New Amazon, Intel Patents Describe Ways of Overcoming CAPTCHA Defeats

Recent Amazon and Intel patents give intriguing insight into the arms race developing in the field of CAPTCHA (Completely Automated Public Turing Test to Tell Computers and Humans Apart) authentication to distinguish between humans who access websites and computer bots with artificial intelligence that are programmed to behave like humans. U.S. patent 9,723,005, granted Aug. 1 to Amazon Technologies, tells in detail how programmers are using increasingly clever techniques in defeating CAPTCHA challenges and what Amazon is doing to try to overcome them. Bots will probably learn to fail like humans, Amazon predicted. So tests can then start using optical illusions or misquotations that humans will ignore subconsciously, but bots won’t, it said. The CAPTCHA questions can also use cookies stored on a device, together with the user’s online search and purchase history, to frame individualized CAPTCHA questions, said Amazon. So a stamp collector who has searched for and bought rare stamps will be more likely to answer personalized questions on philately correctly than a bot with no access to the user’s personal online history record, it said. An Intel patent application (US 2017/0180348) published in late June offers another solution to the escalating problem of CAPTCHA defeats. Intel’s plan is for a website to ask for live video clips of those attempting to prove they are living human beings, it said. The online system tells them to center their faces on the screen, and then poses a few anti-spoofing commands, such as “make a facial expression” or “show a close-up of your ear,” it said. To guard against a bot displaying pre-shot photo images of a face, the camera can be made 3D-capable or enabled to check for infrared or ultrasonic reflections that reveal body heat and facial contours, it said. Amazon and Intel representatives didn't comment Monday on plans to commercialize the techniques described in the patents.