Apple Music will label AI-generated tracks, a SubmitHub study finds 40% of July releases used AI, Alibaba enters the music generation race with HappyShrimp, and Suno Studio 2.0 brings MIDI into the generative workflow.

Apple Music to Add "Made With AI" Labels

Apple emailed music industry partners on August 20 announcing that visible "Made With AI" labels will appear on Apple Music later this year. The labels will flag tracks materially generated by AI platforms like Suno and Udio, but won't apply to songs that simply used AI-assisted tools during production — mastering, EQ, and similar technical workflows are excluded. The system builds on Apple's existing AI Transparency Tags, introduced in March for labels and distributors to disclose AI use during content delivery.

This is the clearest signal yet that the major streaming platforms are done pretending they can ignore the problem. If you distribute AI-generated music, expect disclosure to become mandatory infrastructure, not optional etiquette. The distinction between "AI-generated" and "AI-assisted" will become the most important line in music distribution this year.

Source: MacRumors, Variety, NME

SubmitHub Study: 40% of July Releases Used AI

SubmitHub's AI detection tool SH Labs analyzed over a million releases from July 2026 and found that 23.2% were fully AI-generated, with another 15.3% showing signs of AI-generated audio modified by humans — putting the total at 38.5%. Perhaps more troubling: 31% of AI-generated songs were tagged by their submitters as not AI-made. SubmitHub founder Jason Grishkoff's position is straightforward: "the path forward for AI music is disclosure."

These numbers have jumped fast. The study landed the same week Apple announced its labeling plans, and the timing isn't coincidental — platforms are responding to a flood they can now quantify. For working musicians, the practical takeaway is that AI detection is getting reliable enough to enforce transparency whether creators opt in or not.

Source: NME, Euronews, MusicRadar, Mixmag

Alibaba Launches HappyShrimp 1.0 Music Generation Model

Alibaba released HappyShrimp 1.0 on August 17, a text-to-music model that generates melodies, arrangements, lyrics, and vocals from natural language prompts. Unlike tools that rely on tag-based input (genre + mood + tempo), HappyShrimp accepts conversational descriptions — tell it about an emotion, a story, or a scene and it produces a full track. It supports Chinese-style music, pop, rock, electronic, and jazz, and launched with a strategic partnership with Tahe Music Group.

This is Alibaba's first serious move into music generation, and it matters for two reasons: it's the first major model built with Chinese music as a first-class genre rather than an afterthought, and the Tahe partnership signals an attempt to build a licensed ecosystem from day one rather than train first and negotiate later. Currently in beta and focused on the Chinese market.

Source: Bloomberg, TechNode, GuruFocus

Suno Studio 2.0 Adds MIDI and Moves Toward DAW Territory

Suno shipped Studio 2.0 on August 13, adding MIDI recording, editing, and import to its browser-based workstation. You can plug in a MIDI controller, play on a typing keyboard, or import clips — and then use those MIDI clips as prompts for AI audio generation. The update also includes a wavetable synth, track automation, custom user-built effects plugins, a session-aware chat collaborator, and higher-fidelity stem separation.

MIDI was the most-requested feature since Studio launched last September, and it fundamentally changes what Suno is. The feature list now overlaps with entry-level DAWs: multitrack editing, MIDI, automation, effects, export. Available to Premier subscribers. Whether traditional DAW users will actually switch workflows remains to be seen, but Suno is clearly building for producers, not just prompt-typists.

Source: Digital Music News, Music Business Worldwide, Unite.AI

Pika Ships Four Audio Foundation Models

Pika released its first audio model family on August 13: Soundtrack (video-to-synchronized-audio), Music (prompt-to-song with lyrics and reference tracks), SFX (text-to-sound-effects), and Speech (a 3B flow-matching transformer producing 48 kHz voice from seconds of reference audio). They're positioning on price — up to 20x cheaper than comparable models — and speed, with all four available through the Pika API.

The interesting play here is comprehensiveness. Most audio AI companies specialize in one slice. Pika is shipping four models at once, covering the full post-production audio stack. If you're building video pipelines that need synchronized sound, this is worth benchmarking against ElevenLabs and ByteDance's Seed Audio. The aggressive pricing suggests they're buying market share.

Source: Pika Blog, The Rundown AI

ElevenLabs Music v2 Adds Finetunes and Audio Reference

ElevenLabs expanded Music v2 this month with two capabilities aimed at professional users: audio reference steering, which lets you feed a ~30-second clip to guide a generation's sonic direction, and Music Finetunes, which trains a custom style model on 5-10 minutes of your own audio. The API (v2.59.0) now exposes full CRUD methods for managing finetunes programmatically, and Music v2 API pricing was cut by up to 50%.

Audio reference is useful but incremental. Finetunes are the real story — if you have a distinctive production style or sonic brand, you can now bake it into the model rather than trying to describe it in a prompt. This is the kind of feature that separates "toy" from "tool" for working producers. Available on all paid plans.

Source: ElevenLabs Blog, Jack Righteous
Tool of the week
Suno Studio 2.0

Studio 2.0 is the update that makes Suno worth evaluating as an actual production environment rather than a novelty. MIDI support means you can sketch a melody on a keyboard and use it as a generation prompt — bridging the gap between human musical input and AI output in a way that text prompts alone never could. The wavetable synth and custom effects plugins push it further into DAW territory.

The practical workflow is: play a MIDI idea, generate variations, edit on the timeline, add automation, export stems. It's not replacing Ableton or Logic for serious production work, but for rapid ideation and demo creation, it's now a credible standalone tool. The chat collaborator — which understands your session context and can suggest edits — is surprisingly useful for getting unstuck.

Free tier available, Premier ($24/mo) for Studio access · Browser-based · suno.com/blog/studio-2
Worth watching
The disclosure infrastructure is arriving all at once

In the span of two weeks: Apple announced visible AI labels for music, Suno committed to audio watermarking and download limits (effective September 3), the EU AI Act's transparency obligations became enforceable, and SubmitHub published data showing a third of AI-generated tracks are being mislabeled as human-made. These aren't isolated moves — they're the pieces of a disclosure stack that's assembling itself in real time.

The practical implication for anyone making or distributing AI-assisted music: the window for ambiguity is closing. Detection tools are improving, platforms are adding labels, and regulations now carry penalties. If you use AI in your workflow — even for assist-level tasks like stem separation or mastering — start documenting what's AI and what's human now. The distinction between "generated" and "assisted" is about to become a legal and commercial boundary, not just a philosophical one.