Exploring Privacy Risks of Inertial Sensing on Wearable Rings: A Feasibility Study on Keystroke Classification

Abstract

The smart ring form factor is gaining popularity in wearable computing, as it offers a unique balance of stable, continuous signal acquisition and minimal obstruction to daily activities. Recent research has demonstrated that ring-worn Inertial Measurement Units (IMUs) are able to capture nuanced motor patterns, ranging from sign language gesture tracking and signature-based authentication to precise touch contact sensing on arbitrary surfaces. In this work, we investigated the feasibility of leveraging fine-grained finger movements to infer typed text on a keyboard using motion data from smart rings. This presents a potential privacy side-channel; if keystrokes can be inferred from a single ring alone, sensor access could be exploited to snoop sensitive user data such as passwords or private communications. To evaluate this threat, we implemented a pipeline leveraging a time-series foundation model to decode unconstrained typing. In a feasibility study with 2 participants, our method achieved an average top-1 keystroke accuracy of 84.5% from two wearable rings and successfully reconstructed semantically coherent English text without requiring strict typing constraints.