Hugging Face Image Models Enable Deepfake Nudes
Researchers have discovered that top image editing models hosted on Hugging Face can be easily manipulated to create nonconsensual explicit deepfakes. An analysis of 1,000 image editing prompts reveals how users are exploiting the software for this purpose. The findings highlight a significant safety issue within the AI development platform, which hosts thousands of open-source models. The study underscores the challenge of moderating AI tools that can be repurposed for harmful applications, raising concerns about platform responsibility and the need for stronger safeguards.
Key facts
- Researchers tested top image editing models on Hugging Face.
- Models could easily create explicit deepfakes.
- 1,000 image editing prompts show how people use the software.
- The findings highlight a safety issue on the platform.
- Hugging Face hosts thousands of open-source models.
- The study raises concerns about platform responsibility.
Entities
Institutions
- Hugging Face
Sources
- Wired AI —