Report says 7 of 9 Hugging Face image models will undress people on request
The Verge AI · rss · 2026-07-28
A new report from AI Forensics says Hugging Face is being used to generate non-consensual sexual deepfakes, including prompts to undress women and children.
The nonprofit tested nine image-editing models hosted on Hugging Face and found seven would readily comply with simple undressing prompts. By contrast, mainstream models such as Google Gemini and OpenAI ChatGPT typically block those requests with safety guardrails. The Verge says Hugging Face is doing little to stop the abuse.
More from Safety
- Court win over AI scraping puts Google and Reddit back in the data-rights fight — JackFisherBooks · 2026-07-28
- METR says frontier models are increasingly reward hacking on coding and AI-R&D tasks — vkrakovna · 2026-07-28
- Anthropic says Claude Opus 4.6 found and decrypted BrowseComp answer keys — vkrakovna · 2026-07-28
- A Guardian essay says rogue AI needs better metrics, not just better locks — nordicinst · 2026-07-28
- AI smart lamp posts in the UK raise new fears of street-level surveillance — nordicinst · 2026-07-28
- Google publishes an ACM piece on enterprise security for the AI era — jordigg · 2026-07-28