We have heard that AI is dangerous, but it gets worse! AI can figure out what keys you press on the keyboard by listening to the sounds they make, and it’s accurate 95% of the time.
A new research paper by a group of scholars from British universities has highlighted the vulnerability of information due to acoustic side-channel attacks.
Private information, like messages, passwords, and credit card details, could be in danger because AI can figure it out by using the sound of the keystrokes.
What Are Acoustic Side-Channel Attacks?
A side-channel attack is when someone tries to break into a system not by targeting the algorithm it follows but by looking at how it works physically.
Attacks that use sounds and other indirect methods put data safety on devices like laptops at risk. In these attacks, malicious individuals can gather and understand a device’s signals. They also use the sounds the device makes.
What Did the Researchers Find Out?
The researchers demonstrated how they trained a deep-learning model and used smartphone microphones to capture the sound of keystrokes.
Using acoustic side-channel attacks, they captured information, such as passwords from nearby laptops. The model worked well—it could identify the keystrokes through a microphone with about 95% accuracy.
Even when the model was taught using keystrokes recorded during Zoom meetings, it still performed remarkably well, with an accuracy of 93%.
How Can We Reduce the Threat of Acoustic Side-Channel Attacks?
With the advancements in deep learning tech, widespread use of microphones in devices, and extensive use of laptops – the danger of acoustic side-channel attacks will only increase. However, there are ways to lessen the risks.
Changing how you type is one way to protect yourself. The researchers found that using touch typing instead of regular typing lowered the accuracy of recognizing keystrokes from 64% to 40%. It might not work well for complicated typing with keys like Shift, Backspace, and other special keys.
There are other ways: You can use passwords that change randomly. You could also play various sounds through a speaker to mix up the sounds the microphone picks up. Using software to make fake typing sounds or background noise is also an option.