AI program can steal your password by listening to the sounds your keyboard makes when you type it

Research published by Cornell University showed that scientists programmed an artificial intelligence system that listened to people typing their passwords and was able to correctly identify the keys with 95% accuracy.

The group programmed an AI system to listen to a typed password on MacBook Pro keys over both a phone and a Zoom call, according to Daily Fetched.

The AI model was trained by pressing each of the MacBook Pro's 36 keys 25 times each and recording the sounds. The sounds were fed into the AI so it could correctly identify each key.

Over the phone, the program correctly identified the keys with 95% accuracy, while over Zoom the number dropped slightly to 93%. The phone was placed about six and a half inches away from the keyboard, according to the Daily Mail.

“When trained on keystrokes recorded by a nearby phone, the classifier achieved an accuracy of 95 percent, the highest accuracy seen without the use of a language model,” the study reportedly said.

“Each key has a unique audio, or voice, that can be fingerprinted to infer what is being pressed," according to the study's co-author, Dr. Ehsan Toreini of Surrey University’s Centre for Cyber Security.

“I can only see the accuracy of such models, and such attacks, increasing," the doctor added.

Another professor, Feng Hao of Warwick University in England, recommended that people refrain from typing “sensitive messages, including passwords, on a keyboard during a Zoom call.”

The professor added that even subtle body movements could give away which keys were being pressed.

“Besides the sound, the visual images about the subtle movements of the shoulder and wrist can also reveal side-channel information about the keys being typed on the keyboard even though the keyboard is not visible from the camera," Hao mentioned.

Dr. Toreini has also made suggestions that Apple or other manufacturers should consider in order to ward off "side channel" attacks.

These included adding random noises to keystrokes for MacBooks and compressing the audio for Zoom calls.

“It gives you a hint of the tremendous improvement that has happened in the past five years in terms of the accuracy of the models, which somehow elevated the accuracy from 70-ish percent to around perfect results,” Dr. Toreini added.

Like Blaze News? Bypass the censors, sign up for our newsletters, and get stories like this direct to your inbox. Sign up here!