Presented in a research endeavor conducted by Cornell University in the US, the AI program activated on a nearby smartphone showcased an impressive 95% accuracy in flawlessly reproducing entered passwords.
A group of UK-based computer scientists meticulously trained an AI model to recognize the distinct sounds of keystrokes on the 2021 version of the MacBook Pro, a laptop commonly found in the market. During a Zoom video conference, the AI tool impressively showcased its prowess by capturing typed content through the laptop’s microphone with remarkable precision. The result was a remarkable 93% accuracy rate in reproducing keystrokes, which in turn set a new benchmark for this method. This points to the fact that AI Can Steal Passwords By Listening To Keystrokes .
The researchers strongly warn, underlining that numerous users remain unaware that malicious actors can secretly monitor their typing patterns to compromise accounts—this type of cyberattack is termed an “acoustic side-channel attack.”
The study emphasized, “The ubiquity of keyboard acoustic emanations makes them not only a readily available attack vector but also prompts victims to underestimate (and therefore not try to hide) their output.”
To gauge accuracy, the researchers systematically pressed 36 keys on the laptop, each key pressed a total of 25 times. This process involved varying pressure and finger positioning, and the AI program adeptly identified the unique traits of each key press, including sound wavelengths.
During the experiment, researchers positioned the smartphone, an iPhone 13 mini, at a distance of 17 centimeters from the keyboard, thereby enhancing the overall comprehensiveness of the study.