The next time you need to have a truly private conversation over the phone, you'll need to account for a whole new way by which people can snoop on you. Researchers at Pennsylvania State University have discovered that the vibrations of the earpiece inside your phone – the one that's active when you're holding the handset to your ear – can be picked up by a radar sensor from a distance, converted into audio, and transcribed using AI.
Thankfully, this new technique, called WirelessTap, isn't yet perfect. It currently manages a maximum word recognition accuracy of about 60%, works up to a range of 10 ft (300 cm), and the accuracy decreases even within that range. But the tech gives us a glimpse into the future of spycraft and surveillance, and shows how AI can prove essential in these fields.
The WirelessTap system relies on millimeter wave (mmWave) sensing tech, where radio frequencies in the 77-81 GHz range can be picked up by radar sensors for a variety of applications including motion detection inside a house or a car, and for helping self-driving vehicles spot obstacles around them. The researchers' work showed how a commercially available mmWave radar sensor can be used to pick up the practically imperceptible vibrations of a phone's earpiece, and the resulting vibrations of the phone itself, from a few feet away.
"If we capture these same vibrations using remote radars and bring in machine learning to help us learn what is being said, using context clues, we can determine whole conversations," explained computer scientist Suryoday Basak, who authored the paper he published with his colleague in Proceedings of WiSec 2025 this June. "By understanding what is possible, we can help the public be aware of the potential risks."
The duo turned the noisy data captured by the sensor into recognizable speech using a customized version of OpenAI's Whisper speech recognition and transcription model. While WirelessTap demonstrated nearly 60% accuracy at a close range of about 1.5 ft away (50 cm), the researchers note this could be improved by incorporating elements of the context of the conversation in question – similar to how lip readers can more accurately interpret what's being said by people they're observing when they know what their targets might be talking about.
The team notes this system goes beyond other radar-based eavesdropping technologies, which have either relied on the victim's use of a loudspeaker, or have been restricted to discerning audio using a limited vocabulary. WirelessTap is therefore a lot more discreet.
The mmWave sensor can be miniaturized to fit into small commonplace objects like pens. Combined with contextual background about the audio being captured, WirelessTap could be a handy tool for espionage operations, or other sneaky ways to steal private information.
"The goal of our work was to explore whether these tools could potentially be used by bad actors to eavesdrop on phone conversations from a distance," Basak concluded. "Our findings suggest that this is technically feasible under certain conditions, and we hope this raises public awareness so people can be more mindful during sensitive calls."
Source: Pennsylvania State University
