Debates over whether advances in artificial intelligence are good or bad aren’t going to subside anytime soon — and these new wrinkles aren’t going to help. British researchers They found that by combining machine learning with keystroke recordings, it is possible to train an AI to steal passwords simply by listening to what keys are being pressed. via the Tone E Computer tab The story comes from a so-called sonic attack that can extract data from keystroke sounds with a remarkable 95% accuracy. According to the article published on August 3, the researchers placed the iPhone a short distance away from the MacBook and recorded the sound of 36 keys as each key was pressed 25 times. The recordings allowed the researchers to make waveforms and spectral programs that would allow them to visualize the acoustic differences between each key.
These images were used to train a deep learning image classification program called CoAtNet, which essentially combines visual recognition with audio to steal information. In theory, hackers could record keystrokes using a malware-infected smartphone or, with nearly equal success, via Zoom (93% accuracy) or Skype (91% accuracy). Basic protection for all devices includes working in isolated environments – a University of Virginia article explains that researchers have been aware of this need for nearly 50 years – and controlling microphone access in general. In their article, the British researchers state that “simple changes in typing style may be enough to prevent the attack” and recommend other measures, such as the use of white noise and “random multi-state passwords”.
THE NAIS IS OFFICIAL EDITOR ON NAIS NEWS