A recent study pointed out that the Artificial Intelligence (AI) can figure out passwords just by the sound of the keyboard. The key test applied in an AI program showed an impressive performance, with more than 90% accuracy.
The research was carried out in the United Kingdom at the University of Durham. The researchers developed a program using the deep learning technique.
see more
Duet AI: New Google AI for Work Announced; meet her
THESE secret codes help you find out if your phone has been…
Thus, the system could be able to detect keyboard noise from cell phones, notebooks and other mobile devices.
Basically, with the help of a microphone, the AI learns the sounds of the keys. Subsequently, the system decrypts strings that have been typed. In addition to the sound, it recognizes keyboard patterns such as intensity and typing time to formulate your responses.
The method proved to be very efficient in identifying the sound of keys, even in video calls.
On Zoom, a virtual meeting program, the accuracy was 93%. On the Skype platform, the AI got the passwords right in 91.6% of the cases.
The hands-on test demonstrated how quickly the machine can learn patterns. Therefore, the researchers made a warning about how a cyber attack can be accomplished with the aid of malware and AI technologies.
To carry out the keyboard noise tests, the scientists trained the system to identify 36 keys.
To do this, letters and numbers were typed with different fingers and typing strengths approximately 25 times on a MacBook Pro.
Then, the sounds were captured by a microphone attached to a iphone 13 mini and transmitted via video call. From the wave data and the spectrogram, the AI program was able to isolate the sound of each key.
“When trained on keystrokes recorded by a nearby phone, the classifier achieved an accuracy of 95%, the greater accuracy seen without the use of a language model,” the scientists said in the research publication.
The results revealed that the AI program got the exact password right 95% of the time through the microphone. In video calls, the accuracy was 93% via Zoom and 91.6% via Skype.
(Image: Freepik/Playback)
The new AI system created at the university pointed to worrying issues of cybersecurity, confirming how the new technology can be used for cyber attacks.
Similar programs manage to invade systems and steal data through video calls or malware, the name given to malicious software that steals data from computers.
However, scientists stress that the program cannot identify passwords from any keyboard. First, he needs to learn typing patterns.
Finally, it is worth noting that the study of AI has never been applied outside of British research. Anyway, the result serves as an important alert.
Thus, researchers recommend that people develop methods to change their typing style. Likewise, it's important to add extra layers of protection to passwords.
For example, we recommend changing them periodically, using two-factor authentication, creating complex passwords, and muting your microphone when not talking on video calls.