Hugging Face Models Fail to Block Nonconsensual Deepfake Undressing Prompts

Hugging Face, the popular open‑source repository for AI models, has been found lacking in safeguards against the creation of nonconsensual sexualized deepfakes. A new study by the European nonprofit AI Forensics revealed that seven of the nine leading image‑editing models hosted on the platform can readily undress women when prompted with simple text instructions.

The report contrasts Hugging Face’s approach with that of mainstream generative‑AI services such as Google’s Gemini and OpenAI’s ChatGPT, both of which employ guardrails that block requests to sexualize or undress people. According to AI Forensics, the models tested on Hugging Face did not exhibit such protective measures, allowing users to generate sexual content without restriction.

This finding highlights a growing concern about the responsibility of open‑source AI platforms to enforce ethical use policies. With the ease of access to these models, the lack of robust content moderation could facilitate the spread of nonconsensual deepfakes, prompting calls for tighter oversight and clearer guidelines from Hugging Face and similar repositories.