Spoofing Eavesdroppers with Audio Misinformation
Zhambyl Shaikhanov, Mahmoud Al-Madi, Hou-Tong Chen, Chun-Chieh Chang, Sadhvikas Addamane, Daniel M. Mittleman, Edward Knightly
University of Maryland - College Park
Rice University
Los Alamos National Laboratory
Sandia National Laboratories
Brown University
IEEE S&P (Oakland) 2025: Proceedings of the IEEE Symposium on Security and Privacy
Wireless eavesdropping on phone conversations has become a major security and safety concern, especially with advancements toward 5G and beyond featuring higher frequencies and higher sensing resolution.
As demonstrated recently, the eavesdropper can direct a millimeter-wave radar system towards the victim, Alice, who is talking on her phone, to detect few-micron-scale vibration patterns. With these tiny displacements being strongly correlated with the generated sound waves, the eavesdropper recovers audio conversation, intercepting sensitive information, all without the victim ever noticing.
Unfortunately, this analog micro-information cannot be encrypted or protected by existing methods, as the eavesdropper deliberately exploits the physical propagation of sound to steal information during interactions between users and the analog world. As such, the acoustic signal is the weakest link in the communication chain of a phone call, making it very challenging to protect with existing methods.
We present a new class of proactive defensive architecture that not only thwarts such attacks but also enables Alice to counter-attack by spoofing eavesdroppers with audio misinformation.
Our system strategically modifies the eavesdropper's fundamental sensing observations, concealing native signals while encoding alternate synthetic data. It incorporates a low-profile, reconfigurable metasurface and double-inference principles to dynamically generate artificial audio-vibration signatures, injecting deceptive misinformation. This new approach transforms defensive measures from merely reactive to proactively deceptive, giving the defender an advantage and the capability to delude attackers into trusting false information.
This proposed architecture provides significant strategic and security advantages across various scenarios. For instance, attackers could be misled into intercepting fake business contracts and negotiations (corporate), false bank account details and social security numbers (civilian), or incorrect coordinates and mission details (military).
Audio Samples from Experiments Demonstrating the Attack and Countermeasure
Original audio played on smartphone speaker
Eavesdropper recovers audio by remote vibration sensing via radar
Amazon Transcribe: Speech-to-Text
“Hi, I’m James Kelly. Sure. My bank account is 1 2 3 5 6 7. My social security number is 8 9 10”
Alice can scramble the eavesdropper's inference
Amazon Transcribe: Speech-to-Text
Alice declares victory, spoofing the eavesdropper with an altered name, band account details, and SSN
Amazon Transcribe: Speech-to-Text
“Hi, I’m John Wick. Sure. My bank account is 7 7 3 8 0 1. My social security number is 3 5 9”