Runway, a company developing artificial intelligence-based video generation tools, is at the center of a scandal. 404 Media published an investigation alleging that Runway used thousands of YouTube videos to train its AI model, including content from well-known media companies and popular bloggers, without permission.
Runwayml.com
According to The Verge, 404 Media gained access to a spreadsheet containing data allegedly used by Runway to train its AI. It contains links to YouTube channels from companies such as Netflix, Disney, Nintendo and Rockstar Games, as well as popular bloggers such as MKBHD, Linus Tech Tips and Sam Kolder. In addition, the table contained links to news publications, including The Verge, The New Yorker, Reuters and Wired.
According to a former Runway employee, this table consists of a list of channels that “are the result of a company-wide effort to find quality videos to build the model.” These links were then used as input to a web crawler, which downloaded the content using proxy servers to avoid being blocked by Google.
In addition to YouTube channels, data was found containing links to pirated sites that show licensed films for free. It’s not yet clear whether Runway used these films to train its Gen-3 Alpha model, but as The Verge explains, it will most likely be impossible to know. The publication asked Runway to comment on the situation, but has not yet received a response.
It’s worth noting that Runway isn’t the only company using YouTube to train AI models. Earlier this year, OpenAI CTO Mira Murati said she wasn’t sure whether their AI model for generating Sora’s Full HD videos was trained on content from YouTube. Additionally, a recent joint investigation by Proof News and Wired found that Anthropic, Apple, Nvidia, and Salesforce trained their AI models on more than 170,000 YouTube videos.