Reproduce “WebGPU-SPY: Finding Fingerprints in the Sandbox through GPU Cache Attacks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ethan Ferguson, Adam Wilson, Hoda Naghibijouybari
Researchers from Binghamton University developed 'WebGPU-SPY', a novel attack demonstrating that the WebGPU standard enables GPU-internal cache side-channel attacks capable of bypassing existing browser security mitigations. The technique utilizes a hardware-based timer on the GPU's Shared Local Memory to spy on a victim's web rendering via the GPU's L3 cache, achieving up to 90.6% accuracy in website fingerprinting.
Key insights
- WebGPU enables a new class of microarchitectural side-channel attacks entirely within the GPU context, effectively bypassing browser mitigations designed for CPU-centric timing attacks.
- GPU Shared Local Memory can be leveraged as a precise, hardware-based timer, providing a low-noise mechanism for cache hit/miss differentiation independent of software APIs.
- The GPU's internal L3 cache acts as an effective cross-stack leakage channel, allowing a WebGPU compute process to infer activities from the GPU's graphics rendering process.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.