New File Notification Attack Lets Hackers Track User Activity Across Linux, Windows and macOS
Graz University of Technology researchers show file-notification APIs (inotify, ReadDirectoryChangesW, FSEvents) leak user activity across Linux, Windows, macOS without privileges.
Researchers at Graz University of Technology published 'File Notification Attacks,' showing OS file-notification services (inotify on Linux, ReadDirectoryChangesW on Windows, FSEvents on macOS) let unprivileged local processes surveil user behavior. Semi-automated templates identified terminal commands, keystroke timing (F1 93.1–100% for seven users), SSH pseudo-terminal activity (100% F1), website visits, printing, and USB/Bluetooth/VPN changes at 0.2–11.5ms resolution with under 0.21% CPU overhead. On Windows, watching C:\ exposed other users' paths and Firefox browsing across 975 top-1,000 sites at 97.8% F1; on Linux the team demonstrated a KDE Plasma 6 pkexec UI-redress attack and Firefox website fingerprinting at 87.9% F1. Microsoft called the behavior by design (EnforceDirectoryChangeNotificationPermissionCheck is disabled by default) and Linux shipped only a partial mitigation in early 2026.
- inotify, ReadDirectoryChangesW, FSEvents leak file activity without privileges
- Keystroke-timing detection hit F1 93.1–100% for seven Linux users
- Windows Firefox browsing inferred across 975 sites at 97.8% F1
- KDE Plasma 6 pkexec UI-redress authentication attack demonstrated
- Microsoft says by design; Linux shipped partial mitigation early 2026
From the original article
3 min read at cybersecuritynews.com
A cross-platform side-channel attack that abuses file-notification services to monitor user activity on Linux, Windows and macOS. The technique turns inotify, ReadDirectoryChangesW, and FSEvents, which alert applications when files are changed, into a surveillance mechanism requiring no elevated privileges.
By Guru Baran · cybersecuritynews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.