The rise of artificial intelligence (AI) tools may have a big down-side, too, apparently. Hackers can potentially train AI to listen to your keystrokes and decipher the sound you make while typing to predict sensitive data like passwords with a great deal of accuracy. The “startling” findings are part of a new combined study done by researchers from Durham, Surrey, and Royal Holloway universities.
The researchers have found that it is possible to train AI to detect and decode keystrokes through microphones with an accuracy far exceeding 90 percent. In short, it can get scary close to sneaking on you and stealing your personal data, without your knowledge. To prove this startling revelation, the researchers took a MacBook Pro and pressed each of its 36 keys 25 times while recording the sound that accompanied each interaction.
Also Read | OnePlus Ace 2 Pro launched with up to 24GB RAM, 150W fast charging and more: Specs, price
They took a two-pronged approach, one where they recorded these sounds coming through a recording done using a Zoom video call and other where they did it directly using an iPhone 13 Mini. The sound data was then fed into the AI model for training purposes. What’s surprising is that the AI was able to master these keystrokes with a great deal of accuracy regardless of whether it was direct or recorded. Basically, it took all that data and was able to accurately predict what the researchers had been typing with an accuracy of over 90 percent.
Also Read | Honor to make India comeback with former Realme senior exec at helm; to soon launch new smartphone
“With recent developments in deep learning, the ubiquity of micro-phones and the rise in online services via personal devices, acoustic side channel attacks present a greater threat to keyboards than ever,” the research paper notes, adding that “Our results prove the practicality of these side channel attacks via off-the-shelf equipment and algorithms.”
More specifically, it was able to successfully interpret the keystrokes with an accuracy of 95 percent from data captured directly using an iPhone while the indirect route which used data sourced from a recorded Zoom call, fetched an accuracy rate of 93 percent.
Follow FE Tech Bytes on Twitter, Instagram, LinkedIn, Facebook.
