According to Cornell University researchers, Keystrokes are now being leveraged by AI tools to steal the passwords of individuals leaving their banking and personal information exposed.

In a new research paper, researchers describe how an AI assault can correctly guess passwords with around 95% accuracy by observing what individuals type on keyboards.

By training an AI model to recognise the sound of keystrokes and using it on a nearby phone, researchers found AI consistently uncovered passwords with the MacBook Pro having the highest accuracy.

This was the greatest accuracy researchers have observed not including the application of a large language model.

In real life, an attack would materialise as malware installed on phones or a nearby device with a microphone where it would get data from keystrokes and feed them into an AI model by listening on a user’s microphone.

In the experiment, the researchers employed CoAtNet, an AI image classifier, and instructed the model on 36 keystrokes on a MacBook Pro pressed 25 times each.

For consumers looking to safeguard their devices, the Bleeping Computer suggests avoiding typing your password in at all by using features like Windows Hello and Touch ID.

A good password manager also allows for more protection, but a new keyboard won’t make a difference.

This new potential threat is the latest in a long list of attacks that can be facilitated by AI tools with even the FBI has cautioned about the threats of ChatGPT being leveraged for criminal activities.

Recent research has seen an escalation in these attacks and warns about the effects of adaptive malware that can rapidly transform with the use of AI tools like ChatGPT.