Researchers have found new hacking method using artificial intelligence (AI) by listening to how people type on their keyboard. This is possible with an AI trained to recognize the sound profile of the keys on a keyboard, including the sound, intensity and time of each keystroke.
A research paper by Joshua Harrison et al. presents a practical implementation of state-of-the-art deep learning model. Whereas it is able to “classify laptop keystrokes, using a smartphone integrated microphone.”
The classifier achieved a 95% accuracy when trained on keystrokes recorded by a nearby phone, the highest accuracy seen without using a language model.
Researchers also tested it during a Zoom call by recording the keystrokes through the laptop’s microphone. They found that the AI model accurately reproduced the keystrokes with 93% accuracy. Similarly, in another test using Skype, the model showed an accuracy rate of about 92%.
Advancements in number of microphones within the acoustic range of keyboards has made it easier for this new cyberattack method to exploit.
To test the concept, researchers pressed 36 individual keys 25 times each on a MacBook Pro to help the system recognize patterns. They then used an iPhone 13 mini placed 17 cm away from the keyboard to record the keystroke audio for their first test.
The researchers then recorded the laptop keys over Zoom using the MacBook’s built-in microphones.
The researchers warn that this new technique, which uses the trio of AI, microphones, and video calls, poses a greater threat to keyboards.
However, the AI model needs to be trained separately for each keyboard in order to understand what character each keystroke corresponds to.
To prevent these types of attacks, the study suggests changing one’s typing style. They also recommend using randomized passwords with multiple cases. As well as using biometric passwords like fingerprint or face scanning instead of typed ones.
