Recently, researchers taught a deep learning model to interpret keystrokes from a remote location using nothing more than audio recordings. By training the model with recorded keystrokes, they were able to accurately predict what was typed with a 95% accuracy rate. The accuracy dropped slightly to 93% when using the Zoom video conferencing platform to train the system. This research suggests that sensitive information could be intercepted by anyone within earshot of someone typing on their laptop. These types of attacks, known as acoustic side-channel attacks, have become easier due to the prevalence of devices with microphones that can capture high-quality audio. Combined with advances in machine learning, these attacks are more dangerous than previously thought. To protect against these attacks, avoid logging into accounts on phone calls or in public places, and use biometric tools like fingerprint or face scanning.
The whytry.ai article you just read is a brief synopsis; the original article can be found here: Read the Full Article…