Hugging Face Models Enable Nonconsensual Deepfake Nudes, Report Says

Hugging Face is a major repository where developers publish and run open-source AI models, datasets and applications through tools including Spaces. Unlike closed systems from Google and OpenAI, which generally apply centralized prompt and output safeguards, moderation on Hugging Face can depend on each developer. That distinction matters because general-purpose image editors can be repurposed to fabricate nonconsensual intimate deepfakes, a form of abuse increasingly linked to harassment, blackmail and the sexual exploitation of women and children.

European nonprofit AI Forensics said on July 28, 2026 that seven of nine prominent image-editing Spaces it tested produced topless versions of a clothed synthetic woman after a simple six-word request, without researchers trying to bypass safeguards. In a separate weeklong honeypot, the group collected more than 1,000 submitted images and prompts; 73% were sexual, and 83% of those sought to undress or sexualize a subject. Women accounted for 95% of targets, while 6.7% of sexual requests involved apparent children. Hugging Face acknowledged a developer-safeguard gap but disputed the methodology and representativeness.