Muslum Ozgur Ozmen (Purdue University), Ruoyu Song (Purdue University), Habiba Farrukh (Purdue University), Z. Berkay Celik (Purdue University)

In smart homes, when an actuator's state changes, it sends an event notification to the IoT hub to report this change (e.g., the door is unlocked). Prior works have shown that event notifications are vulnerable to spoofing and masking attacks. In event spoofing, an adversary reports to the IoT hub a fake event notification that did not physically occur. In event masking, an adversary suppresses the notification of an event that physically occurred. These attacks create inconsistencies between physical and cyber states of actuators, enabling an adversary to indirectly gain control over safety-critical devices by triggering IoT apps. To mitigate these attacks, event verification systems (EVS), or broadly IoT anomaly detection systems, leverage physical event fingerprints that describe the relations between events and their influence on sensor readings. However, smart homes have complex physical interactions between events and sensors that characterize the event fingerprints. Our study of the recent EVS, unfortunately, has revealed that they widely ignore such interactions, which enables an adversary to evade these systems and launch successful event spoofing and masking attacks without getting detected.

In this paper, we first explore the evadable physical event fingerprints and show that an adversary can realize them to bypass the EVS given the same threat model. We develop two defenses, EVS software patching and sensor placement with the interplay of physical modeling and formal analysis, to generate robust physical event fingerprints and demonstrate how they can be integrated into the EVS. We evaluate the effectiveness of our approach in two smart home settings that contain 12 actuators and 16 sensors when two different state-of-the-art EVS are deployed. Our experiments demonstrate that 71% of their physical fingerprints are vulnerable to evasion. By incorporating our approach, they build robust physical event fingerprints, and thus, properly mitigate realistic attack vectors.

View More Papers

Blaze: A Framework for Interprocedural Binary Analysis

Matthew Revelle, Matt Parker, Kevin Orr (Kudu Dynamics)

Read More

PISE: Protocol Inference using Symbolic Execution and Automata Learning

Ron Marcovich, Orna Grumberg, Gabi Nakibly (Technion, Israel Institute of Technology)

Read More

Cooperative Perception for Safe Control of Autonomous Vehicles under...

Hongchao Zhang (Washington University in St. Louis), Zhouchi Li (Worcester Polytechnic Institute), Shiyu Cheng (Washington University in St. Louis), Andrew...

Read More