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

As AI-based deepfakes technology continues to advance, it is becoming increasingly difficult for people to judge “which content is AI forged” through the naked eye. In order to prevent images from being misinterpreted after being modified by crawlers, MIT recently announced a technology called Photoguard, which can process images to make them difficult to be recognized and tampered with by AI

It is reported that this "PhotoGuard" technology mainly prevents AI from recognizing the image content by making slight modifications to the image. This technology provides two solutions to prevent pictures 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; the other is the "diffusion" method to make the AI ​​think the picture is A is picture B, thus generating the wrong image.

However, IT House learned after inquiries that the application space of this technology is actually limited. Researchers warned that PhotoGuard technology is more suitable for anti-crawler scenarios. The technology itself also has certain limitations. Attackers only This technique can be cracked by cropping or flipping the processed image.

Researchers pointed out that to solve the problem of AI fraud, in addition to technical means, model developers, social media platforms and relevant departments also need to work together to provide a comprehensive "anti-AI fraud" solution

The above is the detailed content of MIT develops "PhotoGuard" technology to protect images from malicious AI editing. For more information, please follow other related articles on the PHP Chinese website!