A recent study has unveiled a concerning discovery: AI systems can “hear” and identify keystrokes with a staggering 95% accuracy, opening the door for potential misuse by hackers to decipher passwords.
UK-based tech experts utilized a standard 2021 MacBook Pro laptop for their experiment. Their findings, which were disclosed in a publication from Cornell University, revealed that when an AI model was activated on a nearby iPhone, it astonishingly replicated the entered password based on the sound of the keystrokes.
Additionally, the AI’s accuracy remained high even when it was analyzing typing sounds captured by a laptop microphone during a Zoom call, achieving a 93% success rate – an unprecedented figure for such a setting.
The study highlights the potential of what they termed “acoustic side channel attacks.” A significant concern is that users might not be adequately aware of this threat. As the researchers noted, “The common nature of keyboard sound emissions means they’re not only an accessible route for attackers, but also a risk often overlooked by users.”
The experiment involved typing on 36 distinct keys on the laptop 25 times, each with varying pressure and using different fingers. This allowed the AI to detect distinct sound patterns of each key. The setup included placing an iPhone 13 mini 17 centimeters away from the typing area.
This groundbreaking work was led by Joshua Harrison from Durham University, Ehsan Toreini from the University of Surrey, and Maryam Mehrnezhad from the Royal Holloway University of London.
This discovery sheds light on another potential pitfall of AI advancements. Esteemed voices in the industry, including OpenAI’s Sam Altman and business magnate Elon Musk, have previously stressed the imperative for caution with AI. Without the right precautions, the technology’s potential dangers to society might be vast.