File notification APIs leak user activity across major OSes

Researchers at Graz University of Technology in Austria have shown that the file-change notification features in Linux, Android, Windows and macOS can be used to spy on other users of the same system. The information exposed ranges from the rhythm of someone's typing to the websites they visit.

All four operating systems let applications ask to be told when files are created, modified or deleted. Text editors, file managers, sync clients and antivirus tools rely on this. It needs no elevated privileges, only read access to the watched location.

The attacks never reveal what is inside a file. The researchers found that file names and the timing of events are enough to piece together what users, applications and the system are doing. Most scenarios assume an attacker who can already run code on the machine under a separate account. On Android, that attacker could be an app that asks for no permissions at all.

Keystrokes, fonts and fake prompts on Linux

On Linux, a user who is not allowed to watch a protected file can still get its events by watching the folder that holds it, as long as that folder is readable. When the researchers applied this to the device files that represent keyboard input, they could see when a key was pressed, though not which key.

That still matters. More than twenty years of research has shown that the gaps between keystrokes can help reveal what is being typed. In tests with seven typists, the attack detected keystrokes reliably and scored between 93.1% and 100% on a standard accuracy measure. Input that does not appear on screen, such as a hidden sudo password, is not captured.

The team also fingerprinted websites by watching which system fonts Firefox loads for a page. This identified sites from the top 100 with 87.9% accuracy.

A third Linux attack targeted the KDE Plasma 6 desktop on Wayland. A malicious process running as the victim notices when the genuine authentication dialog is about to open and draws a fake one on top of it to collect the credentials.

WhatsApp activity on Android, browsing on Windows

The Android tests ran on a Google Pixel and on Samsung Galaxy phones. An app with no permissions was able to watch another app's private storage folder. Against WhatsApp, this showed when photos, videos and documents were sent or received, what the files were called, and when they were later deleted.

On Windows, the biggest problem appears when a user monitors the root of the system drive. Windows then reports the full path of every changed file on the machine, including files in other users' home directories that the monitoring account cannot open.

Firefox stores data for many websites in folders named after each site. An unprivileged user can therefore see, in real time, which sites another user visits. Across the top 1,000 websites, accuracy reached 97.8% for Firefox and 48.5% for Edge, which creates folders for far fewer sites.

macOS leaked the least, because only globally readable files can be monitored. Even so, the researchers could track app launches, interactions with apps and changes to settings.

Few fixes, and a "by design" answer

The Linux kernel has been partly hardened so that device files no longer produce access and modify events. The researchers say this deals with the most severe issues. The fix is tracked as CVE-2025-68788. No fixes are listed for Android or macOS.

Microsoft told the researchers the Windows behavior is by design and exposes only file names and paths in another user's profile, not contents or sensitive data. In a statement to SecurityWeek, a spokesperson said the company had determined "that this is not a security vulnerability," noting that the technique "requires an attacker to already have the ability to run code locally on a device under a separate user account." Microsoft also pointed to protections it documented in April 2025 for some file-path disclosure scenarios involving directory change notifications, which administrators can turn on.

Apple and Google did not respond to SecurityWeek's request for comment. The researchers say they know of no in-the-wild exploitation, but proof-of-concept code has been published on GitHub.

Our Take

The research is a reminder that metadata can be as revealing as content. None of these attacks read a single file, yet they infer typing patterns, browsing habits and messaging activity.

The Android findings are probably the most relevant for everyday users. The attacker there is simply an app with no permissions, a lower bar than the multi-user scenarios on desktops. Given how often malicious Android apps reach real phones, it is worth watching whether Google responds and restricts access to other apps' storage folders.

Microsoft's "by design" position is consistent with its view that local code execution is already a breach of trust. For shared machines, terminal servers and multi-user environments, admins may still want to look at the optional protections Microsoft documented. With public proof-of-concept code available, the gap between research and abuse could narrow.