A group of scholars from Cornell University has developed an artificial intelligence model that can recognize typed text (and consequently also passwords) from the sound of the keyboard. The experiment was conducted on a MacBook Pro, and achieved an astonishing level of accuracy: 95%.
A piece of news as fascinating as it is disturbing. However, there is one factor to consider: the AI is trained on a specific laptop (in this case a MacBook Pro), so the success rate drops dramatically if you use a different computer.
How the artificial intelligence that guesses the password from the keyboard works
The mechanism is relatively simple: each character you type produces a unique sound, slightly different from the others. An absolutely difference imperceptible to the human earbut which the machine is able to recognize and process.
The researchers explained that training the AI is also quite simple: it was enough press each key 25 times (to allow the AI to assimilate its sound) and associate each sound obtained with the reference letter. This last step is particularly important because the layout of the letters on the keyboard can vary according to the reference market.
Although the 95% success rate is high, the experiment revealed some critical issues. For example if the writer changes, the AI shows more difficulty recognizing sounds. This is because each user has his own typing style (some people hit and some touch the keys softer). With different styles the success rate in guessing the text is between 40 and 60%.
Should we worry?
At the moment the hypothesis of a cybercriminal stealing our password simply by recording our keystrokes is rather remote. As said thethe artificial intelligence in question must be calibrated on the reference computer and to be accurate it should be trained on our typing style.
However, looking forward, the technology is definitely a concern. The bad guys could train AI on the most common laptops (or perhaps those most common in offices), thus increasing the chance of success. The risk is even greater if you use them mechanical keyboards which tend to be noisier.
In short, if tomorrow someone asks you to send them an audio while typing on the keyboard of your laptop, maybe don’t do it.