Researchers find image editing tools on Hugging Face can be used to make explicit deepfakes

Researchers examining image editing systems hosted on Hugging Face say several of the most widely used tools could be pushed into creating explicit deepfakes. The findings, reported this week, focus on how easily the models can be directed to produce nonconsensual nude images.

According to Wired, the testing also included 1,000 prompts that shed light on how people are using the software in practice. The examples were used to show the range of requests that can be made of the models, including abusive ones.

A separate TechCrunch report said the episode has revived debate over how much control AI developers should build into increasingly capable systems. That discussion has centered on whether the priority should be stronger alignment, tighter containment, or both.

Hugging Face is a major online hub where developers share and download machine learning models, including tools for image generation and editing. The reports add to broader scrutiny of how open model marketplaces handle harmful uses of generative AI.