an exclusive report revealed that google is using its massive library of youtube videos to train artificial intelligence models, raising concerns over intellectual property rights and creator consent.
article contents:
- introduction
- the discovery
- scope of use
- creator reactions
- technical details
- legal implications
- industry impact
- conclusion
introduction
google is using its vast library of youtube videos to train its artificial intelligence models, including gemini and the veo 3 video and audio generator, according to what cnbc has learned. the company confirmed to cnbc that it relies on the youtube video vault to train its ai models, but google said it only uses a subset of its clips for training.
this use of youtube videos has the potential to spark an intellectual property crisis for content creators and media companies, experts said. the tech company is turning to its catalog of 20 billion youtube videos to train these new-age smart tools, according to a person not authorized to speak publicly about the matter.
content creator challenges thieves: data poisoning to protect creativity
the discovery
google confirmed to cnbc that it relies on the youtube video vault to train its artificial intelligence models, but the company said it only uses a subset of its clips for training and honors specific agreements with content creators and media companies.
a youtube spokesperson said in a statement: “we have always used youtube content to make our products better, and that hasn’t changed with the advent of ai. we also recognize the need for controls, which is why we’ve invested in robust protections that allow creators to protect their image and likeness in the age of ai—something we are committed to continuing.”
scope of use
while youtube says it previously shared this information, experts who spoke with cnbc said it is not widely understood by content creators and media organizations that google trains its ai models using its video library.
youtube did not mention how many of the 20 billion videos on its platform or which ones are used for ai training. but given the scale of the platform, training on just 1% of the catalog would amount to 2.3 billion minutes of content, which experts say is more than 40 times the training data used by competing ai models.
creator reactions
cnbc spoke with several leading creators and intellectual property professionals, none of whom were aware or informed by youtube that their content could be used to train google’s ai models. luke arregoni, ceo of lotti, a company that protects the digital identity of content creators, said: “they are likely taking data from a lot of creators who spent a lot of time, energy, and their own thought process into putting these clips together.”
he added: “it helps the veo 3 model make an artificial version, a weak imitation, of these creators. that is not necessarily fair to them.” the disclosure that youtube is training on user clips is notable after google announced veo 3 in may, one of the most advanced ai video generators on the market.
technical details
in its reveal, google showcased cinematic-grade video scenes, including one of an old man on a boat and another showing pixar-like animals talking to each other. the entire scenes, both visual and audio, were entirely ai-generated. according to youtube, an average of 20 million videos are uploaded to the platform daily by independent creators and nearly every major media company.
many creators say they are now concerned they may be unwittingly helping train a system that could eventually compete with or replace them. arregoni said: “it doesn’t hurt their competitive advantage at all to tell people what kind of clips they are training on and how many.” “the only thing it will truly affect is their relationship with creators.”
legal implications
even if veo 3’s final outputs do not directly replicate existing work, the generated content feeds commercial tools that can compete with the creators who made the training data possible, all without credit, consent, or compensation, experts said. when uploading a video to the platform, the user agrees that youtube has a broad license to the content.
the terms of service state: “by providing content to the service, you grant to youtube a worldwide, non-exclusive, royalty-free, sublicensable, and transferable license to use that content.” dan neely, ceo of vermilio, which helps individuals protect their likeness from misuse and also facilitates secure licensing of approved content, said: “we have seen an increasing number of creators discovering fake copies of themselves spreading across platforms—new tools like veo 3 will only accelerate this trend.”
industry impact
neely’s company is challenging ai platforms to generate content that allegedly violates the intellectual property of its clients, both individuals and corporations. neely says that although youtube has the right to use this content, many creators posting on the platform are unaware their clips are being used to train generative ai video software.
vermilio uses a proprietary tool called trace id to assess whether an ai-generated video has significant overlap with a human-created video. trace id gives scores on a scale from zero to 100. any score over 10 for a video with audio is considered meaningful, neely said.
in one example cited by neely, a video from youtube creator brody moss closely matched content generated by veo 3. trace id gave a score of 71 for the original video, with the audio track alone scoring over 90.
conclusion
this disclosure opens an important debate about the future of creator rights in the age of ai. while tech giants leverage available big data to develop their models, creators find themselves in a difficult position where they unwittingly contribute to developing technologies that may replace them. the need for greater transparency and better protection for creator rights has become more urgent than ever. the future requires a fair balance between technical innovation and protecting the rights of individuals who contribute to building these technologies.
