Research Reveals File Notification Vulnerabilities in Major Operating Systems

Researchers at Graz University of Technology have identified significant vulnerabilities in the file-notification systems of Windows, Linux, and macOS that can be exploited to monitor user activity. On Windows, for instance, a standard unprivileged account was able to detect 95.7 percent of another user’s visits to popular websites using Firefox. This capability allows an attacker to reconstruct a victim's browsing history based solely on file notifications.

The study reveals that file-notification systems are designed to alert programs when files change, which can be useful for legitimate purposes such as updating documents or syncing files. However, the researchers demonstrated that these notifications can be manipulated to track user actions across different accounts. They utilized tools like Linux’s inotify, Windows’ ReadDirectoryChangesW, and macOS’s FSEvents, which can be accessed by any account with read permissions.

In their experiments, the researchers mapped user activities by recording file events while performing various tasks. They created templates that covered a wide range of actions, including terminal commands, keyboard and mouse inputs, and interactions with virtual machines. The monitoring process was found to consume minimal CPU resources, with a maximum overhead of 0.21 percent.

The researchers placed a watch on the root of the C: drive on Windows, which allowed them to receive notifications about file creations, modifications, and renames across the drive, including in other users’ home directories. This capability raises serious privacy concerns, as it enables the tracking of activities that should remain confidential.

In their browser tests, the team found that Firefox created site-specific folders for 95.7 percent of the tested sites, while Microsoft Edge had a lower detection rate of 32 percent. The difference was attributed to the way these browsers manage data storage. The findings indicate that even routine actions, such as running a Windows Defender scan, can inadvertently expose sensitive information about other users.

Microsoft acknowledged the findings but stated that the behavior is by design, revealing only file names and paths within another user’s profile directory. The researchers argue that this design choice should be reconsidered, as it poses a significant risk to user privacy.

Vulnerabilities Exploited in Linux Systems

On Linux, the vulnerabilities are even more pronounced, as an account with read access to a folder can receive notifications for all files within it, including sensitive device files. This allows attackers to infer keystroke timing, which can be used to reconstruct typed text. In tests, the attack successfully captured nearly every keystroke from multiple typists.

The same method was effective over SSH, where a process on a server was able to capture keystrokes from a remote user. The researchers also demonstrated that malware running in a user’s account could overlay a fake password prompt over a legitimate one, further compromising security.

The team disclosed their findings to security teams across various platforms, including Linux, Android, Windows, and macOS, between August and October 2025. All four acknowledged the vulnerabilities, and Linux kernel developers implemented a partial mitigation in early 2026, assigning the findings the identifier CVE-2025-68788. However, the researchers noted that the patch only addressed part of the issue, leaving other areas vulnerable.

In contrast, macOS exhibited the least leakage among the three operating systems tested, attributed to its more restrictive file-notification system. However, even on macOS, the researchers observed some data exposure related to app launches and device changes.

Intelligence Reviewed

The study also extended to Android devices, where a zero-permission app was able to bypass scoped storage protections, revealing sensitive files and media. This highlights the ongoing challenges in securing user data across various platforms.

The authors have made their code publicly available for all four platforms, encouraging further research and mitigation efforts to address these vulnerabilities.