ARTICLE: AI cyberattack could figure out your password from keyboard acoustics

Hacking passwords by recording the sound of your keystrokes is not new, but AI has enabled researchers to do so with greater precision.

Durham University, University of Surrey, and Royal Holloway University of London computer scientists simulated a cyberattack in which a deep learning model classified keystrokes using Zoom and smartphone microphone audio recordings. Researchers obtained 93 percent accuracy when trained on keystrokes using Zoom, and 95 percent accuracy when using a smartphone. Using commercially available hardware and software, they demonstrated that this type of attack is feasible.

This cyberattack, known as an acoustic side channel attack (ASCA), was studied in the early 2000s but has received less attention recently. However, due to the rise of video conferencing, remote workers in cafés and public spaces, and recent advancements in neural networks, the researchers noted that this threat could become more widespread. The study, which was presented and accepted as part of the 2023 IEEE European Symposium on Security and Privacy Workshop, was conducted to demonstrate its viability and bring attention to ASCAs now that deep learning tools are more readily available.

How then can individuals defend against acoustic side channel attacks? The simplest solution is to avoid entering your password near microphones or on Zoom. However, this is not always possible. Since it is difficult to recognize the release of the Shift key, the researchers recommended using two-factor authentication, biometric registration, and randomized passwords with upper and lower case characters.