NVIDIA has been caught scraping videos from YouTube, Netflix, and other sources to train its AI models, raising significant legal and ethical concerns.

According to documents, messages and emails shared with 404 Media, NVIDIA instructed its employees to scrape videos from platforms including YouTube and Netflix to build datasets for its AI projects, including the Omniverse 3D world generator, self-driving car systems, and digital human products.

The initiative, internally named "Cosmos", aims to create a comprehensive AI model capable of simulating light transport, physics, and intelligence for various applications.

NVIDIA reportedly is using between 20 to 30 virtual machines on Amazon Web Services to download approximately 80 years' worth of video content daily, without the knowledge or consent of their owners.

Employees raised concerns about the legality of scraping copyrighted content without explicit permission. Despite these concerns, NVIDIA management assured them of compliance with copyright laws, claiming they had top-level clearance. 

Nvidia maintains that its practices are in "full compliance with the letter and the spirit of copyright law," arguing that copyright law protects specific expressions but not the underlying facts, ideas, or data used for model training.

Google (YouTube) and Netflix expressed concerns about the discovery, with YouTube's CEO Neal Mohan stating that scraping their content for AI training is a clear violation of their terms of service.

This finding is part of a broader trend in the AI industry in which companies adopt a "scrape first, ask forgiveness later" approach to data acquisition for AI training.