MIT develops "PhotoGuard" technology to protect images from malicious AI editors

CTOnews.com, July 25 (Xinhua) with the continuous improvement of deep forgery technology based on AI, it is becoming more and more difficult for people to tell "which content is forged by AI" by the naked eye. MIT recently announced a technology called Photoguard, which makes it difficult for images processed by Photoguard technology to be identified and tampered by AI directly, helping to prevent images from being misread after being modified by crawlers.

It is reported that this "PhotoGuard" technology is mainly through minor modifications to the image to prevent AI from recognizing the content of the image. This technology provides two solutions to prevent the picture from being modified by AI, one is the "encoder" method, which interferes with the algorithm model's ability to understand the image by modifying the pixels of the image, and the other is the "diffusion" method which makes AI think that picture An is picture B, thus generating the wrong image.

However, after inquiry, CTOnews.com learned that the application space of the technology is actually limited. Researchers warn that PhotoGuard technology is more suitable for anti-crawler scenarios, and the technology itself has some limitations. Attackers only need to crop or flip processed images to crack the technology.

In addition to technology, solving AI fraud requires the joint efforts of model developers, social media platforms and relevant departments to provide a comprehensive "anti-AI fraud" solution, the researchers said.

▲ diagram source Pixabay