WiFi Can See You (coffee.link)

Researchers have shown that everyday WiFi signals can be turned into surprisingly detailed images and rich behavioral telemetry. A June 2025 arXiv paper introduces LatentCSI, a system that maps Channel State Information (CSI) — the radio-wave fingerprint created by reflections off walls, furniture, and bodies — directly into the latent space of Stable Diffusion to generate 512×512 images of people’s positions, orientations and poses. By avoiding classical GAN pipelines, LatentCSI trains about three times faster and produces higher-quality reconstructions that are robust in low-light and non-line-of-sight conditions; generated images can even be restyled with text prompts while preserving true pose. The authors position the technique as a camera-free alternative for uses like elder-care monitoring and security, but intentionally blurring faces does not eliminate privacy risks. LatentCSI sits atop a rapidly maturing field: WiFi sensing already achieves near-perfect activity recognition, fine-grained gesture and breathing monitoring, keystroke inference, gait and device fingerprinting, and through-wall localization. Lightweight models (WiFlexFormer), public datasets (WiMANS), self-supervised methods (CAPC) and LLM integrations (Wi‑Chat) are lowering deployment barriers, and vendors have begun embedding sensing into consumer routers. That combination—powerful inference, invisibility of radio sensing, and little regulatory oversight (IEEE 802.11bf excludes privacy)—creates a privacy paradox. Technical mitigations like “cover signals” exist, but the community now faces urgent choices about consent, transparency and legal guardrails to prevent silent surveillance even as useful applications proliferate.