Injected and Leaked: Actively Inducing Side-Channel Leakage Using Electromagnetic Injection and Hardware Nonlinearity

Scientific Impact Assessment

Core Contribution

This paper introduces the concept of Injection-Induced EM Side Channels, unifying two historically distinct branches of electromagnetic security: EM *injection* (an integrity/availability threat) and EM *side-channel leakage* (a confidentiality threat). The central insight is elegant and non-obvious: the same hardware nonlinearity (in amplifiers, ADCs, MOSFETs, power converters) that lets adversaries *demodulate* injected carriers into false analog inputs can also *modulate* internal secret analog signals onto an externally injected carrier, upconverting otherwise-unradiatable low-frequency secrets (10 Hz–10 kHz) into measurable EM emissions. This dissolves the long-standing "frequency mismatch" barrier that made low-frequency analog secrets (audio, power traces, actuator control) effectively unreachable by passive eavesdroppers. The concrete instantiation, InjectEave, demonstrates through-wall audio eavesdropping up to 30 m, smart-home activity inference, and — most strikingly — a closed-loop "eavesdrop–synthesize–inject" manipulation chain against a landline phone.

Methodological Rigor

The work is methodologically comprehensive. The theoretical Injection–Modulation–Emission model derives leakage amplitude from a series-expansion of nonlinear transfer functions, yielding testable predictions (Eq. 7) about near-linear scaling with injection amplitude and secret amplitude. These predictions are validated empirically. Importantly, the authors include appropriate controls: passive-only conditions produce no correlated leakage, and replacing nonlinear components with linear resistors eliminates the effect — directly isolating nonlinearity as the mechanism. The evaluation spans 4 component classes, 11 COTS devices, and systematic environmental characterization (distance, antenna angle, barrier material), plus in-the-wild case studies with interference and stealthiness measurements. This is a strong, well-controlled experimental design. Minor gaps: audio-quality evaluation relies on a limited set of metrics (SNR, STOI, WER on synthesized TTS speech rather than diverse natural speech), and the diffusion-based enhancement is trained purely on simulated data with cross-device evaluation that, while promising, is somewhat narrow.

Potential Impact

The impact potential is high. The paper defines a new attack surface that manufacturers and the hardware-security community have essentially ignored, explicitly puncturing the assumption that low-frequency analog interfaces are "safe" from EM leakage. It is likely to seed a subfield of follow-up work characterizing other susceptible interfaces (the authors themselves preview microphone inputs and internal communication lines) and defensive research (twisted-pair mitigation, active RF anomaly detection). The closed-loop manipulation capability — moving from passive confidentiality compromise to active content falsification — is conceptually significant and likely to be cited as a demarcation point. Published at USENIX Security '26, a top venue, it will reach the relevant audience.

Timeliness & Relevance

Highly timely. It sits at the confluence of two mature but siloed literatures (EM injection à la Foo Kune, and TEMPEST-style EM leakage) and adjacent recent work on active/backscatter side channels (Echo Tempest, LeakyOhm, backscatter serial eavesdropping). The authors carefully differentiate their contribution: prior active approaches used binary impedance-change models limited to digital data, whereas this work provides the first waveform-level nonlinear model for continuous analog secrets. The integration of modern generative speech enhancement is also current.

Strengths & Limitations

Strengths: (1) A genuinely new conceptual framing that reframes an assumed-hard problem; (2) tight coupling of theory, component-level feasibility, device-level evaluation, and real-world case studies; (3) clean mechanistic controls; (4) impressive quantitative results (30 m range, through-wall, closed-loop manipulation); (5) thoughtful ethics and mitigation discussion.

Limitations: (1) Real-world ranges are modest for most devices (2–6 m default; 30 m only with a $415 amplifier and best-case devices); (2) microphone-input eavesdropping is limited to ~30 cm, revealing the amplitude-dependence constraint; (3) reproducibility is intentionally hampered — vulnerable injection frequencies and active-injection control logic are withheld for ethical reasons (defensible, but it raises the barrier to independent replication); (4) the speech-enhancement generalization claims rest on simulation-trained models with limited real-device validation; (5) mitigations remain preliminary (twisted-pair is one contributing factor, not a complete defense).

Additional Observations

The work is a building-block/foundational contribution by intent — the authors explicitly frame it as laying groundwork for a research program, and the Injection-Modulation-Emission model is reusable. Resource requirements are moderate (SDR, spectrum analyzer, antennas, optional power amp) — accessible to a competent RF/security lab. The refutation value is meaningful: it directly contests the "false sense of security" around low-frequency analog leakage. Interdisciplinary reach is real but concentrated in hardware security + RF/EM engineering, with a lighter touch of ML for signal recovery.

Overall, this is a high-quality, conceptually original security paper likely to influence both offensive and defensive research on analog-interface EM security.