In 2026, platforms began sorting music by how it was made

Until 2025, a track delivered to a streaming platform was simply a track. In 2026, it began to be classified as human-made, AI-assisted or AI-generated. Whether declared by a distributor or determined by a detector, that classification now governs how the track is treated. This first edition of VERIMUSIC's monthly news review traces the shift from January to September 2026, event by event. The figures are those reported by the platforms themselves and, to our knowledge, have not been independently audited. We make no accusations and aim simply to show artists, one development at a time, how the ground has shifted beneath their feet.

In January, Deezer puts its detector on the market

On 29/01/2026, Deezer announced that it was now selling its AI detection technology, "thereby offering the whole music sector an option to promote transparency in music streaming". The platform said it had previously "carried out successful tests with industry leaders, including Sacem", and it signed a commercial agreement with Sacem that same day. Deezer also released its own figures. Some 60,000 AI-generated tracks were being delivered to the service each day, roughly 39% of daily deliveries. More than 13.4 million tracks had been detected and tagged in 2025, and fraudulent streams had accounted for "up to 85%" of streams of fully AI-generated tracks that year. Those tracks, it added, are "automatically removed from algorithmic recommendations and do not appear in editorial playlists". Sources: Deezer newsroom, 29/01/2026; Zonebourse via Boursorama, 29/01/2026.

In March, Apple Music asks labels and distributors to declare AI use

On 04/03/2026, Apple Music introduced its "Transparency Tags", four metadata fields (artwork, track, composition, music video) that labels and distributors can fill in when AI has been used to create a substantial part of the content. Apple, which does not carry out its own detection, "defers to content providers to determine what qualifies as AI-generated content". The tags were optional at launch, and Apple said they would become mandatory for new deliveries without specifying a date. In late August, the trade press reported that this would happen before the end of 2026, although Apple's own notice had not been made public. Sources: Music Business Worldwide, 04/03/2026; Exron Music, 25/08/2026 (secondary source).

In April, Spotify launches AI credits and AI tracks reach 44% of Deezer's uploads

On 16/04/2026, Spotify launched its "AI credits" in beta. When a distributor uses the DDEX metadata standard to declare AI involvement, that information appears in the track's credits, with separate fields for vocals, instrumentation and post-production. DDEX is the standard used by the music industry to exchange metadata. DistroKid was the first distributor to connect to the system. The declaration is purely informational and does not trigger a warning or lower the track's ranking. Sources: Spotify newsroom, update to the post of 25/09/2025; Billboard, 22/04/2026; Undetectr, 2026.

On 20/04/2026, Deezer updated its figures. Nearly 75,000 AI-generated tracks were now being uploaded to the service every day, roughly 44% of daily uploads and more than two million a month, although they still accounted for only 1% to 3% of total streams. The platform also said it had stopped storing high-resolution versions of these tracks. Source: Deezer newsroom, 20/04/2026.

In late April, a distributor restricts AI tracks and Spotify introduces a badge for human artists

On 30/04/2026, Believe and TuneCore announced that they were automatically blocking the distribution of tracks produced partly or entirely using unlicensed AI generators, which they described as "pirate studios", including Suno. Believe's CEO described its detection technology as "99% reliable" at identifying the source model. Tracks made with licensed tools such as Udio and ElevenLabs remained eligible for distribution, and no appeal procedure was described. For the first time, a major distributor was deciding which generator a track could come from before it was even delivered to a platform. Source: Music Business Worldwide, 30/04/2026.

Also on 30/04/2026, Spotify launched the "Verified by Spotify" badge, a green tick reserved for artist profiles with an identifiable presence both on and off the platform (concerts, merchandise, linked accounts) and sustained listener interest. Spotify stated that "Profiles that primarily represent AI-generated music or AI-persona artists are not eligible." According to the platform, more than 99% of the artists listeners actively searched for were expected to be verified at launch. Source: TechCrunch, 30/04/2026.

In June, a free AI detector becomes available to the public

On 11/06/2026, Deezer launched a free online detector that lets listeners check whether playlists from 20 streaming services contain AI-generated tracks. Available in 27 languages, the tool recognises fully AI-generated music with 99.8% accuracy, according to Deezer. The company also says that 97% of listeners cannot distinguish it from music made by humans in a blind test. A technology sold to the music industry since January had become a consumer product. Source: Deezer newsroom, 11/06/2026.

On 24/06/2026, the dance music store Traxsource announced that, from 01/07/2026, every track in its catalogue would carry either a "Human-Made" or an "AI-Assisted" label, while fully AI-generated music would be removed. The classification relies on two detection partners, SH Labs and SoundPatrol. The announcement also provides for "a rebuttal process for content providers to submit disputes if they feel their music has been mislabeled". To our knowledge, no music store had previously introduced a specific label for hybrid creation. Source: Record of the Day, 24/06/2026.

On 29/06/2026, TIDAL immediately stopped paying royalties and passing on direct payments from fans for releases it classifies as fully AI-generated. It also announced that these releases would carry a visible "AI" badge from mid-July, while tracks impersonating artists would be removed. The service expects distributors to flag AI content before delivery, but its announcement named neither a detection partner nor an appeal procedure. An appeal form only appeared later on its help page, updated on 20/07/2026. Artists who use AI tools to support human creative work are explicitly not covered by these measures. Sources: The Next Web, 29/06/2026; TIDAL, AI Policy; TIDAL Support, 20/07/2026.

In July, AI tracks exceed half of Deezer's uploads and research turns to hybrid creation

On 21/07/2026, Deezer reported that AI-generated tracks now accounted for more than 50% of daily uploads, around 90,000 a day at the June peak, while still representing only 1% to 3% of streams. The platform followed this with detection figures reporting 99.8% accuracy, around two AI-generated tracks missed per thousand and fewer than one human-made track wrongly flagged per 10,000. These are Deezer's measurements, obtained on Deezer's test set. The platform also announced that it would remove AI-generated tracks used for fraud, along with those that had not been streamed for six months. Source: Deezer newsroom, 21/07/2026.

On 29/07/2026, a research paper entitled "Detection of AI-generated stems within hybrid human-AI music" examined whether an AI-generated part remains identifiable once it has been mixed with stems, or separate tracks, performed by humans. Its authors found that the artefacts of an AI-generated stem "are not reliably recovered by generic source separation systems". They proposed a different architecture whose results at track level they describe as "encouraging". Detecting an AI-generated layer in music made by humans is therefore a different problem from detecting a fully AI-generated track, and that problem remains unsolved. Source: arXiv 2607.26874, 29/07/2026.

In August, EU rules take effect and Spotify announces AI Persona badges

On 02/08/2026, the transparency obligations under Article 50 of the AI Act, the European Union's regulation on artificial intelligence, took effect. Providers of generative AI systems must now mark synthetic content in a machine-readable format so that it can be detected. Systems already on the market have a grace period until December 2026, and the Commission published guidelines on 20/07/2026. These obligations apply to providers and deployers of AI systems, not to individual artists. Whether the machine-readable marking will ever reach the listener's screen remains a question the regulation leaves open. Source: European Commission, quick facts on the transparency rules, updated 29/07/2026.

On 11/08/2026, Spotify announced an "AI Persona" badge, rolled out from mid-September, for artist profiles whose identity "may be AI-generated and does not represent a real person". Artists can declare their profiles themselves, but Spotify also examines profiles on its own initiative, through human review and its own tools. Profiles carrying the badge are excluded from editorial and algorithmic recommendations unless a listener follows them. Artists whose profiles receive the badge following this review "will be notified and given the opportunity to self-disclose or appeal". Source: Spotify newsroom, 11/08/2026.

Also on 11/08/2026, the distributor iMusician published guidance for artists challenging the labelling of their tracks. The company acknowledges that "there is no standardized industry-wide appeals process", advises contacting the distributor first and points out that "the final decision lies with the streaming platform itself". It recommends keeping digital audio workstation (DAW) project files, stems, session exports and dated drafts. Source: iMusician, 11/08/2026.

In September, Believe changes its policy and a provenance registry launches

On 08/09/2026, Believe and Suno announced a global licensing deal. Tracks made with Suno's forthcoming "industry partner model" become eligible for distribution through Believe and TuneCore on an opt-in basis, while tracks produced with Suno's existing models remain blocked. Four months after closing the door, the distributor is partly reopening it. The question is shifting from whether AI was used to which model was used and under what licence. Source: Music Business Worldwide, 08/09/2026.

On 10/09/2026, SIQA (The Sonic Intelligence Academy), a US company backed by 1517 Fund, launched the "SIQA Verified Registry" and its companion app. Artists register their works there, and labels, distributors, platforms and listeners can look up whether a track is classified as fully AI-generated, AI-assisted or a hybrid of human and AI work. The privately run scheme is based on registration and sits outside platform policies. We mention it because it is the first public registry built on a three-level scale rather than a simple yes or no. Source: PR Newswire via Morningstar, 10/09/2026.

The implications for artists

These nine months of changes have three practical implications for artists.

First, a track's origin has become a metadata field supplied when the track is delivered, whether you fill it in yourself or a detector does it for you. Apple and Spotify rely on what your distributor declares, while Deezer, TIDAL, Traxsource and Believe use detection or rely on its results. Without knowing which approach each platform uses, it is difficult to understand a decision about your own track.

Distributors have also become central both to getting music onto platforms and to challenging decisions. Declarations go through them, some tracks are blocked at the distribution stage, and every appeal must pass through them too. Choosing a distributor now also means choosing an AI policy and an appeal procedure.

Then there is the question of evidence. What matters is the material you keep before releasing your music (timestamped sessions, stems, dated drafts), together with an honest record of the tools you used. None of the procedures described above specifies what evidence it will accept, but all are easier to follow with those records to hand. Our artist's guide takes you through the process step by step.

Unanswered questions

Answers on the right of appeal remain scarce. Only Deezer publishes an appeal procedure that specifies how long to wait before following up, without committing to a response time. Spotify offers an appeal against the AI Persona badge, TIDAL has provided an appeal form since July 2026 and Traxsource mentions a procedure, while Believe's blocking policy provides for none. No platform publishes a list of the documents it accepts, its response time or the proportion of decisions it reverses.

To our knowledge, none of the detectors has been audited. The accuracy figures of 99.8% from Deezer and 99% from Believe come from the companies using these tools, and we have found neither an independent evaluation nor results broken down by genre, language or production budget. Yet overall accuracy on a test set says nothing about a recording made in a bedroom, in a minority language.

There is still little provision for hybrid creation. Most platforms continue to classify music in binary terms, even though the research published in July shows that detecting an AI-generated layer in music made by humans remains an open problem. Traxsource and SIQA have adopted three-level scales, but there is no agreement on where to set the thresholds.

Finally, there is the question of consent. Cloned voices and training on copyright-protected works feature in these policies only in relation to fraud or impersonation. None of the measures listed here specifies whose consent is required or how someone can prove that a voice is their own.

We will update this review every month. If one of these decisions has affected your music, share your experience with the observatory of disputed decisions. Bringing together documented cases will help answer these questions with facts, not just figures.

All sources were accessed on 02/10/2026. Platform figures are those reported by the platforms themselves; to our knowledge, they have not been independently audited. We have excluded any event for which we could not find a dated source. This article provides general information and does not constitute legal advice. Quotations from French-language sources are our own translations.