Hugging Face Platform Found Hosting Widespread Nonconsensual Deepfake Nude Tools

AI Forensics examined nine of Hugging Face's most used image-editing Spaces and discovered that seven could readily produce topless versions of clothed women using a simple prompt. In a separate honey-pot test, the group logged more than 1,000 prompts over a week, finding that 73% were sexual in nature, with 83% seeking to undress the uploaded person and 95% of those requests involving women; 6.7% appeared to target children.

The researchers observed that many Spaces do not identify themselves as nudifying tools, instead presenting as general image editors, and that the platform's safety mechanisms are largely absent at the platform level. Additional review identified multiple pages advertising nudifying technologies and models capable of generating sexualized images of public figures. Hugging Face declined to comment on its moderation processes, though it maintains policies prohibiting child sexual abuse material and non-consensual deepfake content. The findings highlight a gap between policy and enforcement on a major open-source AI repository, raising concerns about the ease with which harmful sexual deepfakes can be created and distributed.