A Munich court ruled that Suno's AI training infringes copyright in Europe's first binding decision on the issue, the EU's synthetic audio labeling rules went live, major labels and IFPI moved to ban unlicensed AI from charts worldwide, and Google shipped Lyria 3.5 inside a vastly expanded Flow Music studio.

Munich court hands Suno its first defeat in Europe

On July 31, the Munich Regional Court ruled that Suno infringed copyrights by training its v3.5 and v4 models on GEMA-represented music without a license. The court held that storing copyrighted songs inside the model violates the reproduction right, and serving AI-generated outputs to users violates the making-available right. Suno must cease reproducing the protected works, disclose revenues linked to the infringement, and pay damages yet to be quantified. The case centered on six tracks including "Forever Young," "Daddy Cool," and "Mambo No. 5." Suno said it disagreed and would evaluate options including an appeal.

This is Europe's first binding court decision finding that an AI music generator infringed copyright. The signal is clear: EU courts are not buying the "training is transformative use" defense that US courts are still weighing. If you build or use tools that train on copyrighted audio, the licensing question just stopped being theoretical in Europe.

Source: Variety, Music Ally, Music Week, Music Business Worldwide, Music Times

Labels and IFPI launch global framework to keep unlicensed AI off charts

On July 29, a coalition including Sony, UMG, Warner, HYBE, Believe, and BMG proposed a six-point framework for AI music chart eligibility. A track qualifies only if it uses a properly licensed AI service, is substantially human-made, discloses AI usage, and shows no signs of streaming fraud. The framework introduces two voluntary tags — "AI-generated" for fully synthetic tracks and "AI-assisted" for mostly human work with some AI elements. IFPI immediately began rolling the principles out across its managed charts in Latin America, the Middle East, Africa, and Southeast Asia, with plans to expand to 20+ additional chart programs.

This creates a formal two-tier system. Licensed AI tools (Google, ElevenLabs, Stability AI) produce chart-eligible tracks; unlicensed ones don't. Billboard and Rolling Stone are already using Deezer's AI audio detection technology to flag synthetic tracks. No AI-generated song has hit the Hot 100, but several have appeared on niche Billboard charts. If you're releasing AI-assisted music, the disclosure and licensing provenance of your tools now directly affect distribution.

Source: IFPI, Billboard, Variety, Hollywood Reporter, Digital Music News

Google ships Lyria 3.5 and turns Flow Music into an all-in-one studio

Google DeepMind dropped Lyria 3.5 into Flow Music on July 29, replacing Lyria 3 Pro with richer melodic structures, better lyric prompt adherence, and vocals with more expression and cleaner pronunciation. This is the third significant Lyria release in five months. Creative controls now cover tempo, vocals, drums, bass, and track length. Google trained Lyria exclusively on licensed material — a distinction worth noting given what happened in Munich two days later.

The bigger move is what happened to Flow Music Spaces. Google added song generation and editing (covers, section replacement, track extension), stem splitting, image creation, lyrics generation, and video generation — all in one workspace. Spaces is positioned as "vibe coding for music," letting developers build custom browser-based audio plugins directly on the Lyria API. It's the most ambitious attempt yet at a single-platform AI music production pipeline. Free tier available; paid plans run $8–$64/month.

Source: Google Blog, Android Central, Digital Trends, Music Ally, Music Business Worldwide

EU AI Act Article 50 goes live — synthetic audio must carry machine-readable marks

As of August 2, Article 50 of the EU AI Act is enforceable. Providers of generative AI systems that produce synthetic audio must mark their outputs in a machine-readable format so they can be detected as artificially generated. Deployers who publish AI-generated audio to the public — including deepfaked voices of real people — must disclose its artificial nature to the audience. Penalties reach €15 million or 3% of global turnover. Existing providers get until December 2 to comply; anything launching after today must be compliant from day one.

For practitioners: this means every AI music and voice tool serving the EU market needs watermarking baked in. The requirement is machine-readable marking, not a visible badge — listeners won't see a label, but detection tools will find one. If you ship AI-generated audio commercially in Europe, verify whether your generation tool is Article 50 compliant, because if it isn't, you're carrying the liability.

Source: Production Expert, SSL.com, Truescreen, Bratby Law

Spotify's AI slop problem spawns grassroots tracking databases

404 Media reported on July 28 that Spotify still won't label AI-generated music, so third-party databases have stepped in. SoullessMusic.com catalogs "AI artists hiding on Spotify" while SlopTracker.org monitors AI content in Spotify-curated playlists. The scale is grim: an estimated 44% of all new music uploaded to Spotify is now AI-generated, and the company says it has removed over 75 million synthetic tracks — yet the flood continues faster than cleanup. Meanwhile, Tidal quietly started demonetizing fully AI-generated tracks on July 15, becoming the first major platform to cut royalties rather than just label.

The contrast matters. Tidal acts, Deezer detects and reports (90,000 AI tracks per day, over 50% of uploads), and Spotify does neither. For working musicians, the practical impact is that AI tracks compete for the same algorithmic recommendation slots and playlist placements on the platform with the largest audience.

Source: 404 Media, Digital Music News, Futurism

Fish Audio raises $52M to challenge ElevenLabs in voice AI

Fish Audio (Hanabi AI Inc.) announced a $52 million seed round on July 28, led by Coreline Ventures and Capital Today. The company, founded by former NVIDIA researcher Shijia Liao, went from zero to $21 million in annual recurring revenue and 8+ million users in its first year. Its open-source project Fish Speech has 31,000+ GitHub stars. The round positions Fish Audio as a direct challenger to ElevenLabs, which raised $500M at an $11B valuation earlier this year.

If you work with voice AI — narration, voice cloning, real-time TTS — Fish Audio is worth evaluating. The open-source angle means you can self-host and fine-tune, which matters for latency-sensitive applications and projects where you don't want audio leaving your infrastructure. At this funding level, expect aggressive feature development and pricing competition with ElevenLabs.

Source: TechCrunch, SiliconAngle, PR Newswire

Musicians' union expands AI licensing lawsuit against UMG and Warner

On July 27, the American Federation of Musicians filed an amended complaint against Universal Music and Warner Music, adding Warner Records and Atlantic Recording Corp as named defendants. The AFM alleges the labels licensed their members' recordings to Suno and Udio as part of settlement deals without compensating the session musicians who performed on those tracks. At the heart of the case is Article 21 of the Sound Recording Labor Agreement — the "new use" provision — which requires labels to pay musicians when recordings are put to purposes not originally covered. The AFM argues that licensing recordings to train generative AI models qualifies.

This case cuts differently from the copyright suits. Even if labels license their catalogs to AI companies lawfully, the musicians who played on those recordings may still be owed payment. If you're a session musician whose work appeared on major-label releases, this lawsuit's outcome directly affects whether you see any money from AI training deals.

Source: Digital Music News, Music Business Worldwide, RightsTech Project
Tool of the week
SoundBoost

SoundBoost replaces traditional mastering knobs and sliders with natural-language text prompts. Describe what you want — "more analog warmth without crushing the dynamics" or "decrease drums by 2dB, wider stereo field but keep the lows in mono" — and get a streaming-ready master in under a minute. Its Mastering Engine v3.5 tailors adjustments to genre-specific needs without presets, and you can refine iteratively through follow-up prompts. It also includes stem separation and six AI engineer profiles. Over 426,000 masterings completed so far.

The natural-language interface is the draw here. For producers who know what they want but find mastering plugins intimidating, this lowers the barrier significantly. At the unlimited tier, it's roughly 7x cheaper than LANDR per master and includes stem splitting that other services charge $5–$10 per track for separately. The quality won't replace a dedicated mastering engineer for a release you care deeply about, but for demos, content music, and quick iterations it's genuinely useful.

From $4/mo; unlimited from $16/mo · Web, iOS, Android · soundboost.ai
Worth watching
The licensed-versus-unlicensed divide just became structural

This was the week where three independent forces — courts, regulators, and industry — all moved against uncontrolled AI music at once. Munich said training needs a license. The EU said synthetic audio needs a watermark. The labels and IFPI said unlicensed AI doesn't belong on charts. Each alone is significant; together they define a new operating environment that structurally advantages tools built on licensed data.

Google, ElevenLabs, and Stability AI have all positioned on the licensed side. Suno now faces rulings in both the US and EU. Udio pivoted to a fully licensed remixing platform (Starstruck). For anyone making sound and music professionally, the tool you choose is no longer just a creative decision — it's a legal and commercial one. Expect distributors to start requiring proof of licensed-model provenance. The window for "move fast and sort out rights later" is closing.