FOR IMMEDIATE RELEASE

July 30, 2026

Contact: minority_jecpress@jec.senate.gov

Senator Hassan Presses AI Model Platform on Need to Combat Nonconsensual Deepfakes

Request to Hugging Face Comes After Public Reporting and Research on the Company’s Failure to Curb the Creation of Explicit Imagery on the Platform

WASHINGTON – U.S. Senator Maggie Hassan (D-NH), Ranking Member of the Joint Economic Committee, today pushed Hugging Face, a platform for sharing and using AI tools, to prevent people from using the platform to create nonconsensual deepfakes that depict sexual content of women and children. This comes after public reporting and research from the technology watchdog AI Forensics that found serious problems with the company’s approach to protecting against this type of activity.

Ranking Member Hassan requested detailed information on “what immediate and long-term steps Hugging Face will take to address the reported use of your technology to easily create deepfake sexual content of women and children.”

Ranking Member Hassan continued, “According to a recent report in WIRED, AI-powered tools hosted on Hugging Face have been used to generate nonconsensual intimate imagery – including sexual content depicting women and children – despite company policies that prohibit sexual content created ‘without explicit consent’ and prohibit ‘any sexual [c]ontent involving minors. This reporting concluded that Hugging Face ‘has a widespread problem with nonconsensual deepfakes,’ which raises serious concerns about what if anything [the] company is doing to protect against the exploitation of women and children.”

This request is part of Ranking Member Hassan’s ongoing comprehensive effort to combat scams and criminal fraud, often enhanced by AI. As part of her scam prevention efforts, Hassan has opened investigations into the roles that AI and voice cloning companies, federal agencies, data brokers, telecommunication companies, music streaming services, satellite internet providers, and online dating platforms have in protecting people from scams.

Read the full text of Ranking Member Hassan’s request here or below.

Dear Mr. Delangue:

I write today to request information about what immediate and long-term steps Hugging Face will take to address the reported use of your technology to easily create deepfake sexual content of women and children. According to a recent report in WIRED, AI-powered tools hosted on Hugging Face have been used to generate nonconsensual intimate imagery – including sexual content depicting women and children – despite company policies that prohibit sexual content created “without explicit consent” and prohibit “any sexual [c]ontent involving minors.” This reporting concluded that Hugging Face “has a widespread problem with nonconsensual deepfakes,” which raises serious concerns about what if anything your company is doing to protect against the exploitation of women and children. 

The Hugging Face platform hosts AI models capable of generating text, images, and interactive content, and its “Spaces” feature allows users to access these tools directly through a web browser. For image or video generation models, Hugging Face also offers model creators the option of including examples of generated images or videos directly on their pages. According to WIRED, the technology watchdog AI Forensics found that seven of nine highly ranked image-editing Spaces readily undressed an image of a clothed woman using a “simple, six-word, prompt.” In a separate review of nearly 500 Spaces, the researchers found that only three percent “showed evidence of output moderation,” while “none moderated input prompts or images.” They concluded that Hugging Face had implemented “virtually no safeguards” against the generation of nonconsensual intimate imagery.

To test the prevalence of such misuse, the AI Forensics researchers created their own Spaces on Hugging Face. Although the researchers did not advertise their Spaces as tools for generating sexual content, 73 percent of the requests that they received were sexual in nature. Of those requests, more than 80 percent sought to undress or otherwise sexualize a person, 95 percent of whom were women. The researchers also found that nearly seven percent of sexual requests involved a minor. (As WIRED noted, nonconsensual intimate imagery is often used to “blackmail, harass, and harm women and girls around the world.”)

In light of these results, a lead AI Forensics researcher stated that Hugging Face’s lack of “the means or will to moderate” such tools should not exempt the company from responsibility for their deployment and “predictable outcomes.” Given the foreseeable harm and apparent serious flaws with Hugging Face’s moderation efforts, I seek responses to the questions below:

1. At the time of the WIRED report’s publication:

1. What review and approval process did Hugging Face apply before an image-editing Space or other hosted third-party tool became publicly accessible?
2. How did Hugging Face ensure that these Spaces and other hosted third-party tools complied with the platform’s policies?
3. How did Hugging Face identify and respond to prompts that violated the platform’s policies?
4. Did Hugging Face maintain any additional safeguards against the use of Spaces or other hosted third-party tools to create nonconsensual intimate imagery?
5. Did Hugging Face maintain any additional safeguards against the use of Spaces or other hosted third-party tools to create sexual content involving minors?

2. Since the publication of the WIRED and AI Forensics reports: 

1. What steps has Hugging Face taken to remedy the issues identified in those reports?
2. Hugging Face has reportedly acknowledged a “gap” in safeguard adoption among some developers of image-editing tools and stated that it is working to address the issue. Please describe this work and the timeline for it.
3. How many Spaces or other third-party tools has Hugging Face identified as capable of generating nonconsensual intimate imagery?
4. How many Spaces or other third-party tools has Hugging Face removed from the platform or otherwise enforced against?

3. Hugging Face reportedly stated that it would not be technically possible to filter prompts or scan outputs across the platform. Has Hugging Face attempted to implement either measure? If so, please describe those efforts and the specific technical limitations that prevented implementation.

1. Does Hugging Face maintain machine learning and/or rule-based filters? If so, please describe these measures.

4. Please describe the architecture of Hugging Face’s Spaces and identify the respective roles of Hugging Face, the Space operator, the underlying model developer, and any third-party service provider in preventing misuse of hosted tools.

5. Describe the process through which a person may notify Hugging Face that a Space, model, repository, or other tools hosted by Hugging Face were used to generate non-consensual intimate imagery involving that person.

###