UC Irvine

A smart TV is an Internet-connected TV with computational capabilities. These enhancements to the traditional TV set enable the smart TV to stream content from the Internet and run interactive applications (apps). While appealing from a functionality standpoint, smart TVs unfortunately also introduce new privacy risks. For example, unlike traditional TVs which can only receive TV channel broadcasts, the smart TV may use its Internet-connectivity to exfiltrate information about the user, including, but not limited to, viewing history. Despite massive user adoption of smart TVs, there is surprisingly little work on the privacy implications of smart TVs. Using a network measurement approach, this dissertation seeks to close this gap in the literature.

The first part of this dissertation presents a large-scale measurement study of the smart TV advertising and tracking ecosystem. Network traffic collected from smart TVs, when used by real users and when instrumented in the lab, reveal that smart TVs connect to well-known and platform-specific advertising and tracking services (ATSes). Automated tests of the top-1000 apps on two popular smart TV platforms unveil that (i) a subset of apps communicate with a large number of ATSes, and some ATS organizations only appear on certain platforms, showing a possible segmentation of the smart TV ATS ecosystem across platforms; and (ii) hundreds of apps exfiltrate personally identifiable information to third parties and platform domains. Furthermore, an evaluation of DNS-based blocklists shows that even smart TV-specific blocklists miss ads and incur functionality breakage.

Next, the dissertation investigates if an in-network adversary can identify what smart TV app is launched. Automated tests are used to collect multiple samples of the network traffic generated by each of the top-1000 apps on the three most popular smart TV platforms. Network fingerprints are extracted from this dataset using three established fingerprinting techniques. The results show that smart TV app network fingerprinting is feasible and effective: even the least prevalent type of fingerprint manifests itself in at least 68% of apps of each smart TV platform, and up to 89% of fingerprints uniquely identify a specific app when two fingerprinting techniques are used together. It is also shown that apps that exhibit identical fingerprints often stem from the same developer or "no code" toolkit, and that apps that are present on all three smart TV platforms exhibit platform-specific fingerprints.

Finally, inspired by the observation that joint use of multiple fingerprinting techniques improves fingerprint distinctiveness, the dissertation proposes a general fingerprinting framework that can identify fingerprints that are based on any combination of several features, such as server identities and the sizes, directions, and/or order of packets. Through customizable parameters, the framework provides support for both joint and separate use of prior fingerprinting techniques, as well as any fingerprinting technique that can be formulated as a problem of identifying similar packet exchanges. To demonstrate its versatility, the framework is used to implement and evaluate two different fingerprinting techniques. Fingerprints for smart TV apps and for events on simple Internet of Things (IoT) devices, such as smart plugs and smart light bulbs, are extracted using the two fingerprinting techniques. The relative performance of the two fingerprinting techniques is established by comparing the number of fingerprints each fingerprinting technique identifies, as well as how distinct the extracted fingerprints are from other traffic.