YouTube CEO Neal Mohan recently took a firm stance against the use of YouTube videos to train OpenAI’s text-to-video generator, Sora, citing a violation of the platform’s terms of service.
According to Bloomberg, Mohan emphasized that downloading transcripts or video clips from YouTube for training AI models like Sora is a clear breach of these terms. The debate surrounding the use of various web content by AI tools has been ongoing, with generative AI tools leveraging diverse data to improve their abilities in creating new content formats such as videos, photos, and narrative text.
Despite inquiries, OpenAI, with support from Microsoft, has not provided immediate comment on the sources of Sora’s training data. OpenAI’s Chief Technology Officer, Mira Murati, previously expressed uncertainty about whether Sora was trained on user-generated videos from platforms like YouTube, Facebook, and Instagram.
There have been reports suggesting discussions within OpenAI about potentially using transcriptions of public YouTube videos to train its next-generation large language model, GPT-5.
Mohan also shed light on Google’s approach to using YouTube videos to train its own powerful AI model, Gemini. He stressed the importance of respecting individual contracts between YouTube and content creators. While acknowledging that some YouTube content may be used for training purposes, Mohan assured that Google ensures compliance with creators’ agreements and YouTube’s terms of service.
