Scientists in the UK have achieved success in training artificial intelligence to “hack” data by recognizing the sounds that occur when pressing keys on the keyboard.
Scientists in the UK have achieved success in training artificial intelligence to “hack” data by recognizing the sounds that occur when pressing keys on the keyboard.
Thus, AI is able to determine the sequence of keys pressed with an accuracy of 95%.
A team of British researchers and programmers has found that artificial intelligence is able to recognize the keys being pressed by analyzing the sound acoustic signals of typing. They conducted a series of experiments in which they pressed each of the 36 keys on a laptop 25 times using different fingers and pressure levels. The sounds were recorded both during Zoom calls and on a smartphone placed next to the keyboard. Based on the collected data, a machine learning system was developed that learned to recognize the characteristic acoustic signals for each key.
While it’s not known exactly what characteristics the system used, Joshua Garrison, principal investigator at Durham University, suggested that the placement of keys on a keyboard could also influence the character of sounds.
Further tests of artificial intelligence on other data confirmed that the system is able to accurately determine the key pressed based on sounds: in 95% of cases when recording during a phone call and in 93% of cases during Zoom calls.
This study is not the first to confirm the possibility of recognizing keystrokes by sound. However, the researchers note that their work uses the most up-to-date methods and reminds of the importance of ensuring one’s information security.
It is important to consider that laptops in public places are at risk of data leakage. Similar eavesdropping techniques can be applied to any keyboard. The researchers also advise using biometric passwords where possible, as well as enabling two-factor authentication.
Professor Feng Hao of the University of Warwick stressed that one should be careful to avoid entering sensitive data, including passwords, on the keyboard during Zoom calls. He also noted that in addition to sounds, visual signs of shoulder and wrist movements can also reveal information about the keys being pressed, which emphasizes the importance of protecting data on the side channels.
