Hugging Face Models Enable Nonconsensual Deepfake Nudes

A new report from the European nonprofit AI Forensics reveals that Hugging Face, the leading open-source repository for artificial intelligence models, hosts numerous tools that readily generate explicit images of women and children without consent. Researchers tested nine of the most popular image-editing models on the platform and found that seven complied without hesitation when given the straightforward prompt to make a clothed woman topless.

The findings highlight a stark contrast with major AI companies like Google and OpenAI, which have implemented basic restrictions against such requests. Hugging Face models showed no such barriers, allowing users to alter photos with minimal effort. The platform's Spaces feature, where people can run and share models directly, appears particularly vulnerable.

To gauge real-world activity, AI Forensics set up decoy image-editing tools on the site that collected prompts but produced no outputs. Over seven days, these received more than one thousand submissions. Seventy-three percent of the prompts were sexual in nature. Of those, eighty-three percent involved attempts to undress or sexualize submitted photos, with ninety-five percent targeting women. Nearly seven percent focused on apparent children.

Paul Bouchaud, a lead researcher at AI Forensics, noted that most tested Spaces can produce nonconsensual intimate images and that users are actively doing so. This comes as authorities in the United States have seized deepfake hosting sites and officials in the European Union and United Kingdom prepare to restrict nudify applications by year's end. Yet the open-source nature of Hugging Face, valued in the billions, continues to facilitate easy access to these capabilities.

The report underscores how readily available models can be repurposed for exploitation. Simple commands suffice to strip clothing from images, bypassing any meaningful oversight. Users appear drawn to the platform precisely because it lacks the filters found elsewhere, allowing experimentation that mainstream services block outright.

Critics argue that such lax environments accelerate cultural harms, particularly when children enter the picture. The data shows clear demand for material involving minors, raising questions about accountability on platforms that position themselves as hubs for innovation. Hugging Face has faced prior scrutiny over content moderation, but the new evidence suggests the issue persists at scale.

Law enforcement actions against dedicated deepfake sites indicate growing recognition of the problem. However, the decentralized model of repositories like Hugging Face makes enforcement difficult. Researchers warn that without intervention, these tools will remain accessible to anyone seeking to create intimate images against someone's will. The report calls attention to how open-source sharing, while promoted for collaboration, also spreads methods for digital violation.

Broader trends show increased focus on regulating AI-generated content that invades privacy or targets the vulnerable. Still, the speed at which users adopt these models points to underlying demand that tech platforms have yet to curb effectively.