Researchers from Tohoku University, Kyoto University and the Nara Institute of Science and Technology developed an AI-based acoustic side-channel attack that reconstructs laptop typing from keystroke sounds, CyberInsider reported on July 27, 2026. The method differs from earlier keyboard-sound attacks because it does not require advance training on recordings from the victim's specific keyboard, making it more practical in the scenarios described by the researchers.

The attack uses only audio access and does not require malware, privileged access or prior knowledge of the victim's device. The researchers described 3 attack scenarios: placing a smartphone near a victim in public, using a contact microphone on a shared desk or wall to capture vibrations, and recovering keystroke sounds sent as background noise during online meetings.

The system isolates individual keystrokes from an audio recording, groups similar sounds, and uses a Transformer-based language model (context-aware text prediction system) to infer the most likely character sequence. The researchers also added a feedback mechanism and an LLM to improve decoded text iteratively by using surrounding context, rather than relying only on acoustic signals.

In lab tests with a smartphone next to a 2019 MacBook Pro, reconstruction accuracy exceeded 99% after only 100 to 150 observed keystrokes. The technique also worked across Dell, HP, Lenovo and Apple laptops, though some models required more captured keystrokes. Vibration recordings from approximately three meters away with a contact microphone on the same desk, or through a wall, frequently exceeded 90% accuracy after around 150 to 250 keystrokes. Tests over Google Meet, Microsoft Teams and Zoom showed that meeting audio with noise suppression disabled could often reconstruct typed text with high accuracy after a few hundred keystrokes. The researchers recommended avoiding sensitive typing while unmuted, enabling noise suppression and watching for nearby recording devices.

Researchers from Tohoku University, Kyoto University and the Nara Institute of Science and Technology developed an AI-based acoustic side-channel attack that reconstructs laptop typing from keystroke sounds, CyberInsider reported on July 27, 2026. The method differs from earlier keyboard-sound attacks because it does not require advance training on recordings from the victim's specific keyboard, making it more practical in the scenarios described by the researchers.

The attack uses only audio access and does not require malware, privileged access or prior knowledge of the victim's device. The researchers described 3 attack scenarios: placing a smartphone near a victim in public, using a contact microphone on a shared desk or wall to capture vibrations, and recovering keystroke sounds sent as background noise during online meetings.

The system isolates individual keystrokes from an audio recording, groups similar sounds, and uses a Transformer-based language model (context-aware text prediction system) to infer the most likely character sequence. The researchers also added a feedback mechanism and an LLM to improve decoded text iteratively by using surrounding context, rather than relying only on acoustic signals.

In lab tests with a smartphone next to a 2019 MacBook Pro, reconstruction accuracy exceeded 99% after only 100 to 150 observed keystrokes. The technique also worked across Dell, HP, Lenovo and Apple laptops, though some models required more captured keystrokes. Vibration recordings from approximately three meters away with a contact microphone on the same desk, or through a wall, frequently exceeded 90% accuracy after around 150 to 250 keystrokes. Tests over Google Meet, Microsoft Teams and Zoom showed that meeting audio with noise suppression disabled could often reconstruct typed text with high accuracy after a few hundred keystrokes. The researchers recommended avoiding sensitive typing while unmuted, enabling noise suppression and watching for nearby recording devices.