Experts from Durham University, Surrey University and Royal Holloway, University of London recorded the sound by pressing each of the 36 keys on the keyboard of an Apple MacBook Pro laptop 25 times. This information is then fed into an AI program, which is able to identify the vocal patterns of each key.

A participant takes a photo of the artificial robot Sophia at the Global Summit on Artificial Intelligence for Good in Geneva, Switzerland, July 7, 2023.

They then placed an iPhone 17cm away from the same Apple laptop to record someone typing. They were able to deduce the text with 95 percent accuracy. When they recorded with the Zoom conferencing software, the accuracy dropped to 93 percent.

"Each key produces a unique sound, and this sound can be recorded and used to infer which key is being pressed," said Ihsan Torreni of the University of Surrey's Cyber Security Centre, one of the study's authors.

"We're using the most advanced models available, and it gives you a sense of the tremendous improvement in the accuracy of AI models over the last five years, from about 70 percent accuracy to near-perfect accuracy," Torreini said.

This means that the techniques used to carry out "side channel" attacks are now widespread.

"Side channel" attacks are those that attempt to steal signals from communications devices and may exploit electromagnetic waves, acoustics, and power consumption. Torreni said Apple might consider adding random noise to keyboard tapping to block such attacks.

The researchers also said that the target's smartwatch may have been compromised and then used to record keyboard tapping. Scientists have previously shown that it is possible to identify typing by analysing wrist movements recorded by smartwatches with 93.75 per cent accuracy.