WiFireNet works by analysing channel state information, the pattern of how an object distorts and reflects ambient Wi-Fi signals based on its shape, orientation and material. The full setup needs a Wi-Fi transmitter, a receiver and the AI itself. To train it, the team recorded video and reflected Wi-Fi signals of people carrying objects from phones to handguns, then used the footage to match each concealed weapon with its wireless signature. Training deliberately included a plastic handgun, a metal water bottle and a phone so the model could tell a real weapon apart from everyday items of similar shape or material.

The results are promising but uneven. In controlled trials tracking up to four moving people in a small indoor space, the system achieved an average handgun-detection precision of 72.5%. Under stricter requirements, such as pinpointing exactly where on the body the weapon sits, precision fell to 53.8%, and the network struggles in indoor layouts it has not seen before. The prototype runs on a modified implementation of the 1999-era 802.11a standard as proof of concept, with planned phases moving to Wi-Fi 7 and Wi-Fi 8 networks, whose higher-resolution channel data should improve detection.

The technique also has a broader shadow. Security researchers have previously shown that ordinary Wi-Fi routers can identify people with 99.5% accuracy, and critics argue that Wi-Fi sensing's ubiquity and the lack of regulatory frameworks raise serious privacy concerns. A tool built to spot guns and a tool built to track anyone in a room rest on the same underlying principle.