A naive-Bayes website-fingerprinting classifier achieves AUC > 0.94 against vanilla Tor for 8 of 9 Alexa top-ten sites (e.g., Wikipedia 0.9991, YouTube 0.9947). Against StegoTorus-HTTP, AUC drops to ≤ 0.75 for 7 of 9 sites (YouTube 0.4125, Facebook 0.5413, Google 0.6928), which the authors argue is too low for practical perimeter-scale deployment where near-perfect precision is required to avoid error floods.

The StegoTorus HTTP module degrades severely with network latency: it can sustain only a 50 kB/s stream at latencies below 200 ms and fails entirely at higher rates or latencies, because the HTTP request-response pattern transfers only one or two 512-byte Tor cells per round-trip. Plain Tor and chopper-only StegoTorus show no measurable throughput degradation at latencies up to 450 ms. Increasing parallel HTTP connections improves low-latency throughput but does not recover high-latency performance.

§6, Figure 6 evaluation

HTTP steganography in StegoTorus expands upstream traffic by a factor of 41× and downstream by 12× compared to a direct connection (uploading 966,964 bytes vs. 23,643 bytes to transfer a 1 MB file). Chopper-only operation adds only ~2.7× upstream overhead, comparable to plain Tor. Maximum achievable goodput with the HTTP module is ~27 kB/s (~4× a 56 kbps modem), which the authors attribute to a minimum expansion factor of 8× inherent in contemporary steganographic schemes.

§6, Table 3, Figure 7 evaluation

Tor's fixed 512-byte cells packed into TLS 1.0 records produce a characteristic TCP payload of 586 bytes (512 + 74 bytes of TLS overhead). A perimeter filter running a simple exponential moving average (τ ← ατ + (1−α)1ₗ₌₅₈₆, α=0.1, T=0.4) identifies Tor flows within a few dozen packets; this attack succeeds at backbone rates of ~540,000 packets/second on commodity hardware. Obfsproxy does not alter packet sizes or timings and therefore does not defeat this classifier.

§5.1, Figure 3 detection

StegoTorus distributes a fixed set of packet traces and HTTP covertext databases with the software, but allows users to record their own; classifiers trained on the distributed covertext will not generalize to user-generated databases. The paper further notes that reusing a small number of traces repeatedly creates a statistical fingerprint because censors can learn conversation patterns from packet sizes and timings alone, implying that trace diversity must be maintained over time.

§4.1.1, §5.2 defense

Censorship classifiers and traffic analysis attacks consistently exploit the initial seconds of a proxy connection, where packet-size, inter-arrival-time, and burst features are maximally discriminative. Cited work demonstrates that website fingerprinting classifiers trained solely on the first few seconds of Tor traffic achieve high accuracy, and real-world GFW detection of fully-encrypted protocols also targets early-connection bytes.

2025-pereira-extended §1 traffic-shapewebsite-fingerprintml-classifierdpifully-encrypted-detect

The framework's GAN-based schedule generator trains on short session windows (e.g., the first 10 seconds) of real browsing traffic from the Tranco Top 1000 sites, learning joint distributions of packet sizes, inter-arrival times, and burst patterns to produce realistic synthetic schedules. This repurposes GAN architectures previously used for traffic analysis (e.g., GANDaLF) as a defense-side cover-traffic generator.

2025-pereira-extended §2.2 traffic-shapewebsite-fingerprintml-classifier

When a user splits traffic across N paths, a censor observing a single path sees only a partial trace, substantially reducing the accuracy of classifiers trained on complete network traces. Prior Tor traffic-splitting work (TrafficSliver, CoMPS, multipath Tor studies) has validated this defense against website fingerprinting outside the PT context.

2024-lorimer-extended §1, §4.2 website-fingerprintml-classifiertraffic-shape

The host-profiling censor (passive traffic analysis: count connections per server, block those exceeding a threshold τ within a window w) blocks essentially all circumvention user traffic within 30 time steps for all classifier qualities tested (ρ_TP ∈ {0.9, 0.95, 0.99}), while causing far less collateral damage than zig-zag (never exceeding ~30% innocent server blocking). Snowflake resists this attack well: with w=3, τ=3, over 94.48% of users receive a proxy within 2 steps even with worst-classifier rules, and final unblocked server rates are 91.24–99.04%. The host profiling approach is feasible for passive censors who cannot enumerate the distribution channel.

2026-fares-game §5.2, Table 2, Table 3 traffic-shapeml-classifier

Balboa's synchronous leaf-content replacement adds non-negligible timing differences that allow censors to identify its activity with up to ~90% accuracy over different network conditions. The timing anomaly arises because Balboa performs data substitution directly at each data exchange, delaying the server's response while covert data is prepared.

2026-kamali-huma §I (Introduction), §II-A traffic-shapeml-classifier

Without chunk-based padding, an XGBoost classifier identifies the target website from covert data-chunk sizes with 91% accuracy (Tranco top-100). Chunking at 2 MB reduces accuracy to 12% at a 21.3% bandwidth overhead, while 16 MB chunks reduce accuracy to near random guessing at a 480.3% overhead. Chunks as small as 64 KB already reduce accuracy to 64%, demonstrating a monotonic fingerprinting–overhead tradeoff.

2026-kamali-huma §V-C, Figure 4 website-fingerprintml-classifier