Hugging Face criticized as AI models enable non‑consensual nude images

A study by the non‑profit AI Forensics tested nine image‑editing AI tools hosted on the open‑source platform Hugging Face. Seven of the models could generate nude images of individuals using simple text prompts, without technical expertise. Researchers created a “honey‑pot” interface that recorded 1,081 user requests in one week; 73 % were sexual in nature and almost all sought to undress women, with 6.7 % targeting minors.

Hugging Face said it has taken the report seriously and will adjust its moderation practices, while disputing claims of systemic lack of safeguards. The findings echo earlier controversy over xAI’s Grok chatbot, which also produced non‑consensual sexual images, prompting investigations in the United Kingdom and the European Union. The European Parliament has voted to ban AI services that can nude people without consent, a measure supported by member states.

The study highlights a gap in safety mechanisms on the platform, noting that unlike large‑scale models from OpenAI or Google, many Hugging Face tools lack built‑in content filters, leaving moderation responsibility to individual developers.