Towards Preserving Privacy and Building Trust among the Users of Cyber-Physical Systems

Abstract

“Smart” cyber-physical systems (CPS), from smart homes to smart grids to smart farms, are increasingly embedded in our lives. Each day, they generate large volumes of data through their smart sensors. These data support various artificial intelligence applications such as activity recognition, preventive and predictive maintenance, operational efficiency, and energy optimization, all of which are typically done in the cloud and outside the control of the person/system generating the data. This leads to concerns about the security of these systems, the privacy/confidentiality of the data they generate, and the risks of compliance violations due to the unauthorized use or exposure of personal information. In evolving CPS applications, multiple IoT-based smart systems must collaborate as a part of a distributed infrastructure. This leads to additional concerns about trusting the data/information in the presence of adversaries. We present an approach that solves these issues to support secure, compliant, and resilient operations in CPS environments using a policy-driven access control framework that combines dynamic access control with truth maintenance through context. First, we present the design and implementation of our PROTEGO framework that preserves the security of the system and the privacy of data collected by creating policies on data access grounded in Attribute-Based Access Control. These policies describe who can access the smart system and its data and in what context. When sharing data, the policy describes whether data can be shared based on the context that includes user preferences to infer if and how the data needs transformation for varying levels of privacy before sharing externally. While our framework is agnostic to how data is transformed, we show that transformed data, when run on downstream cloud-based applications such as anomaly detection, has minimal impact on accuracy. Second, we present our Bee-Thoven framework, which detects regulatory violations for sensitive data generated by smart sensors by tracing information flows using Extended Berkeley Packet Filters. This allows real-time monitoring and enforcement of data compliance policies like GDPR, CCPA, and HIPAA without modifying the application code in a transparent, application-agnostic manner with minimal overhead. Third, we design and implement ResilIoT to establish trust between agents by exploring and identifying false information sent across the ecosystems while incorporating context via user preferences. We build on existing truth maintenance systems to create policies that identify false information generated by sensors influenced by adversaries. Using two simulated real-world datasets, we show how this detects and resolves conflicts with minimal additional resources.