Abstract

TLS (i.e., HTTPS), since TLS encrypts only the application payload, leaving network- and transport-layer headers and traffic characteristics (packet sizes, timings, and ordering) observable to on-path adversaries. Website fingerprinting leverages this residual metadata to reconstruct browsing patterns and compromise user privacy. Website fingerprinting (WF) attacks can infer which webpage a user visits from encrypted HTTPS traffic alone, compromising privacy even without decryption. WF defenses commonly shape traffic through noise, padding, delays, or flow splitting yet they are most often studied from the perspective of encapsulating protocols like Tor or VPN rather than at the application layer (HTTP). In response, fingerprinting defenses aim to make the IP/TCP/UDP metadata uninformative. Popular obfuscation techniques rely on the Tor and VPN protocols, which encapsulate traffic and obscure its content from on-path observers. However, most web traffic does not use Tor or VPN, leaving HTTP packet metadata directly exposed to on-path adversaries such as ISPs or network operators. For this reason, we focus on fingerprinting scenarios that can be defended by clients or servers at the application-layer (HTTP). In this work, we focus on application-layer defenses enabled by the most widely deployed version of HTTP HTTP/2. We demonstrate how known defenses can be emulated through HTTP/2 features at the client side (HTTPOS, LLaMA, FRONT, Tamaraw) and the server side (ALPaCA, Tamaraw). We further show that HTTP/2 features such as proactive resource suggestion, multiplexing, and flow control offer untapped potential for lightweight yet effective defenses deployable at both endpoints. Application-layer fingerprinting defenses generally pursue one of two strategies: (1) uniformity, wherein the defense transforms webpage metadata into fixed packet lengths and rates; or (2) unpredictability, wherein the defense ensures that metadata differs across subsequent page loads. To follow these strategies, fingerprinting defenses commonly rely on four obfuscation functionalities: (1) noise traffic insertion, (2) packet padding, (3) packet delay, and (4) packet splitting [63]. These mechanisms do not hide IP-layer metadata or the destination domain and therefore complement (not replace) Tor or VPN solutions when network-layer anonymity is required. Instead, they target subpage fingerprinting over ordinary HTTPS by perturbing transport-visible features such as packet sizes, timing, and ordering, thereby reducing leakage about the specific content accessed within a known domain. This distinction is relevant when the domain itself is not sensitive, but the visited subpage is, such as a particular product or medical page. We evaluate these defenses using a unified blueprint that calibrates defense parameters per dataset, then combines practical attacks, information-theoretic leakage estimates, and overhead measurements. For each defense, this framework identifies the strongest hyperparameter-tuned fingerprinting model and estimates the residual uncertainty induced by the defense using two information-theoretic leakage estimators all while accounting for the defenses privacyoverhead trade-offs.