Mozilla: Firefox 143 fingerprint protections cut uniquely identifiable users from 75% to 46%

Mozilla researchers, together with colleagues from the University of Toronto and UC San Diego, published a study of browser fingerprinting based on telemetry from almost 25 million Firefox release users, according to the companion site fingerprinting.study fingerprinting.studyfingerprinting.studyfingerprinting.studyOpen source ↗. The protections designed from these data shipped in Firefox 143. The authors report that they reduced the share of uniquely identifiable users from 75% to 46% while maintaining web compatibility. The measurements used privacy-preserving technologies such as OHTTP, so no individual user could be identified or tracked across the web.

What makes a browser unique

The authors report that fingerprintability is driven largely by a small set of high-impact characteristics and their combinations, not by the accumulation of many independent measurements. For Fingerprint Pro, the most discerning fingerprinter in the model, 80% of the users it sees as unique are already unique after 8 metrics, although it collects over 50. Signifyd collects over 80 metrics but identifies fewer users as unique: 80% of those it sees as unique are identified after 9 metrics.

The authors advise caution with any single percentage for "how many users are unique", because it merges populations that behave differently. The operating system is always exposed to a fingerprinter: spoofing it breaks shortcut keys and gives wrong downloads, among other problems. The team therefore segmented the population by operating system. Each system has different important metrics, and every aggregate figure has a per-OS breakdown.

Beyond Shannon entropy

Shannon entropy is the standard yardstick in this literature, but the study shows that a high value does not reliably predict fingerprinting impact fingerprinting.studyfingerprinting.studyfingerprinting.studyOpen source ↗. Earlier analyses counted unique users and did not examine everyone else. Those users could be 1 million people sharing one fingerprint or 500,000 distinct fingerprints shared by 2 people each.

To capture that difference, the authors use the Simpson Index, also known as the Herfindahl–Hirschman Index, to compare sets of cohorts, meaning groups sharing a fingerprint. This lets them compare defenses not only by how much uniqueness they eliminate.

The site also publishes the paper, appendix data, privacy-preserving aggregate data and analysis code, a description of data collection, an interactive test that compares a user's browser with the measured population, and a discussion board for proposing follow-up research.