Many solutions have been proposed to curb unexpected behavior of automation apps installed on programmable IoT platforms by enforcing safety policies at runtime. However, all prior work addresses a weaker version of the actual problem due to a simpler, unrealistic threat model. These solutions are not general enough as they are heavily dependent on the installed apps and catered to specific IoT platforms. Here, we address a stronger version of the problem via a realistic threat model, where (i) undesired cyber actions can come from not only automation platform backends (e.g., SmartThings) but also close-sourced third-party services (e.g., IFTTT), and (ii) physical actions (e.g., user interactions) on devices can move the IoT system to an undesirable state. We propose a runtime mechanism, dubbed Maverick, which employs an app-independent, platform-agnostic mediator to enforce policies against all undesired cyber actions and applies corrective-actions to bring the IoT system back to a safe state from an unsafe state transition. Maverick is equipped with a policy language capable of expressing rich temporal invariants and an automated toolchain that includes a policy synthesizer and a policy analyzer for user assistance. We implemented Maverick in a prototype and showed its efficacy in both physical and virtual testbeds, incurring minimal overhead.

In this paper, we perform the first multifaceted measurement study to investigate the widespread insecure practices employed by tertiary education institutes (TEIs) around the globe when offering WPA2-Enterprise Wi-Fi services. The security of such services critically hinges on two aspects: (1) the connection configuration on the client-side; and (2) the TLS setup on the authentication servers. Weaknesses in either can leave users susceptible to credential theft. Typically, TEIs prescribe to their users either manual instructions or pre-configured profiles (e.g., eduroam CAT). For studying the security of configurations, we present a framework in which each configuration is mapped to an abstract security label drawn from a strict partially ordered set. We first used this framework to evaluate the configurations supported by the user interfaces (UIs) of mainstream operating systems (OSs), and discovered many design weaknesses. We then considered 7045 TEIs in 54 countries/regions, and collected 7275 configuration instructions from 2061 TEIs. Our analysis showed that majority of these instructions lead to insecure configurations, and nearly 86% of those TEIs can suffer from credential thefts on at least one OS. We also analyzed a large corpus of pre-configured eduroam CAT profiles and discovered several misconfiguration issues that can negatively impact security. Finally, we evaluated the TLS parameters used by authentication servers of thousands of TEIs and discovered perilous practices, such as the use of expired certificates, deprecated versions of TLS, weak signature algorithms, and suspected cases of private key reuse among TEIs. Our long list of findings have been responsibly disclosed to the relevant stakeholders, many of which have already been positively acknowledged.

The widespread deployment of end-to-end encryption protocols such as HTTPS and QUIC has reduced the visibility for operators into traffic on their networks. Network operators need the visibility to monitor and mitigate Quality of Experience (QoE) impairments in popular applications such as video streaming. To address this problem, we propose a machine learning based approach to monitor QoE metrics for encrypted video traffic. We leverage network and transport layer information as features to train machine learning classifiers for inferring video QoE metrics such as startup delay and rebuffering events. Using our proposed approach, network operators can detect and react to encrypted video QoE impairments in real-time. We evaluate our approach for YouTube adaptive video streams using HTTPS and QUIC. The experimental evaluations show that our approach achieves up to 90% classification accuracy for HTTPS and up to 85 % classification accuracy for QUIC.