Jun 20, 2025 11:55:00

It turns out that Google is using videos uploaded to YouTube to train its AI, but the creators don't know much about it

When Google rolls out its chat AI

Gemini and its video generation model Veo 3, which can create videos with resolutions up to 4K, it is using a huge library of videos uploaded to YouTube for training, but it has become clear that many creators are not aware of this fact. YouTube has stated its intention to protect the rights of video creators, but some experts point out that using posted videos to train AI could lead to a crisis of intellectual property.

Google is using YouTube videos to train its Gemini, Veo 3 AI models
https://www.cnbc.com/2025/06/19/google-youtube-ai-training-veo-3.html

Google has been proactive in its efforts with AI tools, such as announcing plans to integrate the video generation AI 'Veo 3' into YouTube Shorts and developing an AI music generation tool that imitates artists' music. On the other hand, YouTube is also actively working to protect creators' rights from being infringed by AI, such as by allowing creators to choose whether to allow third parties to train AI on their videos and by developing tools to label 'realistic AI-generated content'.

In September 2024, Google announced an AI detection tool to help creators prevent their likeness, including their face and voice, from being copied and used in other videos. In a blog post announcing the tool, YouTube said, 'We will continue to take steps to ensure compliance with our Terms of Service, including continued investments in systems to detect and block unauthorized access, including scraping of YouTube content,' pointing out that scraping content on the platform is a violation of the Terms of Service.

YouTube launches AI detection tool to prevent copying of creators' likenesses, including faces and voices, and also severely criticizes scraping for training AI - GIGAZINE

While YouTube prohibits scraping of its content, it also indicates that YouTube content 'may be used to improve product experiences, including through machine learning and AI applications.' Some experts point out that 'Google's use of its video library to train AI models is not widely understood by creators and media organizations,' raising concerns that this could lead to an intellectual property crisis for creators and media companies.

The problem, experts say, is that YouTube hasn't disclosed how many or which of the 20 billion videos uploaded to the platform are being used to train AI. In fact, when CNBC spoke to several top creators and intellectual property experts, none of them were aware that their content could be used to train Google's AI models, and none of them were informed by YouTube.

In an interview with CNBC, Google acknowledged that it uses YouTube videos to train its AI, but claimed that it only uses portions of the videos for training and complies with certain contracts with creators and media companies. A YouTube spokesperson said, 'We've always used YouTube content to improve our products, and this will not change with the advent of AI. We also recognize the need for guardrails, which is why we've invested in strong protections to help creators protect their image and portrait rights, even in the age of AI. We intend to continue this effort.'

'YouTube is potentially taking videos that creators have invested time, effort, and ideas into to train its AI,' said Luke Arigoni, president of the group Protecting Creators' Digital Identities. 'Veo 3 creates synthetic versions of these creators, or inferior copies. That's not necessarily fair to creators. If YouTube were to reveal what videos it trained on, it would have an impact on its relationships with creators.'

Google has included an '

indemnification clause ' in its AI-generated products, including Veo, which stipulates that if a user faces a copyright infringement lawsuit for AI-generated content, Google will be legally liable and cover the associated costs. In addition, in December 2024, 'Third-party training' will be added to the settings for posting on YouTube, allowing users to choose whether to allow others to train AI models on their content. Also in December, YouTube partnered with the Creative Artists Agency to announce a tool that allows creators to identify and manage content that includes their likeness and request the removal of videos, but some experts have questioned the reliability of this tool.

According to a YouTube press release , more than 20 million videos are uploaded to YouTube every day, and there will be more than 20 billion videos on YouTube as of April 2025. Experts say that the platform is so massive that even if just 1% of the videos were used for training, it would be more than 40 times the training data used by competing AI models.