Google opened its Lyria 3.5 music model to all Gemini users and developers via API, SOCAN became the third collecting society to sue Suno over training on copyrighted music, and all three major labels simultaneously invested in Stability AI's $76M round while continuing to sue other AI companies.
Google Opens Lyria 3.5 to All Gemini Users and Developers via API
On September 4, Google made Lyria 3.5 — its best music generation model — available globally through the Gemini app, Gemini API, Google AI Studio, and Google Vids. Previously confined to the Google Flow Music app, the model now generates full songs with verses, choruses, and bridges in 44.1 kHz stereo, up to three minutes long. It includes an image-to-music input mode: upload a photo and it produces a matching track. Developers get a public preview API for text- and image-prompted music generation.
If you build products that need music, this is the first major-company music model available through a general-purpose API alongside text and vision capabilities. The quality is a clear step up from Lyria 2 — richer melodic structures, better vocal pronunciation, and tighter prompt adherence. Free tier gets limited generations; paid Gemini plans get more. Not open source.
SOCAN Sues Suno — Third Collecting Society to File
On September 2, Canada's SOCAN filed suit against Suno in Federal Court of Canada, becoming the third collecting society to sue after Germany's GEMA (which won its case July 31) and Denmark's Koda. The complaint cites 150 Suno outputs that allegedly replicate copyrighted works, including songs by Avril Lavigne, Joni Mitchell, Celine Dion, and The Tragically Hip. SOCAN alleges Suno trained on music available online without obtaining licenses.
The GEMA win in Munich set a precedent that's clearly emboldening other societies. Suno is now fighting copyright claims on three fronts — Germany, Canada, and the US (where the major label suits continue). Each new filing narrows the viable jurisdictions for unlicensed training. For anyone using Suno commercially, the legal risk is now compounding, not receding.
Sony Music and Warner Chappell Sue Anthropic Over Lyrics Training
Sony Music Publishing and Warner Chappell Music filed a multi-billion-dollar copyright lawsuit against Anthropic in California federal court on August 29, alleging it pirated hundreds of song lyrics and compositions — from the Beatles to Taylor Swift — to train its Claude AI. The complaint describes "one of the largest and most blatant ongoing thefts of intellectual property in history" and seeks up to $150,000 per infringed work.
This isn't about music generation tools — it's about general-purpose AI models reproducing lyrics. But it matters to the audio world because the music industry is now opening legal fronts against AI companies well beyond the Sunos and Udios. If successful, it could reshape how any AI model handles musical content, including the metadata and lyrics that audio professionals work with daily.
All Three Major Labels Invest in Stability AI's $76M Round
Stability AI closed a $76 million Series B with a historic investor list: Universal Music Group, Sony Music Group, and Warner Music Group all took equity in the same AI company for the first time, alongside Electronic Arts, AMD Ventures, Coatue, and Lightspeed. Total funding now stands at $232 million. This follows Stable Audio 3.0's launch in May (trained exclusively on licensed AudioSparx and Freesound recordings) and its DAW plugin release on August 18.
The signal is unmistakable: the majors have shifted from pure defense to direct participation in AI audio infrastructure — but only when training data is fully licensed. Stability AI is now the clearest example of the "licensed-data-first" model the industry will accept. Worth noting the criticism, though: as TechTimes put it, "Music labels get equity; artists still get nothing."
Suno Tightens Download Limits and Adds Audio Watermarking
On September 3, Suno rolled out revised terms of service alongside new download limits and audio watermarking. Pro subscribers now get 20 download credits per month, Premier gets 60, and Suno Studio workflows are exempt. All outputs now carry audio watermarks and fingerprinting — a move aimed at mass distribution use cases and likely a nod toward the legal pressure the company faces.
If you've been using Suno for volume production work — game audio, background music libraries, content — the new download limits change the economics. The watermarking is also significant: it means Suno outputs are now traceable back to the platform, which could matter if your client or distributor starts checking provenance. The community is split, but this feels like Suno preparing for a licensed-model future.
Spotify AI Persona Badges Go Live Mid-September
Spotify's "AI Persona" badge system, announced August 11, begins rolling out to profiles in mid-September. AI-generated artist identities get a visible badge and are excluded from editorial, algorithmic, and personalized recommendations by default. Users can report unlabeled AI profiles, and artists can appeal mislabeling. Meanwhile, Apple Music confirmed its "Made With AI" disclosure labels will become mandatory, covering AI involvement in recordings, compositions, artwork, and music video.
The two biggest streaming platforms are now formally separating AI-generated music from human-made music in their discovery systems. For practitioners distributing AI-assisted work, the key distinction is between "AI-assisted" (human artist, AI tools used in production) and "AI Persona" (no real human artist). The former is fine; the latter gets downranked. If you ship music made with AI tools, make sure your distributor's metadata is clean.
Released August 27, Instrument X is a neural acoustic modeling platform from the makers of Synthesizer V. Instead of triggering pre-recorded samples, it generates orchestral instrument sound directly from notation using neural synthesis. It covers 15 instruments across strings, woodwinds, and brass, with up to 11 simultaneous modeled microphone feeds including close mics, a Decca tree array, and room ambience. The engine runs entirely locally — no GPU, no internet, no cloud dependency.
This matters because it eliminates the need for massive sample libraries that routinely consume hundreds of gigabytes of disk space, while offering continuous expressive control over articulations that sample-based approaches can only approximate. If you score for orchestral instruments and you're tired of managing terabytes of Kontakt libraries, this is worth evaluating seriously. The tradeoff: the sound is convincing but still identifiably synthesized on sustained solo passages. For ensemble and layered work, it's remarkably good.
This week crystallized a tension that's been building all year. The same three major labels — UMG, Sony, Warner — are simultaneously suing AI companies (Anthropic, Suno) for copyright infringement while investing equity in Stability AI. They've struck licensing deals with Udio and ElevenLabs while maintaining active litigation against others. The pattern is clear: licensed training gets investment and partnership; unlicensed training gets legal action. The majors aren't anti-AI — they're pro-control.
For audio professionals, this means the market is bifurcating. Tools built on licensed data (Stable Audio, Udio's licensed mode, ElevenLabs) are becoming safer bets for commercial work. Tools with unresolved copyright questions carry increasing legal and reputational risk. The "move fast and train on everything" era of AI music is ending. What replaces it will be more expensive, more constrained, and — if the licensing deals work as intended — will actually pay rights holders. Whether individual artists see any of that money remains the open question.