AI Tool Can Decode Keystrokes on Zoom with 93% Accuracy, Researchers Warn of Growing Threat
Researchers from Durham University, the University of Surrey, and the Royal Holloway University of London have published a study highlighting the growing threat of acoustic side-channel attacks on keyboards during video conferencing. The study reveals that an AI tool developed by the researchers can decipher text, including passwords, from keystroke sounds recorded over Zoom with an accuracy rate of 93%.
The AI model used in the study demonstrated its ability to accurately decipher keystrokes from recordings of a Macbook’s keystrokes made during Zoom sessions. The researchers noted that the accuracy rate increased to 95% when the keystrokes were recorded using an iPhone 13 mini.
Acoustic side-channel attacks involve hackers exploiting information such as the sounds made by keystrokes or the amount of power a computer is using, rather than directly attacking the system’s code. The researchers emphasize that these types of attacks pose a significant threat to keyboard security.
The researchers used a 2021 16-inch MacBook Pro for their experiment, highlighting its consistent keyboard design with other recent MacBook models. The AI tool utilized deep learning, a subset of machine learning that trains computers to analyze data in a manner similar to the human brain.
The study also sheds light on potential countermeasures to mitigate the risk of acoustic side-channel attacks. The authors suggest that passwords containing full words may be at greater risk and that touch typing and adding background noise can lower the accuracy rate of the AI tool.
The authors of the study point out that while these types of attacks are relatively understudied, they have a long history. They mention that acoustic emanations were even identified as a vulnerability in a partially declassified NSA document dating back to 1982.
This study adds to the growing concerns surrounding the use of AI tools to compromise security and privacy. Insider previously reported that AI tools make online scams more challenging to detect, as they enable scammers to personalize their approaches for each target.
In 2019, specialists also raised concerns about how the advancement of AI and 5G technology could increase vulnerabilities in internet-connected devices and amplify cybersecurity threats.
At this time, the study’s authors have not responded to a request for comment from Insider outside regular business hours.
As the capabilities of AI continue to progress, it is crucial to address potential risks and vulnerabilities. Researchers, as well as technology companies and users, must remain vigilant in developing robust security measures to protect against emerging threats. The findings of this study serve as a valuable reminder of the ongoing need for enhanced cybersecurity measures in an increasingly digital world.