Particle.news
Download on the App Store

Report Finds Hugging Face Models Can Easily Create Nonconsensual Sexualized Images

A Tuesday report says gaps in platform-level safeguards leave hosted image-editing models able to produce undressed deepfakes, a risk that could prompt stronger enforcement and regulation.

Overview

  • AI Forensics published a report on Tuesday that tested nine popular image-editing Spaces on Hugging Face and found seven produced topless edits from a simple prompt.
  • Researchers created honeypot Spaces that collected more than 1,000 user submissions over seven days and found 73 percent of prompts were sexual with 83 percent seeking to undress subjects.
  • The study says 95 percent of targeted images were of women and roughly 6.7 percent of sexual requests appeared to involve children, highlighting specific harms to vulnerable people.
  • Hugging Face’s rules ban sexual deepfakes without consent and child sexual content but the company’s enforcement appears inconsistent and some nudifying pages were removed only after media inquiries.
  • AI Forensics recommends platform-level input and output filtering plus scanning of Spaces, and the findings add pressure for clearer rules, stronger tech controls, and regulatory action in the EU, UK, and U.S.