Google has invented a foolproof way to tag AI-generated images

It is becoming increasingly difficult for humans to distinguish AI-generated images from pictures created by humans, so engineers at Google and other technology giants are seriously thinking about developing technical means for this purpose. One solution could be the company's SynthID technology, which allows AI-generated images to be labeled without degrading their quality. It works by introducing subtle changes in individual pixels of images, creating a watermark invisible to the human eye but recognizable by computers. Such a mark is detectable even after adding various filters, making color corrections or adjusting brightness, Google claims. Moreover, SynthID cannot be removed even if the image is cropped.

Google was one of the technology companies that promised the US presidential administration to introduce some sort of "watermarks" for AI content. At the same time, Adobe is engaged in similar work - the company is creating data encoding methods to track the ways of image creation and document subsequent changes. It is possible that attackers with powerful AI platforms will deliberately refuse to label AI images, but if major players make it mandatory to label them, people will be wary of images without watermarks.