Runway, a billion-dollar Google-backed AI startup, is facing intense criticism over allegations that it harvested thousands of YouTube videos without permission to train its latest AI video-generation model. The allegations stem from a leak of an internal spreadsheet obtained by 404 Media.
The document, reportedly shared by a former Runway employee, details a plan to catalog and tag content from over 3,900 YouTube channels, including major media companies like Disney and Netflix, as well as popular individual creators like Casey Neistat and Marques Brownlee (MKBHD).
404 Media reports that the data was used to develop “Jupiter,” now known as Runway's Gen-3 AI video creation model. The spreadsheet also allegedly contains links to pirate video sites.
Runway, which is valued at $1.5 billion and has raised $141 million from investors including Google, has denied the authenticity of the spreadsheet. The company has previously said it uses “curated internal data sets” for training but declined to provide details.
YouTubers, including MKBHD, have expressed their dissatisfaction on social media, saying that over 1,600 of their videos have been removed. Another YouTuber, Mr WhoseTheBoss, also condemned the move on social media, calling it “horrifying,” and revealing that the company has used 1,600 videos from his channel.
Apart from YouTube, the report also suggests that piracy sites such as KissCartoon, which has a vast library of cartoon and animation content, were also included in the training dataset.
Earlier this year, YouTube CEO Neil Mohan told Bloomberg that training AI models on videos uploaded to the platform was a “clear violation” of the company's policies.
Runway is already facing lawsuits from creators over the unauthorized use of their content in AI training.
Recent reports have implicated other major tech companies, including Apple, Anthropic, and Nvidia, in similar practices. OpenAI is also unsure whether its text-to-video generation tool is trained on YouTube videos.
The document, reportedly shared by a former Runway employee, details a plan to catalog and tag content from over 3,900 YouTube channels, including major media companies like Disney and Netflix, as well as popular individual creators like Casey Neistat and Marques Brownlee (MKBHD).
404 Media reports that the data was used to develop “Jupiter,” now known as Runway's Gen-3 AI video creation model. The spreadsheet also allegedly contains links to pirate video sites.
Runway, which is valued at $1.5 billion and has raised $141 million from investors including Google, has denied the authenticity of the spreadsheet. The company has previously said it uses “curated internal data sets” for training but declined to provide details.
YouTubers, including MKBHD, have expressed their dissatisfaction on social media, saying that over 1,600 of their videos have been removed. Another YouTuber, Mr WhoseTheBoss, also condemned the move on social media, calling it “horrifying,” and revealing that the company has used 1,600 videos from his channel.
Apart from YouTube, the report also suggests that piracy sites such as KissCartoon, which has a vast library of cartoon and animation content, were also included in the training dataset.
Earlier this year, YouTube CEO Neil Mohan told Bloomberg that training AI models on videos uploaded to the platform was a “clear violation” of the company's policies.
Runway is already facing lawsuits from creators over the unauthorized use of their content in AI training.
Recent reports have implicated other major tech companies, including Apple, Anthropic, and Nvidia, in similar practices. OpenAI is also unsure whether its text-to-video generation tool is trained on YouTube videos.
