Privacy Sandbox and the Topics API: Why Fingerprinting Still Works in 2026
When Google announced that third-party cookies were going away, many people concluded that cross-site tracking was going away with them. It didn't work out that way. In 2025 Google decided to keep Chrome's existing approach to third-party cookies, and in October 2025 it announced it would retire Topics and most of the other Privacy Sandbox APIs, citing low adoption. Through all of it, browser fingerprinting, which needs no stored identifier, carried on untouched.
This post covers what Privacy Sandbox set out to do, what Topics added while it was live, why fingerprinting was never really in scope, and what that means for privacy-minded users, site owners and teams running several browser profiles.
The Privacy Sandbox Promise vs. Reality
Google announced Privacy Sandbox in 2019. The pitch: phase out third-party cookies, limit cross-site tracking, and preserve ad targeting and measurement for publishers through new browser APIs such as Topics, Attribution Reporting and Protected Audience. Those APIs shipped in Chrome. The cookie phase-out didn't: after years of delays Google kept the existing cookie approach, and the APIs built to replace cookies lost much of their reason to exist.
Fingerprinting was a separate matter all along: Privacy Sandbox never reduced it. Anyone could check this by running a fingerprint viewer in Chrome with the Privacy Sandbox settings on and off. The browser's core signals (canvas, WebGL, audio, fonts, screen) don't change, and Client Hints add more structured data on top.
Topics API: A Signal Wearing a Privacy Costume
Remember FLoC? Google's first attempt at cookie-replacement advertising clustered users into "cohorts" based on browsing history. Privacy advocates objected, the EFF among them, and Google replaced it in 2022.
Topics was FLoC's replacement. Here is how it works.
Topics analyzes the hostnames in your browsing history and assigns you interest categories from a fixed taxonomy of a few hundred topics: sports, personal finance, travel, software, automotive and so on. When you visit a site that uses Topics, your browser can share some of your top topics from the past three weeks.
Google's case that this protects privacy:
- Topics are coarse-grained (only a few hundred categories)
- Some topics returned are random (noise injection)
- Topics expire after three weeks
- You can see and delete your topics
Those are real protections. But Topics still adds a small amount of identifying information, and it combines with everything else. Canvas, WebGL, AudioContext, screen resolution, timezone, Client Hints, Topics: each parameter narrows the population you match.
The noise is also limited. If a site observes Topics across many sessions, the stable topics stand out from the random ones, so over time the real interests are easier to see.
The EFF criticised Topics as a rebrand of FLoC that still shares people's interests with the sites they visit. Whether that is acceptable is a policy question, but it is clearly profiling rather than anonymity.
Google has now announced Topics' retirement, and Chromium's published plan is to deprecate it and then remove it in a later release. Until it is gone from the Chrome versions your visitors run, it is one more signal; after that, it is a footnote. The rest of the fingerprinting surface stays either way.
For anyone running several accounts they are allowed to hold, such as an agency managing client profiles, Topics was another reason to keep each profile's browsing history its own. A profile used for one client's auto-insurance work should not carry interests from another client's unrelated work, and keeping profiles separate means each one's history belongs to that profile alone. Our WebGL fingerprinting deep dive explains how GPU signals combine with other parameters.
Client Hints: More Headers, More Entropy
User-Agent strings were messy: one long string with browser name, version, OS and device info, formatted inconsistently across browsers. Hard to parse, and Chrome has since reduced the detail it carries.
Chrome's replacement is Client Hints (UA-CH). Instead of one string, Chrome sends structured headers:
Sec-CH-UA— browser brand and major versionSec-CH-UA-Platform— operating systemSec-CH-UA-Mobile— mobile or desktopSec-CH-UA-Full-Version-List— complete version numbersSec-CH-UA-Arch— CPU architecture (x86, arm)Sec-CH-UA-Bitness— 32-bit or 64-bitSec-CH-UA-Model— device model (mobile only)Sec-CH-UA-Platform-Version— OS version
That's eight structured headers where one string used to be, and they're easy to parse. Some are sent by default and others only when a server asks for them, which is the privacy argument for the change. But the high-entropy hints (architecture, bitness, platform version, full version list) are exactly the kind of detail that makes a browser distinguishable.
Google's justification is that sites need client capabilities for compatibility, which is true. The fingerprinting implications are a trade-off the design accepted. Client Hints are part of how Chrome reports itself, not an advertising API, so they were not part of the Privacy Sandbox retirement.
What this means for software that presents a browser identity: all UA-CH values must agree with each other and with the User-Agent. An x86 architecture alongside arm-specific hints, or one major version in the User-Agent and another in the version list, is an inconsistency a site can see.
Client Hints are sent as HTTP headers before JavaScript runs, so a content-script extension can't change them. A native engine can generate the User-Agent and UA-CH values from one source. That is how JustBrowser works: each profile's identity parameters come from the engine itself. Our Playwright and Puppeteer REST API guide covers programmatic profile control.
What Privacy Sandbox Didn't Touch
These fingerprinting signals were never in Privacy Sandbox's scope:
- Canvas fingerprinting — GPU-specific rendering differences still expose hardware.
- WebGL parameters — MAX_TEXTURE_SIZE, shader precision and extensions remain queryable.
- AudioContext fingerprinting — audio processing variations still distinguish machines.
- Font enumeration — partially limited, but system fonts still vary by installation.
- Screen resolution and DPR — window size, device pixel ratio, color depth.
- Timezone — a small number of zones with uneven population distribution.
- Navigator properties — platform, hardwareConcurrency, deviceMemory.
- WebRTC — local IP and media device enumeration remain possible unless blocked.
Together these represent a large share of the entropy that makes a browser distinguishable. Privacy Sandbox addressed none of them. Third-party cookies stayed, and the fingerprinting fundamentals didn't change either.
For site owners, this is useful. ClickzProtect, our passive bot detection platform, is an example of a tool that scores traffic on device and connection signals; our JA4 TLS fingerprinting cross-product analysis covers how TLS fingerprints complement browser signals. The audio and font fingerprinting guide covers the entropy breakdown by signal.
Who Gained and Who Lost
Gained, or at least lost nothing:
-
Large platforms with first-party data. Facebook, Amazon, Netflix: you're logged in, so they know who you are. Privacy Sandbox never affected them.
-
Fingerprinting vendors. Their techniques kept working throughout, and may have worked better while users assumed Privacy Sandbox meant they weren't being tracked.
Lost:
-
Publishers and ad-tech firms that built for the new APIs. Integration work for Topics, Protected Audience and Attribution Reporting went into APIs that are now being retired.
-
Privacy advocates expecting strong privacy. Third-party cookies stayed, Topics was profiling, and Client Hints expanded the signal surface.
-
Users who thought Privacy Sandbox made them anonymous. It didn't. The mechanisms changed; the outcome mostly didn't.
What Teams and Engineers Should Actually Do
If you run several accounts you are allowed to hold, such as client accounts at an agency:
-
Keep each profile's history its own. Interest signals, Topics included while it exists, are derived from the browsing a profile does. Separate profiles keep one client's interests from showing up in another's.
-
Ensure Client Hints consistency. Architecture, bitness, platform version and full version list should all agree. A native engine generates them from the same source as the User-Agent.
-
Test your own pages with current tools. Fingerprint viewers that read UA-CH show you what your configuration reports. Use an up-to-date one when you QA how your site behaves for different configurations.
-
Don't assume that changes to cookies mean changes to tracking. Your threat model should assume fingerprinting continues.
If you're building detection systems:
-
Treat Topics as a weak, temporary signal. It is low in information and on its way out, so don't build anything that depends on it.
-
Weight Client Hints properly. UA-CH headers are structured and reliable. Architecture plus bitness plus platform version is quite discriminating.
-
The traditional surface remains. Canvas, WebGL and audio still work, so you don't need to rebuild your stack.
If you care about personal privacy:
Firefox with Resist Fingerprinting mode, Brave or Tor Browser are designed to reduce fingerprinting. Chrome's Privacy Sandbox was designed primarily around ad measurement and targeting, not anonymity.
Our View
Privacy Sandbox was a regulatory compromise that outlived its premise.
Regulatory pressure on cookie-based tracking, from GDPR in the EU and the CCPA in California, made a cookie phase-out look necessary, and Privacy Sandbox was Google's answer. Once Google decided to keep third-party cookies, most of the replacement APIs had nothing left to replace.
Through the whole episode, fingerprinting was untouched. That reads as an ad-measurement project more than a privacy initiative, though the evidence could change that view.
The Fingerprinting Landscape Continues
For teams managing several accounts, nothing fundamental changed. Profile consistency across all parameters, including the expanded UA-CH headers, still matters. Cookies were easy to isolate and every tool could do it. Keeping 40+ identity parameters consistent is harder, and native C++ implementations do it better than API interception.
JustBrowser runs a custom Chromium engine with native C++ integration, patched on every upstream release. Client Hints, canvas, WebGL and AudioContext values are generated at the engine layer, not intercepted in JavaScript, so each profile presents one consistent identity.
Our multi-account browser myths debunked covers where fingerprinting is heading: behavioral analysis, ML scoring, WebGPU. Privacy Sandbox is one change among many.
JustAnalytics takes the opposite approach: privacy-first, cookieless analytics. Different problem, different solution, but both exist because tracking never depended on cookies alone.
Frequently Asked Questions
Does Privacy Sandbox reduce browser fingerprinting?
No. Privacy Sandbox was aimed at third-party cookies, not fingerprinting, and the cookie phase-out itself didn't happen: Google kept Chrome's existing approach to third-party cookies, and in October 2025 announced it would retire Topics and most other Privacy Sandbox APIs. Canvas, WebGL, AudioContext and font fingerprinting were never affected, and Client Hints add structured detail on top of the User-Agent string.
What is the Chrome Topics API, and is it still around?
Topics replaced the abandoned FLoC proposal. It assigns your browsing to interest categories from a fixed taxonomy of a few hundred topics, based on the hostnames you visit, and lets sites read some of your top topics from the past three weeks. Google announced its retirement in October 2025, and Chromium's published plan is to deprecate it and then remove it. While it is present, it adds a little identifying information that combines with other signals.
What are Client Hints and why do they increase fingerprinting entropy?
Client Hints (UA-CH) are structured HTTP headers that report what the User-Agent string used to pack into one line: Sec-CH-UA, Sec-CH-UA-Platform, Sec-CH-UA-Mobile, Sec-CH-UA-Full-Version-List, Sec-CH-UA-Arch, Sec-CH-UA-Bitness and others. Several discrete parameters replace one string, and the detailed ones are exactly what makes a browser distinguishable. They are part of how Chrome reports itself, not an advertising API, so they were not part of the Privacy Sandbox retirement.
Should teams managing several accounts worry about Privacy Sandbox changes?
Worry isn't the right framing — plan is. The Privacy Sandbox wind-down changes little for fingerprinting, because the core signals were never affected. Keep each profile's browsing history its own, and make sure UA-CH headers agree with each other and with the User-Agent. A multi-account browser with a native C++ engine handles this better than extension-based tools because the values are generated at engine level, not intercepted in JavaScript.
Try JustBrowser
A multi-account browser for teams, on a native Chromium engine — not extension-based. Real C++ engine patches at the canvas / WebGL / audio / font / screen layer, so a profile's 40+ identity parameters are consistent with each other. Local REST API and CDP for Playwright, Puppeteer, Selenium. 7-day free trial, card required — then $9.99/month or $99.99/year, unlimited profiles, free team seats.
Get started → · How it differs from Multilogin / GoLogin / AdsPower
Ready to manage multiple accounts?
Seven days free, then $9.99/month — one plan, everything included.