Privacy Sandbox APIs Are a New Fingerprint Surface, and Most Anti-Detect Setups Ignore It
Google built Privacy Sandbox to replace third-party cookies without breaking ad targeting. The pitch is that browsing data stays on-device instead of getting shipped to ad networks. That part is largely true. What gets less attention is that the APIs doing this work are themselves new, queryable, JavaScript-visible surfaces, and most fingerprinting discussions still center on canvas, WebGL, audio context, and font lists. If you run multiple browser profiles for separate accounts, it’s worth understanding what these APIs expose and where they can go wrong for you.
What privacy sandbox actually is
Privacy Sandbox is a set of browser APIs, mostly shipped in Chrome and Chromium-based browsers that opt in, meant to support advertising use cases (audience targeting, conversion measurement, fraud prevention, federated login) without third-party cookies. The main pieces are the Topics API, the Protected Audience API (formerly FLEDGE), the Attribution Reporting API, Private State Tokens (formerly Trust Tokens), FedCM, and an IP-masking relay sometimes called IP Protection. Each one runs computation inside the browser instead of sending raw history to a server. That’s the privacy improvement. But “computation inside the browser, exposed through a JS API” is also the exact shape of every fingerprinting vector that came before it.
The topics API turns your history into a queryable label
The Topics API is the clearest example. Chrome classifies the sites you visit into a fixed taxonomy of a few hundred topics, using a model that runs locally. Any site can call document.browsingTopics() and get back up to a handful of topic IDs drawn from your last few weeks of browsing, one per week, with some of them replaced by random noise to blunt precision. That noise is the privacy mechanism, but it doesn’t remove the signal, it just adds error to it. A script that queries this repeatedly across sessions, or correlates it with other signals on the same page, is doing exactly what a fingerprinting script does: building a profile out of values that are supposed to be too coarse to identify anyone individually.
For multi-accounting, the practical risk isn’t that Topics identifies you by name. It’s that two “separate” browser profiles running on the same machine can end up returning overlapping or identical topic sets if the underlying topic model, which is computed and stored per Chrome profile, isn’t actually isolated between the identities you think are isolated. That’s a correlation bug, not a privacy feature working as intended, and it’s worth testing for directly rather than assuming a fresh profile means a fresh topic history.
Protected audience, attribution reporting, and FedCM
Protected Audience runs ad auctions inside an isolated worklet so a site can retarget you without a third party seeing your interest-group membership directly. The interest groups themselves, though, are stored per-browser and persist across sessions. Their presence and contents are a new piece of local state that didn’t exist before this API shipped, and local state that persists across page loads is, by definition, something a script can try to read or infer from timing side channels even when direct access is blocked.
Attribution Reporting handles ad conversion measurement by storing attribution data in the browser and reporting it in aggregate, with delays and noise added specifically to prevent it from being used as a precise cross-site tracking channel. FedCM changes how federated login works: instead of an identity provider setting a tracking cookie in a third-party context, the browser itself mediates the login request and can tell a site whether you’re signed in to a given IdP. That “are you logged in here” signal is new, browser-native, and something a detection script can probe for even if you never click “sign in with Google” on that particular tab.
None of these are designed as fingerprinting tools. But every one of them adds a queryable bit of state that a Chromium profile spoofed to look like stock Chrome now needs to reproduce correctly, or its absence becomes the tell.
Private state tokens and IP protection
Private State Tokens let a site that trusts you (because you passed some earlier check) issue you a token, which you can later redeem elsewhere to prove “a trusted issuer vouched for this browser” without revealing which browser you are. It’s meant for anti-fraud use cases like distinguishing humans from bots without a CAPTCHA. The risk for anyone running many profiles is that token issuance and redemption both touch persistent per-profile storage, and if that storage isn’t cleanly separated between identities, tokens can become an unintended linkage between accounts that are supposed to be unrelated.
IP Protection routes some third-party requests through Google-operated relays so the visited site can’t see your real IP for certain traffic, while Google’s relay still can. If you’re already routing traffic through your own proxy infrastructure, a browser that’s also trying to double-hop specific requests through Google’s relay creates its own timing and routing footprint, one that a well-built detection stack can watch for.
Why this matters specifically for anti-detect browsers
Multilogin, GoLogin, Kameleo, AdsPower, Dolphin Anty, and similar tools spend most of their engineering effort on the fingerprint surfaces that detection vendors have been checking for years: canvas noise, WebGL renderer strings, audio fingerprints, installed fonts, screen and hardware concurrency values, timezone and locale consistency. That’s the right priority, because those are still the highest-value signals in production anti-fraud systems today.
Privacy Sandbox APIs are newer, Chrome-specific in large part, and not yet a standard line item in most fingerprinting checklists, on either the detection side or the anti-detect side. That’s exactly what makes them worth watching. A profile spoofed to present as “Chrome 128 on Windows” that doesn’t implement browsingTopics() at all, or throws a different error than real Chrome does when the permissions policy blocks it, or always returns an empty topic array regardless of browsing history, is showing a mismatch between its claimed identity and its actual behavior. Detection scripts that specifically probe Privacy Sandbox APIs and compare the result against what a real Chrome install of that version would produce are a logical next step for anti-fraud vendors, because it’s an unaddressed gap.
What to actually check before you trust a profile
If you’re running an anti-detect browser for multi-accounting, don’t take “it spoofs canvas and WebGL” as the full picture anymore. Open the devtools console on a fresh profile and check whether document.browsingTopics exists, whether calling it behaves the way it does on a genuine Chrome install of the version you’re claiming, and whether the result differs across profiles that are supposed to be unrelated identities. Do the same for navigator.credentials.get() with FedCM, and check whether Private State Token storage is actually partitioned per profile rather than shared at the browser-engine level.
None of this makes a browser undetectable, and no anti-detect tool on the market can promise that, regardless of what its marketing says. What it does is close a gap between what your profile claims to be and what it actually does when a script asks. That gap, not the marketing claims, is what detection systems are built to find. Understanding how these APIs work is the difference between running profiles that hold up to scrutiny and profiles that look fine until someone checks the one thing you didn’t.
If you want more breakdowns like this on how browser fingerprinting actually works and how the major anti-detect tools hold up when we test them hands-on, head back to the homepage.
Get new guides and videos first — join the Telegram channel.