ChatterHub

Privacy invasion via smart-home hubs

Hidden by design IEEE SMARTCOMP 2021; Pervasive and Mobile Computing 2022 Network PaperCode

Smart-home hubs sit between low-power devices (door locks, motion sensors, smart bulbs) and the cloud. Their traffic is encrypted, which is often assumed to keep a household’s activity private. ChatterHub shows that this blind spot persists even behind encryption: an adversary who can only observe the hub’s encrypted traffic can still learn which devices are in the home and what they are doing.

ChatterHub overview: offline training and model generation, and attack phase on encrypted traffic
ChatterHub: an offline training phase builds models from labeled traffic, and the attack phase applies them to a target home's encrypted traffic.

How it works

  • Offline training. Packet traces from a hub are labeled with ground-truth device events.
  • Segmentation. A packet filter and dynamic change-point detection isolate the bursts of traffic caused by individual device events.
  • Classification. Models (sequence-to-sequence, LSTM, and random forest) learn to map each burst to a device and an action.
  • Attack. The trained model is applied to a target home’s encrypted traffic, observed by a nearby sniffer, a compromised router, or an Internet service provider.

Key findings

  • Device identity and user actions can be inferred from encrypted hub traffic without decryption.
  • The attack needs no prior knowledge of which devices are installed in the home.
  • The results highlight the need for traffic-shaping defenses in smart-home ecosystems.

Citation. Omid Setayeshfar, Karthika Subramani, Xingzi Yuan, Raunak Dey, Dezhi Hong, Kyu Hyung Lee, In Kee Kim. ChatterHub: Privacy Invasion via Smart Home Hub. IEEE SMARTCOMP, 2021; extended as Privacy Invasion via Smart-Home Hub in Personal Area Networks, Pervasive and Mobile Computing 85, 2022.