Evaluating Wi-Fi Sensing Attack Vectors: AI-Driven Covert Threat Detection and Deterministic Mitigation in Mobile and Edge Ecosystems

Wi-Fi was designed to move data. Paired with modern deep learning, it has quietly become something else: a sensing fabric that detects motion, infers behavior, and reconstructs physical activity from signal reflections alone, using hardware cheaper than a home router. Enterprise and IoT security was never built for this pairing, creating a significant, largely invisible attack surface. This paper presents a rigorous empirical evaluation of covert Wi-Fi sensing attacks in realistic, interference-prone IoT environments, built around a bidirectional LSTM trained to correlate Channel State Information (CSI) sequences with keystroke signatures. A passive adversary with only commodity hardware and this learned model can extract finegrained keystroke intelligence, including PIN entries, without network authentication, physical access, or a single injected packet. The attack reaches an F₁ score of 85.4% in enterprise line-of-sight conditions. It still reaches an F₁ of 61.2% through a standard 15 cm drywall partition — over six times the 10% floor of a ten-class random guess. To neutralize this threat at its source, we introduce RF-Guard: a deterministic, stateful CSI obfuscation framework at the firmware level that gives the attacking model nothing stable to learn from. RF-Guard collapses adversarial accuracy to 9.8% across three IoT testbeds, with gray-box retraining recovering only to 14.3% F₁; a fully adaptive neural adversary remains untested and is named as future work. It adds under 4.2% throughput overhead on a standard ESP32 microcontroller during a continuous 24-hour stress test. No hardware replacement is required.

No artifact released. This paper reports measurements, but its code, corpus and raw outputs are not public, so the numbers above cannot be independently checked. Treat the results as unreproduced.

@misc{poudel2026evaluatingsensing,
  title  = {Evaluating Wi-Fi Sensing Attack Vectors: AI-Driven Covert Threat Detection and Deterministic Mitigation in Mobile and Edge Ecosystems},
  author = {Sushant Poudel},
  year   = {2026},
  note   = {Manuscript},
  url    = {https://sushantpoudel2028.com.np/papers/wifi-sensing.pdf}
}

Claims about his own work that the ledger covers are re-checked weekly against their public source by sushant-me/reputation — 27 claims today, 19 checked live. Coverage is stated rather than implied: it does not yet extend to every manuscript listed here. The reproducibility note above is part of that record: where no artifact exists, the page says so instead of implying one.