AI can now steal passwords with a remarkable 95% accuracy by simply “listening” to the sound of someone typing, as revealed by a recent study.
Computer scientists from the UK accomplished this by training an AI model to recognize the sounds produced by keystrokes on a common MacBook Pro laptop.
This AI technique was so effective that when applied to a smartphone nearby, it was able to accurately replicate the typed passwords.
Even during a Zoom video conference, where the AI listened to the typing through the laptop’s microphone, it maintained a staggering 93% accuracy.
The researchers emphasized that many individuals are unaware of the potential threat posed by this method, which they named an “acoustic side channel attack.”
This means that hackers could potentially monitor the sound of keystrokes to gain unauthorized access to accounts.
The study highlighted that people tend to underestimate the risk because they often focus on hiding their screen while neglecting to mask the sound of their keyboard.
To test accuracy, the researchers pressed each of the laptop’s keys 25 times, varying the pressure and finger used. The AI program was able to identify unique elements of each keystroke’s sound, such as the wavelengths.
Placing an iPhone 13 mini smartphone 17 centimeters away from the keyboard allowed the AI to successfully “listen” to these distinctive sound cues.
The study was carried out by researchers Joshua Harrison from Durham University, Ehsan Toreini from the University of Surrey, and Maryam Mehrnezhad from Royal Holloway University of London.
This discovery highlights the additional risks associated with the growing use of AI technology, as it can potentially aid hackers in compromising cybersecurity.