AI Data Theft: Researchers Use Machine Learning To Extract Information From Keyboard Typing Sounds
Artificial intelligence is becoming a serious concern for researchers, but it continues to fascinate everyone with its potential, and one such case has come to light recently. According to reports, researchers have managed to train a deep learning model that detects the pattern you write to steal data with 95 percent accuracy.
These researchers from a British university, cited by BleepingComputer in their report, have succeeded in using an AI model to recognize sound and interpret keystrokes from a laptop or PC microphone. Researchers have used the algorithm in applications such as Zoom and Skype, where the accuracy rate drops to 93 percent and 91.7 percent, respectively.
The first thing that comes to mind about these numbers is the possibility of AI helping hackers decipher keystrokes when you log into a bank account and even get hold of your password using a voice algorithm.
And most worryingly, researchers may inadvertently help hackers to advance such advances by helping them circumvent various Such cases favor virtual keyboards so that your keystrokes cannot be deciphered, thus improving your digital footprint.
Artificial intelligence for keystroke detection – how it was tested
The researchers had a clear plan for testing this voice-based algorithm and the process has been detailed. The team basically pressed 36 keys on the MacBook, 25 times each to record the sound the keys produced, which helped them capture valuable information to train the AI model. They used an iPhone 13 Mini to get the sound of the keystrokes and was held 17 cm away from the MacBook. They used the audio data to train an image classifier called CoAtNet, which could predict keystrokes based on recorded audio.
Tests like these verify the threat posed by AI, and the last thing you want is an AI listening to your keystrokes and helping hackers get into your bank accounts to steal money.
Researchers advise that people can avoid such mishaps by adjusting their spelling or using random, complex passwords. It’s also possible that a keyboard that produces less typing noise could be a safer option in the near future.
