Dejan Strbad - Silent Receipts, Loud Signals: Offensive and Defensive ML for Messaging App Surveillance
Silent delivery receipts in WhatsApp and Signal can be exploited to infer a target's online status, device count, and operating system, all without generating any notification.
Building on the "Careless Whisper" research (https://arxiv.org/abs/2411.11194), we reproduce and extend this attack using practical tooling built with AI agents and custom ML pipelines, pushing the boundaries on both the offensive and defensive side.
On the offensive side, we train a classification model on delivery receipt timing patterns to infer a user's physical context (home, work, commuting, sleeping). We further explore whether timing correlation of receipt patterns can reveal co-location, determining if two target phone numbers share the same network.
On the defensive side, we build a local network monitor that uses traffic analysis to detect when delivery receipt probing is being conducted against devices on your network, turning the attack into a detection opportunity.
We discuss implications for stalkerware detection — reminding that metadata alone can be deeply revealing, and a phone number remains a dangerously powerful attack surface.
# About the speaker
Dejan Strbad breaks things and builds things — sometimes in that order. A serial entrepreneur and security enthusiast, he co-founded Ascalia (IIoT/Industry 4.0) and is a researcher at the Lisbon Council. With over a decade of experience spanning solution architecture, software engineering, and machine learning, his current obsession is solving problems with ML - the messier, the better.