A groundbreaking AI-based tool has emerged with the power to deduce passwords solely from the auditory cues of keystrokes, boasting an accuracy rate close to 90%. This innovation emphasizes the growing risk associated with typing passwords during online video meetings such as those on Zoom.

The cutting-edge research, a collaboration between experts from Durham, Surrey, and Royal Holloway universities, highlights the newfound potential threat. Their findings suggest that with the surge in video conferencing usage, along with the ubiquitous presence of microphone-equipped devices, there’s an escalated risk of audio-based cyber threats.

The research paper, released on August 3, elaborates on the AI tool’s prowess in decoding text—most notably, passwords—by just “listening” to the sounds of keystrokes. Their tool notched up an impressive accuracy rate that surpassed 90% when put to the test.

The success of the AI’s keen auditory discernment can be attributed to the meticulous training it underwent. The researchers, as part of their methodology, recorded sounds of each of the 36 keys pressed on a MacBook Pro 25 times. This collection of sounds was then processed by the AI, enabling it to detect distinct patterns for every keystroke.

In a remarkable demonstration of its capability, the AI showcased a 93% accuracy in deducing keystrokes from recordings gathered during Zoom sessions on a MacBook Pro. The accuracy saw a further boost, reaching 95%, when the tool was tested with keystrokes recorded using an iPhone 13 mini.

In light of these findings, users are advised to exercise caution when typing passwords during live online sessions.