An AI developed by American researchers has been trained to decipher keystrokes based on sound cues emitted by a smartphone placed near a laptop.
During the experiment, the researchers correlated the sound of each keystroke with the corresponding text input. Following the input of a password on the computer, the AI utilized audio cues to match the heard phrases, achieving an impressive accuracy rate of 95% in identifying logins.
Subsequent tests were conducted to assess the feasibility of using this method to intercept passwords during virtual gatherings on platforms like Zoom or Skype, potentially enabling malicious attacks. The AI managed to maintain an accuracy rate of just under 92% for Zoom and slightly lower for Skype.
Password managers such as Keepass automate the entry of credentials in login fields or utilize a key combination, posing a potential vulnerability to interception by AI.
In their study, the researchers utilized an iPhone 13 Mini placed 17 inches away, Zoom and Skype for video conferencing, and a 2021 Macbook Pro equipped with an M1 chip and a 16-inch display.
The audio recordings were converted into waveforms and spectrograms, which were then utilized to train a graphical AI model. As a security precaution, users are advised to adopt the ten-finger typing method to decrease the likelihood of successful recognition of keystrokes.
The complexity of reconstructing passwords using AI is heightened by the presence of uppercase and lowercase letters, special characters, and other variables. Employing a password manager that automatically fills in credentials with a single click provides enhanced security. If you have not yet explored this option, consider referring to our list of recommended password managers for heightened protection.
