CipherSight: Robust Website Fingerprinting via TLS Records
A recent study presents CipherSight, a hierarchical framework based on TLS records designed for effective HTTPS website fingerprinting (WF). This research, accessible on arXiv (2608.13905), tackles the out-of-distribution (OOD) issue stemming from temporal and geographical variations, as well as the difficulty of identifying previously unseen websites in open-world contexts. Current WF techniques depend on raw TCP packet sequences, which are vulnerable to transport-layer artifacts and fail to produce stable website representations. In contrast, CipherSight derives website representations from TLS records by encoding multiple attributes at the record level. Its hierarchical design captures both intra-flow dependencies among TLS records and inter-flow interactions, enhancing performance in real-world scenarios. This work is particularly significant for the digital art community, as it strengthens privacy and security on online art platforms, mitigating the risks of encrypted traffic analysis revealing user activities.
Key facts
- CipherSight is a TLS-record-based hierarchical framework for HTTPS website fingerprinting.
- It addresses out-of-distribution (OOD) problems from temporal and geographic changes.
- It handles previously unseen websites in open-world scenarios.
- Existing methods use raw TCP packet sequences, which are sensitive to transport-layer artifacts.
- CipherSight learns from TLS records by encoding multiple record-level attributes.
- The hierarchical architecture captures intra-flow and inter-flow dependencies.
- The paper is available on arXiv with identifier 2608.13905.
- The research aims to improve robustness and generalizability of website fingerprinting.
Entities
Institutions
- arXiv