Abstract

Transport Layer Security (TLS) is one of the most widely deployed cryptographic protocols on the Internet that provides confidentiality, integrity, and a certain degree of authenticity of the communications between clients and servers. Following Snowden's revelations on US surveillance programs, the adoption of TLS has steadily increased. However, encrypted traffic prevents legitimate inspection. Therefore, security solutions such as personal antiviruses and enterprise firewalls may intercept encrypted connections in search for malicious or unauthorized content. Therefore, the end-to-end property of TLS is broken by these TLS proxies (a.k.a. middleboxes) for arguably laudable reasons; yet, may pose a security risk. While TLS clients and servers have been analyzed to some extent, such proxies have remained unexplored until recently. We propose a framework for analyzing client-end TLS proxies, and apply it to 14 consumer antivirus and parental control applications as they break end-to-end TLS connections. Overall, the security of TLS connections was systematically worsened compared to the guarantees provided by modern browsers. Next, we aim at exploring the non-public HTTPS ecosystem, composed of locally-trusted proxy-issued certificates, from the user's perspective and from several countries in residential and enterprise settings. We focus our analysis on the long tail of interception events. We characterize the customers of network appliances, ranging from small/medium businesses and institutes to hospitals, hotels, resorts, insurance companies, and government agencies. We also discover regional cases of traffic interception malware/adware that mostly rely on the same Software Development Kit (i.e., NetFilter). Our scanning and analysis techniques allow us to identify more middleboxes and intercepting apps than previously found from privileged server vantages looking at billions of connections. We further perform a longitudinal study over six years of the evolution of a prominent traffic-intercepting adware found in our dataset: Wajam. We expose the TLS interception techniques it has used and the weaknesses it has introduced on hundreds of millions of user devices. This study also (re)opens the neglected problem of privacy-invasive adware, by showing how adware evolves sometimes stronger than even advanced malware and poses significant detection and reverse-engineering challenges. Overall, whether beneficial or not, TLS interception often has detrimental impacts on security without the end-user being alerted.