Researchers in the field of cybersecurity have shown how a machine learning algorithm can guess the PIN of a bank card, even if the victim covers the input panel with his hand. Finding out the coveted four numbers turned out in 41% of attempts. This attack vector assumes that attackers will first have to create an exact copy of the attacked ATM. This is important because the algorithm will take into account the distance between the buttons. After that, the criminals must use machine learning and teach the algorithm to recognize keystrokes. You can train on video recordings on which real bank customers enter their codes.
As part of the experiment, specialists collected (PDF) 5800 videos in which 58 different citizens entered PIN codes from their bank cards. In parallel, a Xeon E5-2670 with 128GB of RAM and three Tesla K20m with 5GB of RAM launched the AI model. Using three attempts at each ATM, the researchers were able to recognize the 4-digit PIN 41% of the time. Of course, the location of the camera played a key role in the calculation of the codes, since both right-handed users and left-handers had to be considered. If the camera could record audio, the expert model also noted the sound of each key, which increased the chances of successfully guessing the PIN.
anti-malware.ru
