Hugging Face, once hailed for its open-source models, appears to have overlooked crucial ethical guidelines in their vast repository. A damning report from AI Forensics reveals that seven out of the top nine image editing tools hosted by Hugging Face can easily generate nonconsensual deepfakes with minimal effort.
The nonprofit tested these models using straightforward prompts and found that 73% of generated images were sexual in nature, with a staggering 83% attempting to undress an individual—95% of whom were women. Even more concerning, nearly 7% of requests targeted children. According to lead researcher Paul Bouchaud, ‘No safeguards at all are being implemented at a platform level.' This stark reality contradicts Hugging Face’s own policies prohibiting harmful content.
While the platform can implement prompt-level filtering and output scanning, AI Forensics strongly urges immediate action to prevent further exploitation. The situation highlights a broader ethical dilemma in the development of AI: how much responsibility lies with developers versus users for ensuring content is generated responsibly?
