AI Strikes Again: Researchers Train Learning Model To Steal Data From Keyboard Sound – News18

AI Strikes Again: Researchers Train Learning Model To Steal Data From Keyboard Sound – News18

[ad_1]

Last Updated: August 07, 2023, 17:58 IST

London, United Kingdom (UK)

Keyboard strokes can be used to get a lot of data which the AI can help hackers

Keyboard strokes can be used to get a lot of data which the AI can help hackers

AI is becoming a concern for various reasons and the last thing you want is for the tech to study your keystrokes and help hackers steal money.

AI is becoming a serious concern for researchers, but it continues to fascinate everyone with its potential and one such instance has come up recently. According to reports, researchers have managed to train a deep learning model to detect the pattern of your typing to steal data, and its accuracy is rated at 95 percent.

These researchers from a British university, quoted in its report by BleepingComputer, have managed the AI model to recognise the sound to decipher the keystrokes from the microphone of the laptop or PC. The researchers have used the algorithm on apps like Zoom and Skype, where the accuracy rate goes down to 93 percent and 91.7 per cent, respectively.

Now, the first thing that comes to mind with these numbers is the prospect of AI helping hackers to decipher keystrokes while you’re logging into a banking account and even get hold of the password through the sound algorithm.

And the worrying part is that researchers could inadvertently aid such advancements for hackers, helping them work around varied security measures in place. Such instances advocate for virtual keyboards, so that your keystrokes cannot be deciphered, giving your digital footprint better security.

AI To Detect Keystrokes – How It Was Tested

The researchers had a clear plan to test this sound-based algorithm and the process has been detailed. The team basically pressed 36 keys on a MacBook, 25 times each to record the sound produced by the keys, helping them capture valuable data to train the AI model. They used iPhone 13 Mini to get the audio from the keystrokes, kept 17 cm away from the MacBook. They used the sound data to train an image classifier called CoAtNet that was able to predict the keys that were pressed based on the sound recorded.

Tests like these ascertain the threat posed by AI and the last thing you want is AI listening to your keystrokes and helping hackers access bank accounts to steal money.

The researchers advise that people can avoid such mishaps by tweaking their typing patterns or using random complex passwords. It is also possible that a keyboard which makes less typing sound could be a safer option in the near future.

[ad_2]

Source link

Leave a Reply