FROST: Exploiting OPFS and SSD Timing for Cross-Browser Fingerprinting

Modern web browsers are designed with rigorous sandboxing to ensure that a website in one tab cannot “reach out” and spy on your other activities.

However, new research has demonstrated that these digital walls are thinner than we thought. By exploiting the physical realities of hardware, a new side-channel technique called FROST (Fingerprinting Remotely using OPFS-based SSD Timing) allows malicious websites to track user behavior by monitoring the subtle timing fluctuations of a computer’s Solid State Drive (SSD).

The research, presented by security researchers including Hannes Weissteiner, reveals that a single “drive-by” visit to a compromised page is sufficient to begin passively monitoring a user’s disk activity. This isn’t just about tracking which websites you visit; it’s about sensing the footprint of almost every application running on your machine.

The Mechanics of the Attack: Exploiting OPFS and Hardware Contention

At the heart of FROST is the abuse of the Origin Private File System (OPFS) API. While OPFS is a legitimate, performance-oriented web storage feature, FROST repurposes it to create a “noisy” environment for measurement.

The attack follows a sophisticated technical workflow:

• Bypassing the Cache: The malicious site uses OPFS to create an exceptionally large file within its own origin-scoped sandbox. It then performs continuous, randomized read operations. Because the file is massive, the browser cannot simply serve it from RAM (the page cache), forcing the system to perform actual, physical I/O on the SSD.
• Measuring Contention: Every time a user opens a new app (like Spotify) or a different browser (like Safari) performs a disk operation, that activity competes for the SSD’s attention. This competition causes microscopic “latency spikes” in the malicious site’s own read operations.
• Precision Timing: To catch these tiny fluctuations, the attacker uses “Cross-Origin Isolation” headers. This configuration allows the website to bypass standard browser protections and regain access to high-resolution timers like performance.now(), which are necessary to measure millisecond-level delays.
• Pattern Recognition: The resulting raw data—a messy stream of timing measurements—is processed through a Convolutional Neural Network (CNN). This AI model learns to recognize the unique “rhythm” of specific applications and websites based on how they interact with the SSD.

Proven Effectiveness: From Browsers to Desktop Apps

The researchers demonstrated that FROST is remarkably accurate. In controlled experiments on macOS, the CNN could predict which website a victim was visiting in a completely different browser (e.g., a Safari instance) by simply watching the timing traces from a malicious Chrome tab. The system achieved an F1 score of 86.95 in “open-world” settings, meaning it could identify targets effectively even amidst the noise of general computer use.

Even more alarming is its ability to fingerprint desktop applications. The researchers successfully distinguished between various pre-installed macOS utilities—such as Maps and System Settings—with a staggering 95.83% accuracy. Because the signal originates at the hardware level, the “sandbox” provided by the browser offers no protection; as long as the processes share the same physical SSD, they are visible to the attacker.

Privacy Implications and the Erosion of the Sandbox

FROST represents a shift in the threat landscape. Unlike traditional fingerprinting that relies on software quirks (like screen resolution or fonts), FROST relies on the fundamental physics of hardware contention. This makes it incredibly difficult to mitigate without changing how web storage works entirely.

The primary risk is long-term, silent surveillance. An advertiser or a malicious actor could keep a single tab open in the background, quietly building a profile of a user’s daily habits, software usage, and browsing patterns without ever triggering a permission prompt or a security warning.

Limitations and Potential Defenses

While powerful, FROST is not a universal “skeleton key.” Its effectiveness depends on a few technical factors:

• Storage Architecture: The attack requires the victim’s activity to occur on the same SSD as the OPFS file. Users with complex multi-disk setups might be less vulnerable.
• Memory-Based Profiles: Advanced users who move their browser profiles into RAM (using tools like profile-sync-daemon) can effectively “hide” their disk activity from the SSD-based sensor.
• Resource Visibility: Allocating massive files to the OPFS can be resource-intensive and might be flagged by advanced system monitoring tools.

As of now, the industry response has been measured. Chromium does not classify this type of fingerprinting as a security vulnerability, while Apple and Mozilla have acknowledged the research but have yet to implement definitive countermeasures.

Proposed mitigations include capping the size of OPFS files to prevent them from bypassing the cache, or requiring explicit user consent before a website is allowed to utilize high-resolution timers alongside large-scale storage capabilities.