UK expert: Artificial intelligence can steal passwords through keyboard sounds | Durham University | Keyboard | Password
Reference News recently published an article on The Times website titled "Artificial Intelligence Can Steal Passwords by Listening to the Sound of Your Keyboard". The summary of the report is as follows:
Experts from Durham University, Surrey University, and Royal Holloway College London pressed each of the 36 keys on an Apple MacBook Pro laptop keyboard 25 times and recorded the sound. Then, this information is input into the artificial intelligence program, so that the latter can recognize the sound pattern of each key.
Then, they placed an iPhone 17 centimeters away from the same Apple laptop to record someone typing. They successfully inferred the typing content with an accuracy of 95%. When they recorded using Zoom conference software, the accuracy dropped to 93%.
"Each key emits a unique sound, which can be recorded to infer which key is being pressed," said one of the authors of the research paper, Ihsan Toreni from the Cybersecurity Center at the University of Surrey
Toreni said, "We are using the most advanced model currently available, which allows you to experience the tremendous progress that artificial intelligence models have made in accuracy over the past five years. This progress has led to an increase in accuracy from around 70% to near perfection."
This means that the technology used to implement "edge channel" attacks is now widespread.
Edge channel attacks refer to attacks that attempt to steal signals from communication devices and may exploit electromagnetic waves, acoustics, and power consumption. Torreni said that Apple may consider adding random noise to keyboard typing to prevent such attacks.
The researchers also said that the smartwatch of the target could be breached and then used to record keyboard tapping sounds. Scientists have previously demonstrated that typing content can be identified by analyzing wrist movements recorded by smartwatches, with an accuracy rate of 93.75%.