AI Music Digest

Apple told music industry partners in a message on August 20 that 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.” The tags, introduced in March as an optional metadata field, will become a delivery requirement, and will surface to listeners as “Made With AI” labels on Apple Music later this year. Apple has not given a launch date.

The tagging system covers four elements: Artwork at the album level, Track at the recording level, Composition for AI-generated lyrics and compositional elements, and Music Video. Responsibility for applying the tags sits entirely with labels and distributors, which Apple has said are best positioned to know how a recording was made. Apple’s qualifying standard — content “primarily derived from a generative AI service” — has drawn criticism as open to interpretation, and unlike Deezer, Apple is not deploying its own detection technology to check what it is told.

Why It Matters: The largest paid music service is shifting AI disclosure from voluntary to required. But a mandate enforced through self-reporting, with no independent detection behind it, tests whether the industry’s disclosure model can work on the honor system.

Beatport confirmed on August 12 that it will no longer sell tracks generated wholly or mostly by AI. Such releases are withheld during ingestion, with rights holders notified directly rather than seeing tracks silently disappear. Music made with AI assistance is still accepted provided the finished recording remains majority human-made, and those releases are tagged at ingestion so Beatport’s curation team can see what it is reviewing and featuring.

The policy runs on an expanded partnership with music fraud detection firm Beatdapp, which Beatport says brings updated technology and processes for identifying automated and generative music across its catalogue. Beatport pointed to its own user research in support of the move, saying 60% of surveyed users would not play AI-generated tracks in their sets and 77% expressed a firm preference for supporting human creators. Those figures come from Beatport’s survey of its own community and have not been independently audited.

Why It Matters: Streaming platforms have converged on labeling AI music. Beatport, a storefront serving DJs who buy tracks to perform, went further and refused to stock it — the first significant commercial venue to treat mostly-AI music as unsellable rather than merely disclosable.

Spotify announced on August 11 that it will begin applying an “AI Persona” badge to artist profiles whose identity is AI-generated rather than a real person. The badge starts appearing in mid-September on artist profiles, in search results and on track rows; tapping it tells the listener whether the artist self-disclosed or Spotify made the determination. Artists can declare themselves through Spotify for Artists immediately, and flagged artists are notified and can appeal.

The consequential part is downstream of the badge. By default, Spotify says AI Personas are excluded from editorial and algorithmic recommendations unless a listener actively follows them. Spotify’s review team will independently identify profiles presenting photorealistic AI-generated identities, focusing initially on profiles that meet defined audience thresholds — which the company says covers the majority of artist profiles listeners actually visit.

Why It Matters: Excluding AI personas from recommendations is a distribution penalty, not a disclosure. On a platform where discovery drives most listening, that is a materially different lever from a label — and Spotify is applying it to identity rather than to how a track was made.

ElevenLabs announced Composer, a section-level editor for ElevenMusic, its generative music platform. The company says users can start from their own lyrics, an existing track, a prompt or a blank slate, then build and revise a song section by section — swapping a bridge, doubling a chorus, adding an outro or rearranging the order — rather than regenerating a whole track. Multiple takes of an individual section can be generated and compared before one is kept.

ElevenMusic launched in April 2026 as a consumer creation and streaming service, built on licensing deals with Kobalt and Merlin announced the previous year. Composer is the platform’s most substantial editing update since the Music v2 model arrived in May. Details of the feature come from ElevenLabs’ own announcement; Digital Music News is the only openly readable trade to have covered it so far.

Why It Matters: Generative music tools are converging on editability. Section-level control moves these platforms away from slot-machine generation and toward something closer to a production workflow, which is where the argument for AI as an assistive tool rather than a replacement gets made.

Alibaba launched HappyShrimp 1.0 in beta on August 17, a text-to-song model that the company says generates melody, arrangement, lyrics and vocals from a prompt describing an emotion, story or genre. Alibaba says the model covers Chinese-style music, pop, R&B and soul, hip hop, rock, funk, electronic, classical and jazz, and that users can supply their own lyrics for the model to set to music.

Alibaba also said HappyShrimp will work with Taihe Music Group, one of China’s larger music companies, on artist co-creation and content development, and that the two will explore applications in commercial advertising and film and television production. Alibaba has not publicly detailed what data the model was trained on.

Why It Matters: A company at Alibaba’s scale entering text-to-song puts a second major Chinese platform in the market alongside Kunlun Tech’s Mureka. The Taihe partnership signals a licensed, label-adjacent path rather than the litigation Western generators walked into.