Suno locked in its second major label deal with BMG as it prepares to retire its current models, Spotify announced AI Persona labels that will exclude AI-generated artists from recommendations, and the GEMA ruling in Munich sent a chill through every AI music company operating in Europe.
Suno signs BMG licensing deal, signals shift to fully licensed models
Suno and BMG announced a global strategic alliance on August 12, covering BMG's full recorded and publishing repertoire. The deal requires artist and songwriter opt-in for training use, compensates rights holders, and retroactively settles Suno's prior use of BMG catalog. This is Suno's second major label agreement after the Warner Music Group deal last November.
More significant than the deal itself: Suno confirmed it is retiring its current generation of models and preparing a new model trained entirely on licensed data. The company is effectively admitting its existing models can't survive legal scrutiny. If you've built workflows around Suno's current output quality, expect a reset. The licensed model will almost certainly sound different — whether that's better or worse remains to be seen.
Spotify will label AI-generated artists and cut them from recommendations
Spotify announced on August 11 that it will tag AI-generated artist profiles with an "AI Persona" badge, with a secondary "Likely AI Persona" label for accounts identified through the platform's own review process. Self-disclosure opened immediately through Spotify for Artists; labels start appearing in mid-September. The badge will show on artist profiles, search results, and playlist track rows.
The real teeth: AI Persona accounts will be excluded from editorial playlists and algorithmic recommendations — Discover Weekly, Radio, all of it — by default. Listeners who actively follow an AI artist will still see their music, but the passive discovery pipeline is shut off. This is Spotify's clearest move yet to protect human artists from AI-generated catalog flooding. If you're distributing AI-generated music on Spotify, your reach just got dramatically narrower.
Munich court hands GEMA a landmark win against Suno
The Munich Regional Court ruled that Suno violated German and American copyright law by training its AI models on music represented by GEMA, the German performing rights society. The court found that copyrighted musical works were "reproducibly contained" in Suno's models, which were stored on servers in Germany. Suno must disclose its revenue from infringing use and compensate GEMA members — damages are still being calculated.
The ruling's sharpest edge is jurisdictional: the court asserted authority over a US-based AI company by arguing that outputs served to EU users make the models subject to EU law. That logic applies to every AI music company serving European listeners. Suno said it disagrees with the findings and is evaluating an appeal, but the precedent is set. If you operate an AI audio service with European users, German courts now consider your training data their business.
EU AI Act transparency rules are now enforceable — audio included
The EU AI Act's Article 50 transparency obligations took effect on August 2. AI systems generating synthetic audio must now embed machine-readable markings at generation time, and AI-generated audio published where it could be mistaken for human-made content must carry a label. Providers have until December 2 to fully comply, but the obligations are live now and apply globally — any provider whose output reaches EU users is in scope, regardless of where the company is based.
For audio practitioners: if you're shipping AI-generated voice, music, or sound effects into the EU market, you need a watermarking and disclosure pipeline. Content created before August 2 is grandfathered, but anything generated from here on needs metadata baked in. The enforcement framework is still being tested, but the legal liability is real.
Suno caps downloads to 20 per month ahead of model transition
Starting September 3, Suno will impose strict download limits: 7 lifetime downloads for free users, 20 per month for Pro ($8), and 60 per month for Premier ($24). The limits apply retroactively to all songs on the platform, including tracks created before the cap takes effect. If you have a library of Suno-generated tracks you haven't exported yet, you have until September 3 to download them without restriction.
The timing isn't subtle — this lands alongside the BMG deal and the model transition. Suno is simultaneously restricting its old output while building its licensed replacement. The practical effect is that Suno becomes more of a walled-garden creative tool and less of a music generation API. For anyone who has been using Suno at volume for content production, this changes the unit economics significantly.
ElevenLabs opened its Dubbing v2 model to API access on August 6, ending the months-long restriction to its own ElevenCreative and ElevenProductions apps. The API translates audio and video into 90+ languages while preserving speaker voice, tone, pacing, and emotional delivery. It conditions directly on the original performance rather than generating flat TTS, which makes a noticeable difference on expressive material.
The project-based API exposes source transcripts and translations as editable JSON — you can modify individual segments and regenerate only the regions that changed, which makes it practical for iterative localization workflows. If you're working on multilingual content, game dialogue, or podcast distribution, this is the most capable dubbing API currently available. The main limitation: at $2.20 per minute of source media, it's expensive at scale.
With BMG signed and Warner already on board, Suno is now openly building toward a model trained exclusively on licensed data — and retiring everything that came before. This isn't a version bump. It's a ground-up retrain with a fundamentally different dataset. The output characteristics will change. The creative range may narrow. The legal exposure drops dramatically, but so does the breadth of musical knowledge the model draws from.
This matters beyond Suno. If the licensed model approach works commercially, it becomes the template: train only on what you've paid for, give rights holders revenue share and opt-in control, and let the courts sort out everyone who didn't. Sony Music is still in court with Suno, and the GEMA ruling just made the unlicensed path even riskier in Europe. The question for practitioners is whether licensed models can match the quality and range of models trained on the full corpus of recorded music. We'll find out later this year.