Suno's data breach exposed 55 million users and its training sources, Sony filed a second lawsuit naming 30,000 recordings against Udio, and Deezer reported that AI-generated tracks now exceed half of all daily uploads.
Suno breach exposes 55 million users — and exactly where its training data came from
A November 2025 breach of AI music generator Suno surfaced publicly this week after Have I Been Pwned added 55.3 million affected accounts on July 21. The attacker used stolen employee credentials to access internal repositories and databases, exposing names, email addresses, phone numbers, physical addresses, and partial Stripe payment data. Suno did not notify users directly, saying individual notices weren't required under applicable law.
The leaked source code is the real story for the industry. It allegedly documents scraping pipelines pulling training data from YouTube Music, Deezer, Genius, Pond5, Jamendo, Freesound, the International Music Score Library Project, and podcasts via RSS feeds. This lands while Suno faces a fair-use hearing in Massachusetts and a GEMA verdict in Munich — the kind of evidence rights holders have been trying to surface through discovery. If you use Suno, check Have I Been Pwned and rotate any reused passwords.
Sony files second lawsuit against Udio, names 30,000 recordings
Sony Music filed a new copyright suit against Udio on July 20, alleging the AI music platform copied 30,117 sound recordings for model training without authorization. The figure dwarfs the 333 tracks in the original 2024 complaint. Sony says the newly listed works are only a subset of hundreds of thousands it identified through discovery. After Judge Hellerstein denied folding these claims into the existing case in June — ruling it would prejudice Udio — Sony filed separately, as the judge suggested.
Sony is now the only major label still litigating against Udio. Universal and Warner dropped their claims after signing licensing deals with the startup in late 2025. The case sharpens the central question: is training a commercial generative model on copyrighted recordings fair use or mass infringement? If you're building on Udio's platform, the legal risk profile just changed — Sony is reportedly seeking $4.5 billion.
AI-generated music now exceeds 50% of daily uploads on Deezer
Deezer reported on July 21 that AI-generated tracks hit 90,000 per day at peak in June, crossing 50% of all new music deliveries for the first time. That's up from 75,000 daily tracks (44%) when the platform last disclosed numbers in April. The trajectory is steep — the figure was around 10,000 per day when tracking began in early 2025.
The counterpoint matters: AI music accounts for just 3% of actual listening time on Deezer. The platform excludes AI-generated content from algorithmic and editorial recommendations, and it will now systematically remove AI tracks linked to streaming fraud plus synthetic songs with zero plays for six months. For working musicians, this is the data point to cite when explaining why the flood of AI content isn't (yet) eating real listening share — but the upload volume itself is reshaping how distributors and platforms allocate resources.
RIAA, IFPI, and SAG-AFTRA launch AI labeling system for recordings
Eight industry organizations — IFPI, RIAA, A2IM, WIN, IMPALA, the Recording Academy, SAG-AFTRA, and the Human Artistry Campaign — launched a voluntary labeling framework on July 15 that tags recordings as either "AI-Generated" (entirely AI-produced, including AI lead vocals or primary performances) or "AI-Assisted" (substantially human-created with AI contributing some expressive elements). The labels sit in track metadata and are designed for display on participating digital music services.
The system is voluntary, so adoption depends on platform buy-in. It also doesn't cover AI use in lyrics, composition, music videos, or cover art — just sound recordings. For producers using AI as a production tool (stem separation, mastering, sound design), your work likely falls outside both labels. If AI generated your lead vocal or primary instrumental, that's "AI-Generated" regardless of how much human editing followed. Worth understanding where the lines are, even if enforcement is soft.
Suno ships iMessage integration and rebuilt lyrics editor
Suno released an iMessage keyboard extension on July 15 that lets iPhone users generate and send songs without leaving Messages — including male/female voice selection, sound effects, and a new mashup feature. It's a consumer play, but it signals where Suno sees distribution heading: embedded in messaging rather than siloed in a separate app. The same week, Suno's CPO confirmed the company is exploring a developer API with a curated partner group.
The rebuilt lyrics editor (July 9) is more relevant if you actually use Suno for production. It adds "Lyricist" profiles that lock in your writing style across sessions, natural-language editing ("make this line funnier"), rhyme and reference suggestions on highlighted words, and full-screen mode with autosave. These are the kinds of workflow features that separate a toy from a tool — the Lyricist profile in particular addresses the biggest complaint about AI-assisted writing: inconsistency across sessions.
Speechify's Simba 3.2 hit #1 on the independent Artificial Analysis TTS leaderboard in early July, outranking flagship models from ElevenLabs, OpenAI, Cartesia, and Google DeepMind. It also placed joint second on Voice Arena's blind listener rankings. The model is streaming-native with low time-to-first-byte, designed for real-time voice applications — agents, IVR, live readers, phone systems.
What makes it worth attention for audio practitioners: fine-grained emotional control, SSML prosody support, instant voice cloning from short clips, and native-quality speech across 30+ locales with automatic mixed-language handling. At $10 per million characters on the entry tier (dropping to $6 at scale), it's the cheapest model in the leaderboard's top ten. If you've been paying ElevenLabs rates for voice work, this is the benchmark to test against.
Germany's Munich Regional Court delivers its ruling in GEMA v. Suno on July 31 — potentially the first major European court decision on whether AI companies must license the music they train on. The case centers on six specific works including Alphaville's "Forever Young," Boney M's "Rasputin," and Lou Bega's "Mambo No. 5." Judge Schwager's 42nd Civil Chamber will rule at 9am in the Justizpalast, courtroom 270.
The timing is deliberate pressure. The Massachusetts fair-use hearing in the Sony v. Suno case is running in parallel, and Suno's breach just exposed its training pipeline. A ruling against Suno in Munich would force licensing across the EU under existing copyright frameworks — and hand ammunition to the US plaintiffs. A ruling for Suno would complicate GEMA's leverage and embolden AI companies operating in Europe. Either way, this sets precedent that every AI audio company will need to respond to.