The creators of "smart" consumer electronics, most likely, do not even expect that their many brainchildren will one day become easy prey for hackers who will be able to turn them into espionage tools.
A group of researchers, with the participation of Nirupam Roy, assistant professor of computer science at the University of Maryland, managed to hack the navigation laser system of the robot vacuum cleaner. Then they processed the information extracted from it with the help of AI, after which they restored the speech of people who were next to the robot and calculated the programs of the TV operating in this room.
Scientists have found that with any device that uses detection and distance technology (we are talking about lidar), it is possible to capture sound - even without using a microphone. At the same time, the lidar is an important part of the navigation system - it helps the robot vacuum cleaner to avoid obstacles during operation.
In the course of the study, information security experts expressed the opinion that the navigation maps of such vacuum cleaners are often stored in the cloud, which makes this information vulnerable to intruders - for example, they can be used to roughly estimate the size of the house and the income level of the owners of the robot.
Roy and his team tried to find out if the lidar is capable of carrying a potential threat as a listening device. Sound waves are known to cause weak vibration in nearby objects. In turn, these vibrations contribute to minor changes in the light reflected from objects.
In the 1940s, intelligence agencies began using laser microphones capable of converting such vibrations into sound. However, for their operation, a directed laser beam is required, reflected from flat surfaces, in particular, from window panes.
By hacking into the vacuum cleaner's on-board computer, the engineers demonstrated the ability to manipulate the laser beam and send the received information to their PCs via Wi-Fi, without creating problems for the navigation system of the device itself.
The researchers ran the received signals through deep learning algorithms that were trained to either recognize human voices or identify musical sequences from TV shows. The computer system, dubbed LidarPhone, was able to identify and correlate fragments of the conversation with an accuracy of 90%. She also identified the TV show from the minute recording with the same accuracy.