Hugging Face Hosts Widespread Nonconsensual Deepfake Tools

The open-source AI platform Hugging Face is hosting widespread tools for creating nonconsensual sexual deepfakes, according to new research that exposes significant gaps in content moderation at one of the industry's most prominent model repositories.

AI Forensics, a European nonprofit, tested nine top image editing models available directly on Hugging Face's platform. Seven of these models successfully transformed clothed images of women into topless ones using a simple six-word prompt—with no attempts to bypass safety mechanisms.

Why it matters

Hugging Face has been valued in the billions and serves as a central hub for AI development, making its content policies influential across the broader AI ecosystem. The platform's apparent inability or unwillingness to prevent abuse at scale raises questions about how open-source AI repositories balance accessibility with safety, particularly as law enforcement and regulators increase scrutiny of deepfake technologies.

The scale of abuse

To understand real-world usage patterns, researchers created decoy image editing tools on Hugging Face that logged user requests without generating images. Over one week, they collected more than 1,000 prompts. The findings were stark: 73 percent of prompts were sexual in nature, with 83 percent of those seeking to undress or sexualize people in submitted photos. Women were targeted in 95 percent of these cases.

Most disturbing, 6.7 percent of sexual requests appeared to target children, according to the research.

"This is not an empty threat, but actually people are using Hugging Face for that," said Paul Bouchaud, a lead researcher at AI Forensics.

The prompts revealed abuse extending beyond simple digital undressing. Researchers documented requests to depict women with semen, using sex toys, or performing sexual acts. Some prompts sought to remove hijabs from Muslim women.

Platform-level failures

The tested models did not identify themselves as nudifying services but rather as general image editing tools. Researchers found no platform-level safety mechanisms preventing their misuse.

"No safeguards at all are being implemented at a platform level. Only the developer can, if they want, implement some, and most of them do not," Bouchaud said. "Hugging Face can easily filter what is coming in and coming out of a system."

Additional reporting by WIRED and other researchers found multiple pages on Hugging Face promoting nudifying technologies and AI models capable of generating sexualized images of named celebrities and politicians. Benjamin Shultz of the American Sunlight Project identified dozens of models naming real people, with sample files showing suggestive poses indicating potential misuse.

Hugging Face did not respond to questions about its content moderation mechanisms. The company's content policies prohibit child sexual abuse material and sexual deepfakes created without consent or used for harassment. Some pages promoting nudifying services were removed after WIRED contacted the company, though the connection remains unclear.

The findings come as authorities begin cracking down on deepfake abuse. US law enforcement has seized deepfake hosting websites in recent months, while the EU and UK have proposed bans on nudify apps by year's end.

These details were first reported by WIRED.