Hugging Face Hosted AI Tools That Strip Clothes Off Women's Photos, Researchers Find

Hugging Face bills itself as the GitHub of AI. Billions in valuation, millions of models and datasets, used by researchers and developers worldwide. It also, according to a report published by the European nonprofit AI Forensics, is hosting tools that make nonconsensual nude deepfakes trivially easy to produce.

AI Forensics researchers tested nine of the most popular image-editing "Spaces" on Hugging Face, the interactive tools users can run directly on the platform. Seven of nine easily turned a clothed photo of a woman into a fake topless image, according to the report cited by WIRED.

The researchers didn't stop at a lab test. They built their own honey-pot Spaces on Hugging Face designed to look like functional image editors but that produced no actual output. Over one week they logged more than 1,000 prompts and images submitted by real users.

The numbers are ugly. Seventy-three percent of prompts were sexual in nature, according to AI Forensics. Of those, 83 percent were attempts to undress or sexualize the person in the photo, and 95 percent of those targets were women. Separately, 6.7 percent of the sexual requests appeared to target children, the report found.

"Most of the Spaces [tested] can be used for generating nonconsensual intimate images, and users are actually using it for these purposes," Paul Bouchaud, a lead researcher at AI Forensics, told WIRED. "This is not an empty threat, but actually people are using Hugging Face for that."

WIRED's own review found multiple pages on Hugging Face promoting nudifying tools or models capable of generating sexualized images of named celebrities and politicians. Some of those pages disappeared after WIRED reached out to the company, though it's unclear whether the takedowns were a direct response.

Hugging Face's Response

Hugging Face did not answer WIRED's questions before publication about its moderation practices. After the story ran, the company gave WIRED a document it had originally sent to AI Forensics. In it, Hugging Face says the findings point to a "gap" in safeguard adoption among some developers who build image-editing models on its platform, and says it's working to close that gap.

But the same document pushes back hard on the research itself, according to WIRED, with Hugging Face saying it questions AI Forensics' methodology and arguing the honey-pot approach could have produced false positives. The company does have content rules on the books barring child sexual abuse material and nonconsensual sexual deepfakes used for harassment. Whether those rules are actually enforced against third-party Spaces built and hosted on its platform remains unclear.

An open-source platform faces a structural dilemma. Hugging Face doesn't build most of these image models itself. Developers upload them, and anyone can run them through the site's Spaces feature. That structure is a big part of why Hugging Face has become a hub for AI research. It's also, per this report, why nudifying tools keep showing up on it.

The Human Cost, in the UK

While researchers were running honey-pot tests, a criminal case in England showed what this technology does to a real person. Georgia Horton, from the south of England, discovered that photos pulled from her social media had been manipulated into pornographic images by someone she knew, an old college friend named Ethan Orchard, according to the BBC.

Orchard, from Southsea, was sentenced to three years and four months in prison in July after creating and sharing fake explicit images of Horton and two other women, the BBC reported. He also posted violent sexual fantasies about the women online and solicited other men to share their own, according to Horton's account to the BBC.

"He would make people choose between us, or ask what they wanted to do to us," Horton told the BBC, describing one post that asked whether commenters would "save her," "rape her," or "sell her."

Earlier in 2026, new UK legislation made it a criminal offense to create sexually explicit deepfake images without consent, building on existing laws against sharing nonconsensual intimate images. Horton called that law "a start" but said it hasn't stopped the tools from spreading. "The technology to make these images and videos is still out there and still widely available," she told the BBC. "Self-regulation isn't working."

A spokesperson for the UK's Department for Science, Innovation and Technology told the BBC the government takes the issue seriously, calling nonconsensual deepfakes "utterly vile."

What's Next

The AI Forensics findings are a documented test result: nine Spaces tested, seven produced nonconsensual nude images, and real user prompts logged over a week show overwhelming demand for exactly that. These are measured outcomes from a specific, disclosed methodology.

Hugging Face's countercharge, that the honey-pot methodology could generate false positives, is a fair scientific question to raise and deserves scrutiny. No independent third party appears to have audited AI Forensics' methodology yet, and Hugging Face hasn't detailed exactly what it disputes beyond the general claim. Readers should treat the exact percentages as AI Forensics' own reported figures pending any outside replication, while noting that the underlying capability (Spaces converting clothed photos into nudes) was independently confirmed by WIRED's own review.

The EU and UK have both moved to draft bans on nudifying apps by the end of 2026, and US law enforcement has seized deepfake hosting sites in recent months, according to WIRED. None of that touches the open-source pipeline where models are trained, uploaded, and run through a legitimate research platform valued in the billions. Whether Hugging Face tightens Spaces moderation, or whether regulators force it to, is the next thing to watch.