Google is fighting a lawsuit from independent musicians who say it trained its Lyria music AI on songs they uploaded to YouTube, and the company’s defense boils down to a careful non-answer. According to The Verge AI, a group of independent artists is suing Google over its Lyria 3 model, claiming it illegally used their uploaded tracks as training data. Google has filed a motion to dismiss, and the language it chose says a lot about where this is headed.
What Google actually argued
In its filing, Google leaned on two arguments at once:
- You can’t prove it. The company called the lawsuit “based on the unsupported hypothesis that Google trained on their specific works.”
- Even if we did, we’re allowed. Google says that by uploading to YouTube, the musicians “granted YouTube, and Google… a broad license to use the uploaded content” under the Terms of Service, including the right to “reproduce, distribute, [and] prepare derivative works.”
That’s classic legal hedging, as The Verge AI puts it, you can’t prove we did it, and even if we did, the fine print covers us. When asked directly whether it uses YouTube videos to train Lyria 3, Google declined to comment.
Why the silence is strategic
What stands out here is that Google has already admitted the broader practice. It just won’t say the word “Lyria.”
The Verge AI points to a clear paper trail:
- In an April 2024 Bloomberg interview, YouTube CEO Neal Mohan said “some portion” of YouTube videos may be used internally to train models like Gemini.
- A later company blog post confirmed Google uses uploaded content “to improve the product experience… including through machine learning and AI applications.”
- Google told CNBC it was using YouTube uploads to train both Gemini and Veo.
So the company has confirmed YouTube data feeds its text model and its video model. Lyria, the music model, is the one admission it’s withholding. With active litigation, that gap isn’t an oversight. It’s plausible deniability, and right now Google has little to gain by going on record.
Why this matters
This is significant because it tests whether a platform’s Terms of Service can double as blanket consent for AI training. Most creators clicked “agree” years before generative music models existed. Google’s argument is that the same license they signed to host a video also covers training a system that could compete with their work.
For anyone who has uploaded music, audio, or video to YouTube, the practical takeaway is blunt: Google almost certainly treats your upload as fair game for model training, whether or not it names the specific model. The license language is broad enough to cover a lot.
For the wider industry, the case fits a pattern. Companies building generative models keep landing in court over where their training data came from, and they keep reaching for the same shields: lack of direct proof, and broad licensing terms users already accepted. How a judge treats Google’s ToS argument could shape how every platform-owned AI model defends itself.
What to watch next
A few things will tell us where this goes:
- Whether the motion to dismiss succeeds. If the ToS license argument holds up early, expect other platforms to copy it.
- Whether discovery forces specifics. If the case proceeds, Google may finally have to confirm what trained Lyria 3, on the record.
- Whether creators get new opt-out rights. Pressure like this is what pushes platforms to add training controls, the way some have already done for other models.
Google’s refusal to confirm the obvious isn’t confusion. It’s a calculated move to keep its options open while the lawsuit plays out. The musicians’ case may be the thing that forces a straight answer. You can find the full reporting at the original source on The Verge AI.