This AI recognizes your password based on how you type

Be careful, your cell phone is listening. (Image: Sudarshana, Makism – adobe.stock.com)

Tests and studies have already proven that AI can make our lives easier in many areas. She helps us decipher old stone tablets, saves us money for a photo shoot or manages entire companies without further ado.

British researchers at Cornell University have managed to develop an AI that can distinguish keystrokes based on their sound.

How does this work?

Imagine the following scenario: You sit at your computer, log into your email inbox to write a message, then log in to Thalia because you want to order a book as a Christmas present.

What you don’t realize: Through your cell phone’s microphones, an AI learns your personal information by listening and reading the sound of typing.

It sounds like something out of a spy thriller, but thanks to researchers at Cornell University, it’s already easily possible today.

You can read here how a well-known sports newspaper misused AI for its own purposes:

A feasibility study

The AI, called CoAtNet, was trained to recognize and differentiate unique sounds between each key.

And she did it with a 95 percent success rate. For the test, a smartphone was placed 20 centimeters away from an Apple MacBook.

The artificial intelligence even read out the attacks via Zoom and Skype. (Image: Cornell University)

Dr. Ehsan Toreini, one of the co-authors of the study, assumes that the accuracy of such AI models and corresponding attacks will increase in the future.

Reason for concern?

Before you throw all technical devices out of the house: There are some limitations of the model. CoAtNet needs to be tailored to each keyboard type as the sounds vary from keyboard to keyboard.

During training, each of the 36 keys on a MacBook was pressed 25 times in a row with different fingers and different pressure levels. The AI ​​also confuses upper and lower case letters, numbers and symbols.

Interesting for keyboard modders: By changing the sound with switches, the AI ​​suddenly becomes useless. So if you use a keyboard you put together yourself, you’ll be fine.

This particularly pleases keyboard aficionados like Nils.

The study serves primarily as a proof of concept and was not used for actual password cracking or in real-world environments. It is intended to increase awareness that such attacks are now theoretically possible.

A popular fast food chain is considering using AI to make ordering burgers even easier.

An AI that reads passwords based on the sound of keystrokes sounds like a thing of the future and quite dangerous. Fortunately, the study is only for feasibility purposes. How careful are you with your passwords? How do you think AI will play a part in our everyday lives in the future? Feel free to write it in the comments.