Your phone’s smallest vibrations may be giving away your conversations — and AI can now decode them.Computer scientists at Penn State have developed a system that uses millimeter-wave radar and an AI speech recognition model to capture and transcribe phone calls from up to 10 feet away. By detecting tiny vibrations from a phone’s earpiece, the setup achieved about 60% accuracy over a vocabulary of 10,000 words.

Figure 1. AI-Powered Radar.

The research builds on the team’s 2022 project, which recognized only 10 predefined words with up to 83% accuracy. The new version tackles continuous speech, though accuracy drops due to noisy radar data.

“When we talk on a cellphone, we overlook the vibrations traveling through the earpiece,” explained lead author Suryoday Basak. “If radar can pick them up and AI can interpret them, it’s possible to reconstruct entire conversations.” Figure 1 represents AI-Powered Radar.

The system uses the same radar technology found in self-driving cars and 5G networks. To process the low-quality data, the researchers adapted Whisper, an open-source AI speech recognition model, retraining just 1% of its parameters for radar inputs.

While far from perfect, the results show that even partial transcripts can expose sensitive information, much like lip-reading. “Picking up keywords is often enough in a security context,” noted co-author Mahanth Gowda.

The findings highlight the privacy risks of emerging sensing technologies, showing that eavesdropping is technically possible under certain conditions [1]. The researchers, funded by the U.S. National Science Foundation, hope their work raises awareness and spurs the creation of safeguards to protect phone conversations from remote interception.

reference:

1. https://interestingengineering.com/culture/ai-radar-phone-call-privacy-threat

Cite this article:

Keerthana S (2025), AI-Powered Radar Can Eavesdrop on Phone Calls from 10 Feet Away, Raising Fresh Privacy Concerns, AnaTechMaz, pp.765