Hugging Face Faces Widespread Deepfake Nude Problem
Original: Hugging Face Has a Deepfake Nudes Problem
Why This Matters
Highlights a major content moderation gap on a leading open-source AI platform used by millions of developers worldwide.
European nonprofit AI Forensics tested 9 top image editing Spaces on Hugging Face and found 7 could easily generate nonconsensual topless images. A honeypot study of 1,000+ prompts revealed 73% were sexual in nature, with 6.7% targeting apparent children.
AI Forensics, a European nonprofit, published a report on Tuesday revealing that Hugging Face—an open-source AI platform valued in the billions—has a widespread nonconsensual deepfake problem. Researchers tested nine top image editing Spaces on the platform and found seven could easily convert clothed images of women into topless ones without guardrails. In a separate honeypot experiment, AI Forensics created decoy Spaces that logged over 1,000 prompts in one week. Of those, 73% were sexual in nature; 83% of those sought to undress or sexualize the submitted photo subject, with 95% of targets being women. Notably, 6.7% of sexual requests appeared to target minors. 'Users are actually using Hugging Face for these purposes,' said lead researcher Paul Bouchaud. A WIRED review also found multiple pages promoting nudifying tools and models potentially capable of generating sexualized images of named celebrities and politicians. Hugging Face did not respond to questions about its content moderation. The company's policies prohibit CSAM and non-consensual sexual deepfakes, and some pages were removed after WIRED's inquiry. Unlike mainstream models from OpenAI and Google, many models on Hugging Face lack safety guardrails against undressing-style image generation.