AI Can Now Hack Your Passwords by Analyzing Your Typing Sound

Image credit : Tech.co During your next conference call, it might be prudent to reconsider engaging in computer typing while your microphone is active, as UK researchers have achieved a successful feat in training artificial intelligence to decipher keystrokes during such calls. This breakthrough implies that the AI can deduce your typing even if your screen and keyboard remain hidden from view during the call, as detailed in a report by Hacker News. The research underscores that although individuals frequently take measures to shield their screens or keyboards while inputting passwords, they often overlook obfuscating the auditory cues of their keyboard. Nonetheless, AI possesses the capability to capture virtually all of your computer keystrokes, encompassing potentially sensitive emails or documents. [caption id="attachment_190007" align="aligncenter" width="2048"] Image credit : HackRead[/caption] The study indicates that an AI, trained on keystrokes that were recorded through a nearby phone, accomplished an impressive accuracy rate of 95%. When trained on keystrokes obtained from the video conferencing platform Zoom, the AI exhibited a slightly lower accuracy rate of 93%. Even when the AI didn't achieve perfect accuracy in deciphering typed content, its errors often amounted to only a single key's discrepancy. This slight deviation could be conveniently exploited by potential hackers to overcome the inaccuracy. To facilitate the AI's operation, a training process is necessary. This entails the attacker recording the target's keyboard keystrokes, accomplished either through a microphone or by infecting a smartphone with malware. The AI's training involved the use of a MacBook Pro and an iPhone 13 mini. Researchers pressed each key on the computer 25 times, capturing the corresponding keypress sounds through the phone situated at a distance of 6.5 inches. Also read : Google's AI Could Revolutionize Robotics: Here's How In terms of safeguarding oneself from such vulnerabilities, the researchers propose altering your typing style when entering sensitive information. Alternatively, employing ambient noise or a software-based keystroke audio filter as background sound during typing can help mask the distinctive auditory patterns of your keystrokes.