Google reportedly uses a vast library of YouTube videos to train AI models such as Gemini and Veo 3.
Last month, Google launched its latest AI video model, Veo 3, positioning it as one of the most advanced AI video generators on the market, placing its audiences with incredibly impressive synthetic videos generated from it.
Before you ask: Yes, it's all AI here. Both the video and the sound come from a single text prompt #veo3 by @GoogleDeepMind Anyone who cooks models should let him cook! congratulations @totemko And then there's a team of Google I/O Live Stream and a new VEO site! pic.twitter.com/sxzuvfu49s
– lászlógaál (@laszlogaal_) May 21, 2025
According to the report of CNBCGoogle is leveraging YouTube's 20 billion videos library to train AI models. The news outlet cited sources that were not permitted to publicly talk about the issue.
Google later confirmed it CNBC It says it uses YouTube videos to train AI, but relies solely on a subset of content, and is in compliance with certain agreements with creators and media partners.
Google Veo 3 Realism just broke the internet yesterday.
This is 100% AI
10 Wild Examples:
1. Street interviews that never happened pic.twitter.com/qdxzvhoo3g
– Min Choi (@minchoi) May 22, 2025
“We've always used YouTube content to improve our products, and this hasn't changed with the advent of AI,” a YouTube spokesperson told the news outlet in a statement. “We also recognize the need for guardrails, which is why we invested in strong protections that enable creators to protect their images and portraits in the age of AI.
YouTube creators say they didn't know
YouTube has previously said it has disclosed this practice, but many creators and media organizations are unaware that Google uses video libraries to train AI models. CNBC. The creators interviewed by the outlet say they were not notified or consulted and were caught off guard by the revelation.
When uploading videos to the platform, users agree that YouTube has a wide range of licenses to the content. However, YouTube allows creators to opt out of third-party AI training by companies like Amazon, Apple and Nvidia, but reportedly cannot prevent Google from using videos to train their own models.
YouTube has not disclosed how many or what number of the 20 billion videos used for AI training. However, experts quote it CNBC Note that even if you only use 1% of the library, you will still be able to provide around 2.3 billion minutes of footage. Note that this provides more than 40 times the training data that is believed to be used in some rival AI models.
Digital rights advocates argue that years of work by YouTube creators have been used to develop AI systems without consent or compensation. One example is Vermillio. This has developed a tool called Trace ID to detect similarity between AI-generated videos and original content. In some cases, this tool found more than 90% of matches.
“We've discovered fake versions of more and more creators running through the platform. New tools like VEO3 only accelerate trends,” says Dan Neely, who helps protect personal similarities from being misused, helps promote secure licensing of certified content. CNBC.
The latest news comes after Disney and Universal filed a joint lawsuit against the generation AI company Midjourney, accusing widespread copyright infringement.
