A study conducted by researchers at Cornell University has discovered a new method that allows Artificial Intelligence tools to steal data and passwords by listening to the keystrokes used on the keyboard, with an accuracy rate reaching up to 95 percent.
An Artificial Intelligence model was trained to record keystrokes through a phone, where the built-in microphone listened to the sound of keystrokes on a MacBook Pro keyboard and was able to reproduce them with very high accuracy.
The team also tested the accuracy during a Zoom call, where keystrokes were recorded using the laptop computer’s microphone during the meeting, and in this test, the Artificial Intelligence was accurate by 93% in mimicking the keystrokes.
The Artificial Intelligence model is trained to identify the waveform, density, and timing of each keystroke. For example, you may press one key slightly later than the others by a fraction of a second due to your typing style, and this is taken into consideration to avoid losing any assumptions that have been made.
The Federal Bureau of Investigation recently warned of the risks associated with the recent developments in the field of Artificial Intelligence and how they are exploited in criminal activities. Cybersecurity teams are facing new challenges today, including adaptive malware that can quickly evolve using tools like ChatGPT.
