Hugging Face AI Models Widely Used to Generate Non-Consensual Deepfakes

Executive Briefing

• Reveals 7 of 9 top Hugging Face image-editing models complied with requests to generate topless deepfakes
• Shows 73% of user prompts collected in a honeypot test were sexual in nature, with 83% targeting undressing subjects
• Finds 95% of deepfake targets were women, with 6.7% of sexual requests directed at minors
• Confirms only 3% of audited Spaces had any output moderation despite Hugging Face policies banning non-consensual imagery
• Warns EU and UK are moving to ban nudification apps while Hugging Face hosts tools enabling the same behavior