Google controls YouTube. Google controls Content ID. Google has 50 million music reference files with full metadata. So when Google trained Lyria 3 on YouTube uploads without asking permission, independent musicians noticed something that doesn’t look like an oversight.
On March 6, nine indie artists filed a 118-page lawsuit in the Northern District of Illinois accusing Google of mining YouTube’s catalog to train its AI music generator. The case makes familiar copyright claims, but adds something new: allegations that Google extracted biometric voiceprints from performers’ recordings without consent, violating Illinois’ Biometric Information Privacy Act (BIPA).
What Google’s Own Research Reveals
The complaint cites Google’s published papers describing how the company trained Lyria:
- Gathered approximately 50 million internet music videos
- Extracted 30-second audio clips
- Retained roughly 44 million clips after filtering
- A separate 2023 paper described training on 5 million clips totaling 280,000 hours of audio
Neither paper “mentions a license” or “discusses a consent mechanism,” the plaintiffs note.
“It has the technical infrastructure, financial resources and industry connections to clear rights before training,” the lawsuit states. “Google chose not to do so, not because licensing was impossible, but because copying was faster and cheaper.”
The Voiceprint Angle
What distinguishes this case from dozens of other AI copyright lawsuits is the BIPA claim. The plaintiffs allege Google:
- Extracted voiceprints from their recorded performances
- Collected and used those voiceprints as biometric identifiers
- Did so without obtaining informed written consent
- Failed to publish a public data retention policy
Under Illinois law, each violation can carry damages of $1,000 to $5,000 per instance. With millions of clips involved, the potential exposure is substantial.
The same artists previously filed similar BIPA claims against Suno and Udio in October 2025.
The Plaintiffs
The named artists include:
- Sam Kogon (New York singer/songwriter)
- Magnus Fiennes (Los Angeles composer)
- Michael Mell (Atlanta songwriter/producer)
- Attack the Sound (indie R&B group)
- Stan and James Burjek (folk rock duo)
- Directrix (Chicago band: Berk Ergoz, Hamza Jilani, Maatkara Wilson, Arjun Singh)
- David Woulard
Several previously sued Suno and Udio over training practices.
Google’s Position
Google has responded that it developed Lyria responsibly and in collaboration with the music community. The company claims it uses music that YouTube and Google “have the right to use” under “terms of service, partner agreements, and applicable law.”
But Google won’t disclose what those partner agreements actually cover—which labels, which terms, which rights holders.
The lawsuit counters that YouTube’s terms of service grant Google a license to host and distribute uploaded content, not to copy it into separate AI training datasets. Artists uploading music to YouTube agreed to let Google show it to listeners, not to let Google build a competing music generation tool.
The Content ID Problem
The complaint highlights what the plaintiffs call Google’s “vertically integrated syndicate”:
- Google runs YouTube, where artists upload music
- Google runs Content ID, the fingerprinting system that identifies copyrighted content
- Google built Lyria 3, using that same content
- Google distributes AI-generated music back through YouTube
“Google had every opportunity to develop this product legally,” the lawsuit states, given its “technical infrastructure, financial resources, and industry connections to clear rights before training.”
The plaintiffs specifically allege that Content ID—which artists rely on to protect their work—is being used to “launder” AI-generated tracks “of their proper copyright identifications.”
Lyria 3 Pro: Licensed or Not?
Two weeks after the lawsuit was filed, Google launched Lyria 3 Pro, upgrading from 30-second clips to full 3-minute tracks. Google positions it as the licensed alternative for businesses that cannot afford legal exposure from competitors like Suno and Udio.
All outputs include SynthID watermarks. If users include an artist’s name in prompts, the model takes “broad inspiration” rather than mimicking specific voices.
But whether the underlying training data was actually licensed remains disputed. The lawsuit focuses on the original Lyria 3, not the Pro version.
Why This Case Matters
Three reasons this lawsuit stands out:
1. The BIPA experiment. If courts accept that extracting vocal characteristics from recordings constitutes “biometric information” under Illinois law, it opens a massive new liability category for AI companies. Per-violation statutory damages could dwarf copyright damages.
2. The platform leverage question. Google isn’t just an AI company accused of scraping—it’s the company that built the platform artists uploaded to, the fingerprinting system they trusted, and the distribution channel they depend on. The complaint frames this as a structural betrayal.
3. The DeepMind connection. The lawsuit notes that Udio was founded by four Google DeepMind researchers who left to start their own company, then settled with Universal and Warner. Google then released Lyria 3 and acquired ProducerAI. The plaintiffs characterize this as escalation despite known legal risks.
What Happens Next
The case is in early stages. Key questions:
- Will the court accept the biometric privacy theory?
- Can plaintiffs prove YouTube’s terms of service don’t authorize AI training?
- Will Google’s “licensed data” claims hold up under discovery?
Meanwhile, Suno and Udio have already settled with major labels and are launching licensed platforms. Google claims it was already operating legally. Independent artists aren’t convinced.
What You Can Do
If you’ve uploaded music to YouTube:
- Your content may have been used for AI training
- Consider reviewing YouTube’s current terms of service
- Document your upload history if you want to join potential class actions
If you’re in Illinois:
- The BIPA claims apply specifically to residents or those whose biometric data was processed in Illinois
- BIPA has strict requirements for consent and data retention policies
If you’re watching AI music development:
- This case could establish whether platform terms of service authorize AI training
- The voiceprint theory could reshape how companies approach vocal AI
- Settlement patterns from Suno/Udio suggest how this might resolve
The lawsuit joins a growing wave of AI copyright litigation. But by combining copyright claims with biometric privacy law, these indie musicians are testing whether there’s a faster, more painful path to accountability than traditional intellectual property cases.
Google has 50 million reference files. The question is whether “we had the right to use it” will survive scrutiny when artists never agreed to become training data.