New acoustic attack steals data from keystrokes with 95% accuracy
New acoustic attack steals data from keystrokes with 95% accuracy
A team of researchers from British universities has trained a deep learning model that can steal data from keyboard keystrokes recorded using a microphone with an accuracy of 95%.
Holy shit, accuracy drops to 93% only when done by Zoom call. That's very impressive. Privacy truly is dead.
Looks like I won't be getting a mechanical keyboard after all.
Perhaps the fallout from this will improve the lives of countless cube-mates?
There's plenty of quieter switches, and even some silent ones if you're that concerned. Gateron/Cherry Reds are quiet enough for most folks, but you could spend a little extra to get a board with Zealios or Zilent switches. Keep in mind that bottoming out the key will still make a lot of noise, but with a bit of conscious practice one can learn to type without it, significantly reducing the noise any mechanical key makes.
The article accounts for that already. The noises are more silent, but the model can still recognise the difference between various keys.
Holy shit this is so damn impressive, stealing keyboard clicks with audio, a audio based keylogger. Scary.
Given that there's variance among even the same key switches, as well as differences in sound based on the construction of the keyboard the switches are placed in, possibly even based on the surface the keyboard is placed on, the model would probably have to be trained specifically for a given target.
Since I'm working from home, I have much bigger things to worry about if attackers get access to my physical machine to train their model lol