Google Leverages YouTube Content to Train Its AI Video Generator, Raising Intellectual Property Concerns

Google is leveraging content from YouTube to train its artificial intelligence models, including Gemini and the video and audio generator Veo 3, as reported by CNBC citing sources. According to these sources, the company is tapping into a massive library of 20 billion videos on the platform.

Google confirmed to CNBC that it employs YouTube videos for training its AI models, while also stating that it adheres to specific agreements with content creators and media organizations.

«We have always utilized YouTube content to enhance our products, and this approach remains unchanged with the advent of AI. We understand the need for protective measures, so we have invested in robust security tools that enable creators to safeguard their image in the AI era, and we intend to continue doing so,» stated a YouTube representative.

Experts express concerns that such use of YouTube content could trigger an intellectual property crisis for creators and media companies, who do not fully comprehend that Google is training its AI models using their videos.

YouTube has not disclosed how many of the 20 billion videos on its platform or which specific ones are being used for AI training. However, given the vast scale of the video hosting service, just 1% of its catalog would amount to 2.3 billion minutes of content — a volume that far exceeds the amount of data used by Google’s competitors for training their models, experts note.

Last fall, Google warned about the potential use of YouTube content «to improve product experience,» including through machine learning and AI applications. Users uploading content to the platform effectively agree to the possibility of their videos being used for AI model training without the option to opt-out.

CNBC interviewed several prominent content creators and copyright experts, none of whom were aware that their content could be utilized for training the corporation’s AI models.

Google introduced Veo 3 in May of this year, describing it as one of the most advanced AI tools for video generation available. During the unveiling, Google showcased cinematic-quality videos produced by the AI.

On average, 20 million videos are uploaded to YouTube daily. Many creators have expressed concerns that their work might be used to train AI models that could potentially compete with or replace them. In reality, by uploading a video to YouTube, users consent to the possibility of their content being used for training neural networks.

However, video creators do have the option to seek compensation from Google if they encounter and can demonstrate copyright infringement.