Websites can fingerprint apps and tabs through SSD I/O latency measured from large Origin Private File System files using a convolutional neural network trained on contention traces. The Ars Technica report on the FROST attack cites the researchers' description of random reads from gigabyte-scale OPFS files to capture latency spikes caused by concurrent host SSD activity, enabling classification of user actions on M2 macOS with the same primitive shown functional on Linux. Primary measurements derive from the DIMVA 2025 paper "FROST: Fingerprinting SSD Activity from JavaScript" which details the CNN architecture and notes untested Windows behavior plus the requirement for same-drive storage; related timing attacks documented in USENIX Security 2023 on browser-based cache contention and IEEE S&P 2022 on storage-channel leaks demonstrate the pattern of escalating web-exposed hardware side channels that predate FROST. Browser mitigations such as OPFS size caps remain unimplemented while the attack's dependence on persistent large files links it to existing surveillance practices that favor low-visibility hardware telemetry over detectable network requests.