HomeThe New IntelligenceRadar Can Now Eavesdrop on Your Phone Calls

Radar Can Now Eavesdrop on Your Phone Calls

Penn State researchers use millimeter-wave radar to transcribe phone calls by detecting earpiece vibrations from three meters away, reaching 60% accuracy with AI.

Share
The New Intelligence · Explore this series
August 9, 2025
Key Takeaways
  • Penn State radar transcribes phone calls from three meters away.
  • The system adapts OpenAI's Whisper model using just 1% of parameters.
  • At 60% accuracy, partial transcripts still reveal conversation meaning.

Suryoday Basak, a computer science doctoral candidate at Penn State, pointed a radar sensor at a phone and listened. Not to the sound, but to the vibrations.

His team has demonstrated that phone call interception is possible through millimeter-wave radar that can remotely transcribe phone conversations. It detects the tiny vibrations produced by a phone's earpiece from up to three meters away.

They call it "wireless-tapping." The name is deliberately unsettling.

The Vibrations Your Phone Cannot Hide

Every phone earpiece vibrates during a call. The movements are imperceptible to the person holding the device, but they ripple across the phone's surface in patterns unique to each spoken word.

Basak and his advisor, associate professor Mahanth Gowda, captured these vibrations using off-the-shelf mmWave radar sensors. The same sensors already sit inside self-driving cars and 5G network equipment.

Key figure

60%

Word accuracy rate when transcribing conversations with a vocabulary of 10,000 words

The system reached approximately 40-60% accuracy across a vocabulary of 10,000 words. That represents a remarkable leap from their 2022 prototype, which managed only 10 predefined words at 83% accuracy.

A jump from word recognition to full transcription in three years is, by any measure, striking.

OpenAI's Whisper Learns to Read Radar

The cleverness lies not in the radar hardware. It lies in repurposing an AI speech model for a task it was never designed to handle.

Basak's team adapted OpenAI's Whisper, an open-source speech recognition model, using a technique called low-rank adaptation. They retrained just 1% of the model's parameters, specializing it for radar-derived audio.

That economy is notable. One percent of a model, carefully chosen, turned clean-audio software into a radar interpreter.

[...] Using context clues, we can determine whole conversations.

Suryoday Basak, Penn State

The results, published in the proceedings of WiSec 2025, show the system handles continuous speech rather than isolated words. With contextual knowledge of the conversation topic, accuracy climbs higher still.

Sixty Percent Accuracy Is More Dangerous Than It Sounds

A 40-60% transcription rate might seem too low to be threatening. Basak's team draws a pointed comparison: lip-reading typically captures only 30 to 40 percent of words, yet skilled observers reconstruct full conversations from context.

Radar eavesdropping follows the same logic. Partial transcripts, combined with knowledge of the speaker or subject, can yield surprisingly complete intelligence from phone call interception.

What is mmWave radar?

Millimeter-wave radar operates at frequencies between 30 and 300 GHz. These high frequencies detect extremely small movements, originally designed for automotive collision avoidance and gesture recognition. The same precision that helps a car sense a pedestrian can, it turns out, sense a phone vibrating during a call.

The privacy implications are genuinely troubling.

Gowda notes that mmWave radar sensors continue to shrink. Future versions could conceivably nestle into everyday objects: furniture, office equipment, public infrastructure. The risks AI poses to personal privacy keep expanding in unexpected directions.

The Researchers Want You Worried

Basak and Gowda frame their work as a warning, not a weapon. The research was funded by the U.S. National Science Foundation, and the team published openly to prompt awareness.

More On Surveillance

Super Recognizers Detect Fake Passport Photos Others Miss

Super recognizers outperform trained border officers at detecting morphed passport photos. New research shows their visual strategies could reshape security training.

"By understanding what is possible, we can help the public be more mindful during sensitive calls," Basak explained.

The field is accelerating now. A separate research group recently demonstrated EchoLLM, which targets bone conduction headphones using similar mmWave techniques. Each new variant widens the attack surface.

The three-meter range and imperfect accuracy currently limit practical exploitation. But radar sensors are cheap, and AI models improve steadily.

The quiet assumption that a private call stays private deserves fresh scrutiny.


Sources

Fact Check: Claim-by-Claim Verification Verified

The article accurately reports the Penn State researchers' findings on mmWave radar-based phone call transcription, matching primary sources on key claims, names, accuracy figures, and methodology.

1 Verified
Penn State team led by Suryoday Basak and Mahanth Gowda used off-the-shelf mmWave radar (77-81 GHz) to detect earpiece vibrations up to 3 meters away
2 Verified
System adapts OpenAI's Whisper via low-rank adaptation (1% parameters) for radar data, achieving ~40-60% word accuracy on 10,000-word vocabulary
3 Verified
Builds on 2022 prototype (10 words at 83% accuracy); published in WiSec 2025 proceedings; funded by NSF; framed as privacy warning
4 Verified
Quote "Using context clues, we can determine whole conversations" directly from Basak in Penn State press release

Commentary

  • Peak 60% accuracy likely at closer ranges (~50cm); drops at max 3m range (e.g., 2-4% word-level per secondary reports), though article appropriately hedges as "approximately 40-60%".
  • arXiv preprint (2024) titled "mmWave-Whisper"; final WiSec 2025 paper likely "Wireless-Tap" per proceedings—minor title variance, but content aligns.
  • Lip-reading comparison (30-40% words) and mmWave explanation are consistent with researchers' statements.

Sources used for verification

Academic/Peer-reviewed:

Other reliable sources:

Share
Related Articles
AI Consciousness Is Unlikely, Says Neuroscientist Anil Seth

Neuroscientist Anil Seth argues AI consciousness is unlikely without biology. His TED talk lands amid a widening debate over conscious AI, not intuition.

AI In Science Connects the Dots, But Only In Fields That Are Fragmented

An analysis of 80 million papers shows AI boosts originality where knowledge is scattered and connections are weak, but contributes little novelty in structured science.

"Keep Humanity Safe From AI," Urges Pope Leo XIV

Pope Leo XIV's first encyclical reaches the same verdict on AI as the labs building it, then parts ways over the meaning of human limits.

AI Solves Erdős Math Problem: What's Next for AI in Mathematics?

An AI solved an 80-year-old Erdős math problem by walking a path mathematicians had collectively avoided.