Deep learning-based website fingerprinting (WF) attacks dominate website traffic classification. In the real world, the main challenges limiting their effectiveness are, on the one hand, the difficulty in countering the effect of content updates on the basis of accurate descriptions of page features in traffic representations. On the other hand, the model’s accuracy relies on training numerous samples, requiring constant manual labeling. The key to solving the problem is to find a website traffic representation that can stably and accurately display page features, as well as to perform self-supervised learning that is not reliant on manual labeling. This study introduces the multi-level resource-coherented graph convolutional neural network (MRCGCN), a self-supervised learning-based WF attack. It analyzes website traffic using resources as the basic unit, which are coarser than packets, ensuring the page’s unique resource layout while improving the robustness of the representations. Then, we utilized an echelon-ordered graph kernel function to extract the graph topology as the label for website traffic. Finally, a two-channel graph convolutional neural network is designed for constructing a self-supervised learning-based traffic classifier. We evaluated the WF attacks using real data in both closed- and open-world scenarios. The results demonstrate that the proposed WF attack has superior and more comprehensive performance compared to state-of-the-art methods.

With the frequent occurrence of information security incidents within the Internet of Things (IoT) landscape, there has been an increasing emphasis on anomaly detection in IoT traffic. Recently, supervised machine learning techniques have shown significant potential on this topic. However, the intricate nature of IoT network environments has posed a challenge in acquiring sufficient labeled samples of abnormal traffic. In comparison to supervised learning, unsupervised learning has more lenient sample requirements. Researchers have proposed various unsupervised detection methods, yet limitations persist. Firstly, unsupervised learning, lacking guidance from labeled information, necessitates a more diverse range of traffic perspectives for comprehensive information coverage. Secondly, despite efforts to extract multi-view traffic features from various perspectives, existing methods struggle to integrate these features effectively, limiting interpretability and introducing redundancy and noise. Lastly, conventional unsupervised methods often rely heavily on manually crafted features, potentially leading to biased and limited representations. In this paper, we propose an unsupervised IoT traffic anomaly detection method based on multi-view subspace learning. Specifically, we first construct a multi-view traffic representation, including a protocol field view and a payload semantic view. Subsequently, a multi-view subspace learning algorithm is designed to project the different views of traffic onto a unified and low-rank subspace, optimized using the Augmented Lagrangian Multiplier with Alternating Direction Minimization (ALM-ADM) strategy. Finally, spectral clustering is employed to accomplish IoT traffic anomaly detection. We benchmark the proposed method on multiple IoT traffic datasets and diverse computational platforms. The experimental results demonstrate that the method outperforms other state-of-the-art approaches in terms of accuracy and computational efficiency.

Despite the end-to-end encryption capabilities provided by network protocols such as QUIC in HTTP/3 and the additional tunneling functions offered by proxy tools like virtual private networks (VPNs) and the onion router (Tor), website fingerprinting (WF) techniques can still identify specific network services by exploiting the spatio-temporal characteristics of network traffic. Therefore, defending against WF attacks is crucial for ensuring comprehensive privacy protection for network services. Existing WF defenses typically rely on proxy-based solutions that require coordinated packet manipulations between the client and the proxy node to counteract WF attacks. These symmetric architectures cannot protect network traffic between proxy nodes and web servers from WF attacks. Furthermore, the ability to counter more powerful traffic analysis tools remains a challenging issue. In this paper, we propose WF-A2D, an asymmetric adversarial defense method against website fingerprinting for HTTP/3. WF-A2D employs a two-stage cascading adversarial learning strategy, leveraging packet direction and length patterns to enhance defense performance. Position-based perturbation vectors representing packet operations are generated for packet-by-packet manipulations to achieve real-time WF defense. Experimental results on a real-world HTTP/3-QUIC website browsing traffic dataset demonstrate that WF-A2D can achieve a defense success rate of 97.10% on average against seven state-of-the-art traffic analysis tools, while incurring less than 2% bandwidth overhead. More importantly, WF-A2D can operate independently on the client side and ensure end-to-end protection to web servers.

Network covert channels transmit secret messages by manipulating network traffic, including packet headers, timing intervals, and communication patterns. The growth of network services has spurred interest in exploring these channels. Yet, the practical application of these channels faces challenges in transmission rate and reliability due to unpredictable network interference. QUIC-based websites offer promising opportunities for covert communication, given their inherent dynamic nature from web resource updates and network interferences. Repeated visits or refresh actions on the same website generate substantial statistical redundancy. Furthermore, widely used proxy tools introduce additional traffic morphology changes. This paper presents QuicCourier, a covert channel leveraging web traffic's dynamic characteristics and proxy service encapsulation to hide messages in QUIC packets from the service node to the client. Guided by a generative model for web resource patterns, QuicCourier ensures that covert traffic closely resembles legitimate traffic, employing three packet-wise meta operations. The altered QUIC flows are then encased in proxy protocols, complicating the detection of embedded information. The covert receiver is incorporated into the proxy client. The efficacy of QuicCourier is evaluated using a dataset of over 30,000 web browsing traffic samples, demonstrating its exceptional undetectability against state-of-the-art traffic classification tools and a high covert transmission rate.

Website fingerprinting (WF) attacks analyze encrypted network traffic to exploit side-channel features such as packet sizes, inter-packet timings, and burst patterns, enabling adversaries to infer users’ browsing activities and posing persistent privacy threats even under encryption protocols like TLS. Existing WF defenses primarily rely on static perturbations of coarse statistical features, which fail to reproduce the multi-scale spatio-temporal dynamics of website traffic and are increasingly ineffective against modern deep learning-based classifiers. To address this challenge, we propose WFD-EST, a website fingerprinting defense framework that dynamically emulates spatio-temporal traffic characteristics for fine-grained obfuscation. WFD-EST constructs a multi-scale traffic representation that captures both packet-level dynamics and burst-level correlations. A diffusion-based generator, guided by a fine-tuned large-scale discriminator, synthesizes realistic target traffic templates that preserve structural consistency while reflecting temporal diversity. Based on these templates, a burst-aware manipulation module performs packet padding, insertion, and delay operations to align source flows with target spatio-temporal patterns, generating traffic indistinguishable from real target flows. Evaluations on a real-world dataset comprising 15,000 encrypted samples from three representative websites show that WFD-EST consistently outperforms two state-of-the-art defenses, reducing classification F1 scores by 0.082–0.144 while lowering bandwidth and time overheads by at least 0.086 and 0.054, respectively.