American engineers from Google Research have developed a special neural network whose sole purpose is to bypass protection against bots by entering combinations of numbers, letters and characters in the image (the so-called captcha or captcha). Spotted on this portal ITHome.
It is noted that the developed software is called Pix2Struct. The built-in artificial intelligence can convert the image to text even if the characters in the picture are very distorted, which is designed to prevent robots from identifying the characters.
Captcha is used to verify that the visitor is a human, not a bot. But now bots may be capable of successfully passing this test if new software developments are incorporated into their work.
According to Pix2Struct developers, AI models also learn the basic structure of a site page regardless of what elements are used to create it. This allows you to recognize the characters better because sometimes some of them are stored inside the site code.
Pix2Struct was developed jointly by scientists and Google Research interns. At the same time, the authors assured that this model was not designed specifically for captcha.
Formerly Cloudfare offered to save the world Creating an alternative way of identifying people on the Internet from captcha.