Google has officially admitted to using YouTube videos to train its AI models, including the latest versions of Gemini and the video-generation tool Veo 3.
This revelation has stirred a lot of concern among creators and digital rights advocates, particularly because many were unaware their content was being used in this way.
While YouTube’s terms of service technically allow Google to use uploaded content across its services, most creators did not anticipate their videos would be ingested into massive AI training datasets.
Google clarified that only a portion of publicly available YouTube videos were used for AI training, but it did not specify how many or which videos were selected.
Even a small sample from the platform’s massive video library represents an enormous training pool. The lack of transparency has frustrated users, especially since creators currently have no way to opt out of Google’s internal AI training processes.
An opt-out feature introduced in 2024 only prevents third-party AI companies like OpenAI or Meta from using a creator’s content, not Google itself.
Digital rights experts have raised red flags, noting potential ethical and legal concerns. One particular instance showed AI-generated content closely mirroring a specific YouTube video, with up to 90% similarity in the audio and 71% in visuals.
Critics argue that without clear rules or oversight, creators risk having their style, voice, and originality cloned by machines without credit or compensation.
Google maintains that it has long used platform content to improve services and that it incorporates protective measures to avoid direct copying of creator likeness or content.
However, these claims have been challenged as vague and insufficient. The company has also introduced indemnification for users of its AI products, shielding them from potential copyright lawsuits, which some see as a way to shift accountability away from itself.
