person. Eleven days before that announcement, a quieter event: the transparency obligations of the European AI Act took effect. And in the same two weeks, Deezer released the figure that ran through the press: more than one track in two uploaded to the platform each day is now AI-generated. You could read all this as platforms closing ranks against the machine. That would be a misreading \u2014 and it’s exactly the subject.<\/p>\n The badge will show up in three places: on the artist profile (header and “About” section), in search results, and on track rows inside playlists. Its effect isn’t cosmetic. By default, badged profiles drop out of editorial and algorithmic recommendations: their music no longer lands in personalised suggestions unless the listener has chosen to follow the artist. Spotify assigns the badge in two ways. Artists can declare themselves through Spotify for Artists, but the platform doesn’t rely on good faith alone: its teams review profiles whose name and imagery look like a fabricated photorealistic identity, starting with artists above a certain audience threshold. Profiles badged through that review are notified and can either self-declare or appeal. One point sits at the heart of the current confusion: the badge is about an artist’s public identity, not about how the music was made.<\/p>\n [\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_image src=”https:\/\/papacitoyen.reves-connectes.com\/en\/wp-content\/uploads\/sites\/3\/2026\/08\/musique-assistee-par-ordinateur-mao.jpg” title_text=”musique-assistee-par-ordinateur-mao” _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][\/et_pb_image][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”]<\/p>\n The striking number comes not from Spotify but from\u00a0<\/span>Deezer<\/a>. At the June 2026 peak, more than 50% of the tracks uploaded each day were fully AI-generated \u2014 roughly 90,000 a day. What matters is the slope: around 10% in January 2025, 39% in January 2026, 44% in April, more than half by summer. In sixteen months the volume multiplied ninefold. One detail changes everything, though: this flood is at the door, not in the ear. AI tracks account for just 1 to 3% of actual listening, and Deezer estimates up to 85% of those streams are fraudulent \u2014 flows inflated artificially to siphon off royalties. So the real, large-scale problem isn’t listeners rushing to synthetic music; it’s a catalogue saturating with uploads no one plays. Deezer, which has detected them since January 2025, now removes any that go six months without a play or that serve fraud. A detail that should humble any debate about “authenticity”: in an Ipsos survey commissioned by Deezer, 97% of people couldn’t tell a generated track from a human one.<\/p>\n Public debate piles four different realities under a single label \u2014 “AI in music” \u2014 that share neither the same stakes nor the same culprits. Separating them already answers half the questions.<\/p>\n The first is music\u00a0<\/span>composed<\/em>\u00a0<\/span>by AI: the track itself is generated \u2014 melody, arrangement, sometimes vocals. This is what Deezer detects. The second is AI-assisted<\/em>\u00a0<\/span>production: stem separation, automatic mastering, arrangement help. It’s everywhere, and perfectly legitimate \u2014 every home studio has used it for years without anyone calling it cheating. No one labels it, and rightly so. The third is the AI-generated artist\u00a0<\/span>identity<\/em>: the face, the name, the persona behind the music. This is what Spotify’s badge targets. The fourth is streaming\u00a0<\/span>fraud<\/em>\u00a0<\/span>and the industrial dumping of tracks \u2014 the real economic problem, the one that dilutes everyone’s royalties. A reader who leaves thinking “AI music” is a single category hasn’t been well served: these four things call for four different answers.<\/p>\n Once the boxes are drawn, the rest falls into place. Deezer labels content (the first box): it analyses the audio signal and flags synthetic tracks. Spotify labels identity (the third): it judges whether the face and name on show correspond to a real human. Presenting the two moves as platforms falling into line, as part of the press did, means confusing two distinct problems.<\/p>\n Spotify had in fact promised, back in September 2025, a content mechanism:\u00a0<\/span>support for the DDEX disclosure standard<\/a>, which lets an artist state where AI was involved \u2014 vocals, instrumentation, post-production. It does exist, under the name “AI Credits,” with tens of thousands of declarations a day, but it rests on goodwill and carries no penalty. The mechanism with teeth \u2014 removal from recommendations \u2014 is the identity badge. There’s the real shift: from an optional, consequence-free content disclosure to an imposed, penalising identity label.<\/p>\n The case of the fictional band\u00a0<\/span>Velvet Sundown<\/a>\u00a0<\/span>makes the gap clear. In 2025 this fully generated “band” pulled nearly a million monthly listeners in a few weeks, presenting itself as a folk-rock quartet with slightly too-smooth photos. It was Deezer’s content detection that flagged it, not Spotify \u2014 a year before the badge that now targets exactly this kind of identity imposture. For composer Ed Newton-Rex, founder of the nonprofit Fairly Trained, a case like this is theft dressed up as competition.<\/p>\n Widen the lens to the other platforms and it’s clear no common line exists. Apple Music, a pioneer of online music, introduced\u00a0<\/span>“Transparency Tags”<\/a>\u00a0<\/span>in March 2026: metadata labels that labels and distributors\u00a0<\/span>may<\/em>\u00a0<\/span>apply across four categories \u2014 artwork, recording, composition, video \u2014 when AI played a “material” role. Optional, self-declared, with no detection and no penalty: the lightest stance on the market, modelled on Spotify’s voluntary disclosure.<\/p>\n At the other end,\u00a0<\/span>Tidal<\/a>\u00a0<\/span>put the cursor where it hurts: since mid-July 2026, the platform detects fully generated tracks, flags them with an icon and, above all, stops paying royalties on them. No other major platform goes this far on money \u2014 and it isn’t French. In between,\u00a0<\/span>Qobuz<\/a>\u00a0<\/span>\u2014 French, like Deezer \u2014 published a “zero-tolerance” charter on 100%-AI content in February 2026: in-house detection across the entire catalogue, labelling, and a fully human curation of its 50,000 playlists that keeps industrial content out of the spotlight from the start. There is no better safeguard than a human, the platform sums up \u2014 and it also claims the highest per-stream payout on the market. Five players, five answers: “AI in music” doesn’t call for one policy, but for an editorial choice.<\/p>\n The badge does miss its target at the edges. It hunts the fabricated photorealistic identity: a producer dumping a thousand tracks under an abstract alias and a non-figurative cover slips through, with no fake face to flag. The case already exists at scale \u2014 a Swedish composer feeds Spotify’s ambient playlists under more than 650 aliases, without a single synthetic face. Conversely, a human artist who picks a generated avatar, out of taste or to protect their privacy, will get badged. The system catches what’s easy to judge \u2014 a face \u2014 rather than what matters: where the music comes from and who collects the royalties.<\/p>\n On top of that sits a blind spot no one documents: the false-positive rate. Platforms publicise the accuracy of their detection, never the number of human artists wrongly labelled, nor how long appeals take. Yet demotion applies by default, before any review: for an artist with no existing profile, dropping out of recommendations is disappearing, and the appeal only comes after the harm is done.<\/p>\n The counter-argument an informed sceptic would make is strong. Labelling and demotion hit emerging, unsigned artists first \u2014 the ones who use AI as a production crutch for lack of a studio budget. Established catalogues don’t need algorithmic recommendations: they have the name recognition, the negotiated playlists, the marketing. A badly calibrated “anti-AI” mechanism then becomes socially regressive \u2014 it raises a barrier to entry at the very moment technology had lowered it. And the fact that Spotify is separately building an\u00a0<\/span>“artist-first” AI music partnership with the majors<\/a>\u00a0<\/span>doesn’t ease the suspicion: AI generation blessed by the big players moves ahead, while the generated identity of the small ones is pushed back.<\/p>\n The counterpoint holds: at half of daily uploads, sheer volume mechanically threatens every artist’s discoverability, AI or not. Platforms aren’t pure cynics. But the way out is known, and French law already spells it out. As its May 2025 guidance notes, France’s collecting society SACEM holds that a work generated entirely without human creative input has no author and isn’t protected, whereas a composer who keeps control of their choices while using AI as a tool remains fully an author. The society is in fact\u00a0<\/span>pushing for licensing deals with AI platforms<\/a>. A policy that penalises fraud, declares full generation and leaves assisted work alone would protect both established artists and newcomers. No platform does that today.<\/p>\n This ridge line \u2014 penalise fraud without punishing the tool \u2014 has a spokesman who is hard to dismiss. Jean-Michel Jarre built the studio for\u00a0<\/span>Oxyg\u00e8ne<\/em>\u00a0<\/span>himself, has worked for years with the Sony Computer Science Laboratories on composition algorithms, and calls himself a technological utopian: hard to file among the technophobes. Yet it was he who, a\u00a0<\/span>former president of CISAC<\/a>\u00a0<\/span>for seven years \u2014 the confederation that unites more than 200 authors’ societies worldwide \u2014 carried creators’ voices against the tech giants. His position sums up this article: don’t confuse the tool with its use. He sees AI as a collaborator and a muse, not an enemy, but demands rules and refuses to see artists reduced to data suppliers.\u00a0<\/span>Speaking to franceinfo<\/a>, he boils the stakes down to a phrase: creators deserve their share of the digital pie. Coming from a man who spent his career taming machines, the demand isn’t that we ban them, but that we set rules for their use \u2014 exactly the distinction Spotify’s badge, on its own, fails to make.<\/p>\n Jarre’s stance isn’t isolated; it fits an alignment that, on closer look, is very French. The two platforms most aggressive on content detection are French \u2014 Deezer, first to move in January 2025, and Qobuz, which makes human curation its bulwark. The collecting society pushing hardest for a licensing-and-remuneration model is French: SACEM. And the creator who best embodies the refusal to choose between the machine and copyright is French: Jarre. The common thread isn’t transparency for its own sake, but the value of human creation and fair pay.<\/p>\n Does that make it a national doctrine? A sceptic will raise two fair objections. First, the toughest on money isn’t French: it’s Tidal that demonetises. Second, this “pro-human” stance conveniently suits challengers differentiating themselves from the American giants \u2014 principle and marketing blur together. The nuance holds. Still, no other country lines up such a concentration of consistent players.<\/p>\n At European level, by contrast, there is no path but a floor: since 2 August 2026 the AI Act has imposed a minimum of transparency on generated content, without dictating either on-platform labelling or any remuneration model. The rest plays out platform by platform. Europe sets the threshold; France draws a line \u2014 protect the human without banning the tool. It’s the programme SACEM articulates and Jarre carries. Whether French platforms will hold that line as economic pressure mounts remains to be seen.<\/p>\n Spotify’s badge and Deezer’s detection each answer a real question. Neither answers the one that decides a newcomer’s career: how do you get heard when half of daily uploads are synthetic and the sorting runs on the easiest variable to measure? No single badge will answer that. Meanwhile, the best answer is still listening \u2014 to electronic music made by humans who own what they do, in the lineage of a\u00a0<\/span>pioneer like Jean-Michel Jarre<\/a>\u00a0<\/span>or of the long history retraced in our\u00a0<\/span>journey through electronic music, from the 1950s to today<\/a>.<\/p>\n That’s the spirit of the R\u00ea>ON project: soundscapes composed by hand, gathered in the\u00a0<\/span>Electronica<\/em>\u00a0<\/span>playlist<\/a> . To explore the rest of what we listen to, the webzine’s\u00a0<\/span>music section<\/a>\u00a0<\/span>is open.<\/p>\n To keep up with future explainers and R\u00ea>ON releases, subscribe to the webzine’s newsletter. A question, a correction, a source to share?\u00a0<\/span>Get in touch<\/a>.<\/p>\n [\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”][et_pb_code _builder_version=”4.27.7″ _module_preset=”default” global_colors_info=”{}”]What listeners will see from mid-September<\/h3>\n
The figure that frames the debate: more than one track in two<\/h3>\n
Four things we should stop lumping together<\/h3>\n
Spotify targets identity, Deezer targets content<\/h3>\n
And the rest: Apple, Tidal, Qobuz, one phrase for five strategies<\/h3>\n
Spotify’s test catches the wrong variable<\/h3>\n
Who gains, who pays<\/h3>\n
What Jean-Michel Jarre thinks, and why his view carries weight<\/h3>\n
A French approach, for want of a European one?<\/h3>\n
Meanwhile, listen to humans who dream<\/h3>\n