Nvidia, one of the leading companies in the technology sector, has been accused of scraping content from platforms like YouTube and Netflix to train its artificial intelligence models. This technique, which involves extracting information from websites using specialized software, has been used by Nvidia to advance various commercial projects, according to a report by 404 Media.
Project Cosmos
According to 404 Media, Nvidia carries out these practices under an internal project known as Cosmos. This project aims to train multiple AI systems, including the 3D world generator Omniverse, autonomous driving systems, and "digital humans." The leaked information comes from internal conversations, emails, and company documents.
A former Nvidia employee revealed that the company urged its workers to extract videos from Netflix, YouTube, and other sources, such as the MovieNet platform and internal video game libraries, to train its AI models. They were assured that they had authorization to use this content, which has sparked controversy over potential copyright violations. To perform the scraping, Nvidia used tools like the open-source video downloader 'yt-dlp' and virtual machines that changed IP addresses, thus avoiding being blocked by the affected platforms.
Legal Concerns
Despite some employees' concerns about the legality of the Cosmos project, Nvidia's directors insisted that they had permission to use the content. In a statement, the company emphasized that it respects creators' rights and that its models comply with copyright laws. Nvidia argues that intellectual property laws protect specific expressions, not the data or information itself, and that fair use allows transforming works to train AI models.
Netflix has denied any agreement with Nvidia for the use of its content, stating that its terms of service prohibit scraping. Meanwhile, YouTube also expressed its disagreement, recalling the statements of its CEO, Neal Mohan, who affirmed that using videos to train AI is a "clear violation" of its policies.
