AI video generator Runway used pirated content for training

AI video generator Runway was trained on stolen YouTube content and pirated films, a leaked spreadsheet reveals. The information was obtained by 404 Media and shows exactly where the data training sourced material from.

The AI generator in question was code-named Jupiter and released officially under the name Gen 3. It received investments from Google and Nvidia of $141 million. Yet again, the ethics behind the source material for another AI generator are called into question.

Runway’s Gen-3 Alpha tool gained significant attention for producing highly realistic videos when it was released in June. At that time, Runway said it used both videos and images for training but predictably didn’t specify the sources.

Well well well. Runway AI video generator was trained on YouTube videos without permission, including 1600+ MKBHD videos 🫠https://t.co/U4s5KIq1XZ

— Marques Brownlee (@MKBHD) July 25, 2024

The document acquired by 404 Media shows that the training data included popular content from major YouTube channels like Disney, Netflix, and Sony, along with links to sites known for hosting pirated content.

While 404 Media couldn’t confirm all the assets used, it appears likely that Gen-3 Alpha used this content. This suggests that AI companies might be ignoring copyright laws to train their models. 404 Media also easily created believable videos of well-known YouTube personalities using Runway’s tool.

Runway reportedly used a proxy to avoid being blocked by YouTube. A former employee said finding good-quality videos was a company-wide effort, and a web crawler downloaded videos from those channels.

Recently, OpenAI was called out for using YouTube videos to train its text-to-video generator Sora without users’ permission. Ethics and AI data training have been constant issues, with even so-called ethical companies such as Adobe not being entirely transparent about their sources.

[via futurism]

---