There are technologies that seem taken from science fiction not because they are futuristic, but because they turn something everyday into a knife: it can be a tool, but also a weapon. And Wifi technology belongs to that category. We use it to connect mobile phones, computers or televisions, but it actually does something deeper: It continually fills the space with invisible waves that bounce off walls, furniture… and also against us.
Now, a team from the Karlsruhe Institute of Technology (KIT) in Germany has shown to what extent these signals can reveal information about people. According to a study published in Conference on Computer and Communications, a system based on conventional WiFi networks and artificial intelligence models are capable of identifying people with an accuracy close to 99%. And the most striking thing is that you don’t need to carry a cell phone with you, or be connected to the network, or even have the phone on.
“By observing the propagation of radio waves we can create an image of the environment and the people present – explains Thorsten Strufe, cybersecurity expert at KIT and one of the authors of the work, in a statement -. It works in a similar way to a normal camera, with the difference that we use radio waves instead of light”.
The idea may seem strange, but it has some physical logic. WiFi signals do not pass through a room cleanly: they constantly bounce around and change slightly when encountering objects and human bodies. Each person alters these waves in a different way depending on their size, posture, way of walking or even the proportion and distribution of water and muscles in the body. The result is a kind of “footprint”, invisible, but also unmistakable.
Until now, many experimental systems of this type required specialized sensors or analyzed complex technical data called CSI (Channel State Information). But Strufe’s team has used something much more common: data from beamforming, a standard technology present in modern WiFi routers since the arrival of WiFi 5. Basically this technology, known as beamforming, processes signals that allow wireless antennas to concentrate and direct their signal towards a specific device. And there appears the detail that most worries the authors: that information It is normally transmitted unencrypted and can be captured by nearby devices.
“This technology turns every router into a potential surveillance system – adds Julian Todt, co-author of the study -. If you regularly walk past a coffee shop with WiFi, you could be inadvertently identified and recognized later.”
To see how far the system could go, Strufe’s team conducted tests with 197 participants. The AI model managed to recognize them with an accuracy close to 100%, regardless of the angle from which they were detected or how they walked. Once trained, the system needed just a few seconds to identify a specific person.
Although it may sound new, the idea of “seeing” using WiFi has been explored for years. There are already systems capable of detecting human presence behind walls, monitor breathing or recognize gestures using wireless waves. Other recent projects, such as WhoFideveloped in Italy, had also shown that WiFi disturbances can be used as a kind of alternative biometrics, with accuracies of over 95%.
The difference is that the technology developed by the Strufe team is beginning to leave the specialized laboratory. WiFi networks are practically everywhere: homes, offices, airports, hotels, hospitals, stores or public transportation. Unlike cameras, they are also invisible. They don’t point at anyone. They don’t look like surveillance systems, but they could act like them.
The authors insist that the technology still has limitations and that it is not equivalent to magically “reading” identities. The system requires prior training and works best in controlled environments. Even so, they believe that the problem is no longer technical, but political and legal. That’s why they’re calling for future WiFi standards, like the IEEE 802.11bf standard, incorporate specific protection mechanisms to avoid this type of passive monitoring.