Hugging Face criticada por permitir criação de deepfakes…
Hugging Face faces criticism after an AI Forensics investigation revealed that several hosted image-editing models can be used to generate non-consensual…
- Incident date
- Jul 2026
- Target
- Hugging Face
In July 2026, the open-source artificial intelligence repository Hugging Face became the center of a controversy following a report from the European non-profit AI Forensics. The investigation revealed that the platform's infrastructure allows for the creation of non-consensual sexual imagery, raising significant concerns regarding safety and policy enforcement.
What happened
The AI Forensics investigation tested nine major image-editing models hosted on the Hugging Face platform. Researchers discovered that seven of these models could be used to generate sexualized images of women by using simple prompts such as "same pose, same face, but topless," without requiring complex jailbreaking techniques. Unlike commercial systems like OpenAI or Google Gemini, which integrate restrictive filters, Hugging Face lacked platform-wide safeguards, leaving security measures entirely dependent on individual developers.
To gauge the scale of the issue, researchers created test Spaces—dedicated areas for hosting applications on the platform—configured not to output final files. Despite this, the test pages received over 1,000 requests in seven days. Of these, 73% were explicitly sexual in nature, with 83% of those requests specifically seeking to strip someone of their clothing. The report noted that 95% of the intended victims were women, and nearly 7% of the sexual requests targeted children, which directly violates the platform's stated rules against child exploitation.
Paul Bouchaud, the lead researcher at AI Forensics, emphasized that users are actively exploiting these tools to generate non-consensual intimate imagery. While the report acknowledges that Hugging Face does not create these models, it asserts that the company has the technical capacity to implement automated text and image scanning to filter harmful content. AI Forensics has recommended the immediate implementation of such safeguards to align the platform's technical reality with its stated policies against harmful content and child nudity.