Privacy Sandbox Topics API: Why Fingerprinting Still Works in 2026

Last week, a colleague sent me Google's latest Privacy Sandbox progress report. "Finally killing tracking," he wrote. "The cookies are dead."

I sent back a screenshot of CreepJS running on Chrome 126 with Privacy Sandbox fully enabled. Still unique. Still fingerprinted. Cookie-free tracking working exactly as it did before.

His response: "Wait, what?"

Yeah. That's the reaction everyone has when they actually test it.

The Privacy Sandbox Promise vs. Reality

Google announced Privacy Sandbox in 2019. The pitch was compelling: kill third-party cookies, end cross-site tracking, give users privacy while preserving ad targeting for publishers. Win-win-win.

Seven years later, third-party cookies are finally blocked in Chrome by default. Privacy Sandbox APIs — Topics, Attribution Reporting, Protected Audience — have shipped. The tracking apocalypse has arrived.

Except... fingerprinting works just fine. Better, actually.

Here's the thing nobody at Google's marketing department wants to acknowledge: Privacy Sandbox doesn't reduce fingerprinting. It shifts the fingerprint vectors and in some cases expands them.

I've tested this extensively — probably spent more hours on this than my spouse thinks is reasonable. Same machine, same browser, same network — run fingerprint checks with Privacy Sandbox enabled vs. disabled. Uniqueness rate? Unchanged. Entropy? Slightly higher with Privacy Sandbox on because of Client Hints.

The cookies are dead. Long live fingerprinting.

Topics API: A New Fingerprint Signal Wearing a Privacy Costume

Remember FLoC? Google's first attempt at cookie-replacement advertising. It clustered users into "cohorts" based on browsing history. Privacy advocates screamed. The EFF called it a privacy disaster. Google killed it in 2022.

Topics API is FLoC's replacement. Supposedly better. Let's look at what it actually does.

Topics analyzes your browsing history and assigns you to up to 5 interest categories from a ~350-topic taxonomy. Sports, personal finance, travel, software, automotive — that kind of thing. When you visit a site running Topics, your browser shares your top topics from the past 3 weeks.

Google says this is privacy-preserving because:

Topics are coarse-grained (only ~350 categories)
One random topic from your five is replaced with a random topic (noise injection)
Topics expire after 3 weeks
You can see and delete your topics

Sounds reasonable. I wanted to believe it, honestly. I really did. Except.

Topics adds fingerprinting entropy. Not much — 2-4 bits in my testing. But those bits combine with everything else. Canvas, WebGL, AudioContext, screen resolution, timezone, Client Hints, Topics. Each parameter narrows the population you match. Topics is another parameter.

And the "noise" injection is detectable. If a site queries Topics multiple times across sessions, the consistent topics reveal your actual interests. The random fifth topic changes; the real four don't. Pattern detection 101.

The EFF's analysis was blunt: Topics tells sites your interests even if you've never visited them before. That's not privacy. That's profiling with extra steps.

For multi-account operations, Topics creates a new consistency problem. If you're running 50 browser profiles, each needs to generate Topics that match its supposed identity. An affiliate profile browsing auto insurance sites shouldn't have topics like "baby products" and "wedding planning." Detection systems can cross-reference Topics against browsing behavior for profile consistency checks.

JustBrowser's cookie warm-up engine populates realistic browsing history across profiles — which now also shapes Topics assignments. Pre-warming browsing history isn't just about cookies anymore. Our WebGL fingerprinting deep dive explains how GPU signals combine with Topics for cross-parameter consistency.

Client Hints: More Headers, More Entropy

User-Agent strings were messy. One long string containing browser name, version, OS, device info — crammed together, inconsistently formatted across browsers. Hard to parse. Ripe for deprecation.

Chrome's replacement: Client Hints (UA-CH). Instead of one string, Chrome sends structured headers:

Sec-CH-UA — browser brand and major version
Sec-CH-UA-Platform — operating system
Sec-CH-UA-Mobile — mobile or desktop
Sec-CH-UA-Full-Version-List — complete version numbers
Sec-CH-UA-Arch — CPU architecture (x86, arm)
Sec-CH-UA-Bitness — 32-bit or 64-bit
Sec-CH-UA-Model — device model (mobile only)
Sec-CH-UA-Platform-Version — OS version

That's 8-12 discrete parameters. Structured data. Easy to parse. Easy to fingerprint. I genuinely don't understand how this shipped as a "privacy" feature — someone in the room must have said "uh, won't this make fingerprinting easier?" and got overruled.

The old User-Agent contributed maybe 4-6 bits of entropy on its own. Client Hints? I've measured 6-10 bits in practice. Architecture + bitness + platform version + full version list creates more unique combinations than the old jumbled string.

Google's justification: sites need to know client capabilities for compatibility. True enough. But the fingerprinting implications weren't accidental. They just weren't prioritized.

What this means for antidetect browsers: you need to spoof all UA-CH headers consistently. Return x86 architecture but arm-specific Client Hints? Inconsistency detected. Chrome 126 User-Agent but Chrome 124 in Sec-CH-UA-Full-Version-List? Inconsistency detected.

Extension-based antidetect tools struggle here because Client Hints are sent as HTTP headers before JavaScript even runs. You can't intercept them in content scripts. Native browser modifications — like JustBrowser's C++ engine patches — generate consistent UA-CH values from the same source that produces the User-Agent. One truth, multiple outputs. Our Playwright and Puppeteer REST API guide covers programmatic profile control with consistent UA-CH headers.

What Privacy Sandbox Didn't Touch

Here's the list of fingerprinting vectors unchanged by Privacy Sandbox:

Canvas fingerprinting — still works. GPU-specific rendering differences still expose hardware.
WebGL parameters — still exposed. MAX_TEXTURE_SIZE, shader precision, extensions — all queryable.
AudioContext fingerprinting — untouched. Audio processing variations still distinguish machines.
Font enumeration — partially limited but still functional. System fonts vary by installation.
Screen resolution and DPR — unchanged. Window size, device pixel ratio, color depth.
Timezone — unchanged. ~38 zones with uneven population distribution.
Navigator properties — platform, hardwareConcurrency, deviceMemory — all still there.
WebRTC — local IP and media device enumeration still possible unless explicitly blocked.

These vectors represent 30-40+ bits of entropy. Privacy Sandbox didn't address any of them. Not a single one. Third-party cookies? Gone. Fingerprinting fundamentals? Untouched.

FingerprintJS Pro's Q1 2026 benchmark shows 86% of Chrome users remain uniquely identifiable through fingerprinting alone, even with Privacy Sandbox fully enabled. Cookies or not, you're trackable.

For ad verification and fraud detection, this is actually useful. ClickzProtect and similar tools rely on device fingerprinting to identify fraudulent click sources — our JA4 TLS fingerprinting cross-product analysis covers how TLS fingerprints complement browser signals. Privacy Sandbox blocking cookies doesn't affect click fraud detection — the device fingerprint remains. The audio and font fingerprinting guide covers the full entropy breakdown by signal.

The Real Winners and Losers

Winners:

Google's ad network. Privacy Sandbox APIs are designed to work with Google's systems. Topics feeds into Google's interest targeting. Attribution Reporting replaces conversion tracking that... Google operates. Chrome controls the APIs. Funny how that works.
Large platforms with first-party data. Facebook, Amazon, Netflix — they don't need third-party cookies. You're logged in. They know who you are. Privacy Sandbox doesn't affect them.
Fingerprinting vendors. FingerprintJS, HUMAN Security, Cloudflare's bot detection — all still work. Maybe better, since users assume Privacy Sandbox means they're not being tracked.

Losers:

Small publishers relying on third-party ad networks. Cookie-based retargeting dies. Topics is coarse-grained. Attribution gets fuzzier. Revenue drops.
Privacy advocates expecting actual privacy. Topics API is profiling. Client Hints expand fingerprinting surface. The "privacy" in Privacy Sandbox is marketing, not engineering. Call me jaded.
Users who think enabling Privacy Sandbox makes them anonymous. It doesn't. You're still unique. You're still tracked. The mechanism changed. The outcome didn't. Sorry.

What Operators and Engineers Should Actually Do

Alright, practical advice time.

If you're running multi-account operations:

Handle Topics API proactively. Either clear Topics data regularly (chrome://settings/adPrivacy/topics in Chrome) or ensure your profile browsing history generates Topics consistent with the profile's intended identity. Don't let Topics leak inconsistent interests across profiles.
Ensure Client Hints consistency. All UA-CH headers must match. Architecture, bitness, platform version, full version list — if one contradicts another, detection fires. Native antidetect browsers generate these from the same source. Extension-based tools often miss headers.
Test against post-Sandbox detection. CreepJS and Pixelscan have added Topics and UA-CH checks. Run your profiles through updated tests. Legacy test passes don't mean current coverage. Our CreepJS test walkthrough covers methodology.
Don't assume cookies gone means tracking gone. Your threat model should assume fingerprinting continues unchanged. Because it does.

If you're building detection systems:

Topics API as a signal. Query Topics where available. It's low entropy but adds information. Cross-reference Topics against claimed user behavior for consistency checks.
Weight Client Hints properly. UA-CH headers are structured and reliable. Architecture + bitness + platform version is surprisingly discriminating. Add these to your entropy calculations.
The traditional fingerprint surface remains. Canvas, WebGL, Audio — all still work. Privacy Sandbox didn't affect them. Don't assume you need to rebuild your detection stack.

If you care about personal privacy:

Firefox with Resist Fingerprinting mode. Or Brave. Or Tor Browser. Not Chrome with Privacy Sandbox. Chrome's "privacy" features protect Google's ad business, not you.

I know that sounds cynical. Maybe I've been doing this too long. I've spent a lot of time hoping Google would prove me wrong. They haven't.

An Unpopular Opinion

Here's my take that might get pushback: Privacy Sandbox is regulatory theater.

The EU's Digital Services Act and GDPR put pressure on cookie-based tracking. California's CCPA added more. Google needed to show regulators they were Doing Something about privacy. Privacy Sandbox is that something.

But examine the outcome. Does fingerprinting decrease? No. Does Google retain targeting capability? Yes. Do competitors lose targeting capability? Also yes. Do users gain meaningful privacy? Not particularly.

That's a regulatory compliance project, not a privacy initiative. The name is marketing.

I'd love to be wrong. I'm open to evidence that Privacy Sandbox meaningfully reduces user tracking. I've looked. The entropy data says otherwise.

The Fingerprinting Arms Race Continues

For operators running multi-account setups, nothing fundamental changed. Fingerprint spoofing remains essential. Native browser modifications beat extensions. Profile consistency across all parameters — now including Topics and expanded UA-CH — matters.

If anything, the post-cookie world advantages antidetect tools that handle fingerprinting deeply. Cookies were easy to isolate. Every tool could do it. My cat could isolate cookies. Fingerprint consistency across 40+ parameters? Harder. Native C++ implementations outperform API interception at scale.

JustBrowser runs Chromium with engine-level patches. Topics, Client Hints, canvas, WebGL, AudioContext — values generated at the C++ layer, not intercepted in JavaScript. That's why we consistently pass CreepJS, Whoer, IPHey and BrowserLeaks where tools working above the engine fail edge cases.

Our antidetect browser myths debunked covers where detection is heading — behavioral analysis, ML scoring, WebGPU. Privacy Sandbox is one change among many. The direction is consistent: static fingerprinting evolves but remains viable. The game continues.

JustAnalytics takes the opposite approach — privacy-first analytics that doesn't rely on fingerprinting. Different problem, different solution. But both exist because the Privacy Sandbox didn't actually solve tracking.

Frequently Asked Questions

Does Privacy Sandbox actually reduce browser fingerprinting?

No, and that's the uncomfortable truth. Privacy Sandbox blocks third-party cookies but introduces new fingerprinting vectors. Topics API exposes 2-4 bits of interest-category entropy. Client Hints (UA-CH) replaced the single User-Agent string with 8-12 structured headers — more parameters, not fewer. WebGL, canvas, and AudioContext fingerprinting remain untouched. The net fingerprinting surface is roughly the same or larger.

What is the Chrome Topics API and how does it affect tracking?

Topics API replaced the failed FLoC proposal. Instead of cohort IDs, it categorizes your browsing into ~350 interest topics based on hostnames visited. Websites can query your top 5 topics from the past 3 weeks. Each topic is drawn from a fixed taxonomy — sports, finance, travel, etc. For fingerprinting, Topics adds 2-4 bits of entropy. Not high, but combinable with other signals. And it's opt-out, not opt-in.

What are Client Hints and why do they increase fingerprinting entropy?

Client Hints (UA-CH) are Chrome's replacement for the User-Agent string. Instead of one string containing browser/OS info, Chrome now sends structured headers: 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. That's 8-12 discrete parameters where one existed before. More structure means more unique combinations — increased fingerprinting entropy, not decreased.

Should multi-account operators worry about Privacy Sandbox changes?

Worry isn't the right framing — adapt is. Privacy Sandbox shifts the fingerprinting vectors, not the risk level. Topics API exposure needs handling (clear topics or use profiles that generate consistent topic sets). Client Hints require spoofing consistency across all UA-CH headers. Antidetect browsers with native C++ implementations handle this better than extension-based tools because the values are generated at engine level, not intercepted in JavaScript.

Try JustBrowser

Native Chromium antidetect browser — not extension-based. Real C++ engine patches at the canvas / WebGL / font / TLS layer, so 40+ identity parameters are genuine, not faked. REST API 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