Google is utilising the extensive library of videos uploaded on its platform, YouTube, to train Veo 3, the company’s latest AI-enabled video generation tool. Google confirmed this to CNBC, stating that it uses only a subset and not the entirety of YouTube videos to train its AI models.
The report added that several creators are “concerned that they may be unknowingly” helping train the model, which could “eventually compete with/replace them”, and multiple leading creators were neither aware nor informed by YouTube that their content could be used to train the Veo models.
While YouTube allows creators to opt out of providing their content for training to third-party AI companies, they cannot prevent Google from doing so. Third-party companies, including Apple, Anthropic, Amazon, Meta, and Microsoft, can use YouTube’s data to train their generative AI models.
Veo 3 is the company’s third-generation AI video generation model and the latest one. It was announced last month at Google’s I/O 2025 and can generate eight-second videos.
Moreover, Veo 3 can add audio to the generated videos—a capability not present in several competing platforms, including OpenAI’s Sora. It was initially available in the United States alone, but the company has since expanded access to over 70 countries.
Users on Google Gemini’s Pro plan, which costs $20 a month, get limited access to the model. However, the higher-end Gemini Ultra plan, which is $249 per month, provides the highest usage for Veo 3.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.