Other users can watch your browsing and time your keystrokes through OS file notifications

* Researchers at Graz University of Technology used file-notification systems in Windows, Linux, and macOS to spy on activity in other accounts.
* On Windows, an unprivileged account detected 95.7 percent of another user’s visits to popular websites by monitoring browser data folders.
* On Linux, a weakness exposed keystroke timing from the local keyboard and SSH sessions.
* Programs use file change notifications like Linux’s inotify, Windows’ ReadDirectoryChangesW, and macOS’s FSEvents.
* The attack involved running code under a separate, unprivileged account on the victim’s machine.
* The method mapped terminal commands, keyboard/mouse input, browsing, printing, VMs, Docker containers, Bluetooth, network/VPN changes, and USB device events.
* The drive-wide watch cost at most 0.21 percent of CPU time during measurements.
* Windows reported file creations, modifications, and renames across the drive with full paths, including other users' folders.
* File change events were used to track browser data in folders named after websites for Firefox and Edge.
* Linux leaks keystroke timing from device files, which can reconstruct typed text.
* Patches were developed by security teams, including a Linux kernel mitigation assigned CVE-2025-68788.

The findings reveal a fundamental tension between operating system design prioritizing user separation and the inherent visibility granted by file-system event notifications. The attack leverages legitimate system mechanisms—the very tools designed for application interaction (file watching)—as an information conduit across security boundaries. The observation that metadata, such as file paths or timing rhythms, can reconstruct sensitive activities demonstrates a systemic flaw where isolation is conditional rather than absolute. The development of countermeasures, like filtering file events based on ownership and implementing stricter permission checks (e.g., the proposed Linux solution), highlights a process where vulnerability is identified externally and subsequently patched internally. This raises questions about the efficacy of layered security when low-level system calls can bypass intended access controls, suggesting that sovereignty over data relies not just on access permissions but on the integrity of the notification layer itself.
Bridge Questions: If file system events are inherently leaky across user boundaries, what fundamental architectural shift is required to ensure that OS primitives enforce isolation regardless of application or process context? How should security architects redefine "permission" in a distributed environment when low-level notifications can reveal state changes outside explicit access rights? What mechanisms must be established to audit the integrity of these cross-context notification channels?