Web Workers: where page-level fingerprint patches fail

If you want to understand why patching the page has a ceiling, workers are the clearest demonstration available. A worker is a separate JavaScript realm. Your init script ran in the document. The worker did not inherit it, cannot see it, and answers every question from the browser’s own code.

So a page that wants the truth does not have to defeat your patch. It can ask somewhere your patch never went.

This page is what a worker exposes, the canvas path that bypasses page-level protection entirely, why this is the sharpest case against that layer, what an engine-level value does differently, and how to check your own.

A worker is a separate realm with its own navigator

const w = new Worker(URL.createObjectURL(new Blob([`
  postMessage({
    ua: navigator.userAgent,
    cores: navigator.hardwareConcurrency,
    lang: navigator.language,
    platform: navigator.platform,
    mem: navigator.deviceMemory,
  });
`])));
w.onmessage = e => console.log(e.data);

That runs in a context created by the browser, from the browser’s own values. If you redefined navigator.hardwareConcurrency on the document, this reports the real number, and now the page has two different answers to one question.

The comparison is the finding, not either value. An unusual core count is a weak signal. A browser that reports one number in the document and another in a worker has told the page that something is rewriting values in one context and not another, which is a tampering result rather than a hardware result.

What a worker can read

WorkerNavigator is a subset of Navigator, and the subset is exactly the part that matters here: userAgent, platform, language and languages, hardwareConcurrency, deviceMemory where implemented, onLine, and the storage estimate. Plus performance.now() for timing.

That is most of what a cheap fingerprint is made of, available in a place page-level patches do not reach.

OffscreenCanvas: the canvas path that bypasses the patch

This is the strongest version of the problem, and it is not theoretical.

OffscreenCanvas lets a worker create and draw to a canvas with no DOM element involved, then read the pixels back:

// inside a worker
const c = new OffscreenCanvas(300, 60);
const ctx = c.getContext('2d');
ctx.font = '14px Arial';
ctx.fillText('abcdefghijklmnop', 2, 20);
const bmp = c.transferToImageBitmap();     // or convertToBlob()

Two things follow, and both are documented in the wild:

• Protection that hooks HTMLCanvasElement.prototype misses this entirely, because no HTMLCanvasElement is involved. Most bypass and protection scripts hook the element methods and not the OffscreenCanvas equivalents, so a fingerprint taken this way comes back unmodified.
• It runs off the main thread, so it does not block anything and there is no timing cost to doing it.

Browsers have had to catch up here too. Canvas protections in at least one engine did not apply inside shared and service workers for a period, and Firefox tracked its own OffscreenCanvas gap as a bug to be fixed in a specific version. The surface is young enough that “it is covered” is a claim worth testing rather than assuming.

Why this is the sharpest case against page-level patching

Every argument on the three levels page applies here, and this is the example that makes it concrete.

A page-level patch is code running in one realm. Workers are additional realms, created on demand, as many as the page likes. To cover them a patch would have to:

• intercept worker creation, including from a blob URL and from a data URL,
• inject itself into each one before the worker’s own code runs,
• cover SharedWorker and ServiceWorker as well as Worker,
• and patch the worker-side APIs, including the OffscreenCanvas methods, not only the document-side ones.

Each of those is possible and each is another surface that can be inspected, which is the same treadmill as patching toString.

Meanwhile a value decided in the engine is simply the same value everywhere, because every realm asks the same compiled code. There is no injection, no ordering, and no list of contexts to remember.

That is the whole argument for the level, stated once: not that patching cannot work, but that it has to be complete, and completeness here is an open-ended list.

What an engine-level value does differently

In this project the worker-side navigator values come from the same profile as the document-side ones, set in the browser’s own worker navigator code rather than injected. So the two answers agree because they are the same answer.

The canvas path is handled the same way: the substitution happens where pixels are read back inside the engine, so it applies wherever the readback happens, in the document or in a worker, through an element or through OffscreenCanvas.

Being straight about the limit, as everywhere: this makes the values consistent. It does not make them plausible. A worker reporting the same low core count as the document is consistent and still describes a small virtual machine, which is a different problem.

Checking your own

Two comparisons, both short.

// 1. does the worker agree with the document?
const w = new Worker(URL.createObjectURL(new Blob([`
  postMessage([navigator.hardwareConcurrency, navigator.userAgent,
               navigator.platform, navigator.language]);
`])));
w.onmessage = e => console.log('worker:', e.data,
  '\ndocument:', [navigator.hardwareConcurrency, navigator.userAgent,
                  navigator.platform, navigator.language]);

// 2. does a canvas drawn in a worker match one drawn in the document?
// draw the same shapes both ways and compare the resulting bytes

What you want: identical answers from both places, and a canvas that is treated the same way in both. A difference in either is not a weak signal, it is a direct statement that something is intervening.

Then do the same in a stock browser to see what agreement is supposed to look like, which is the comparison method.

Conclusion

Workers are not an exotic corner. They are a normal part of the platform, trivial to create, and they answer from the browser rather than from your patch. OffscreenCanvas extends that to the single most-used fingerprinting surface there is.

If you are evaluating a stealth tool, this is the cheapest meaningful test you can run: ask it something in the document, ask the same thing in a worker, and see whether the answers match. If they do not, you have learned where its ceiling is in two lines.

Short answers to the questions that lead here

Can a website fingerprint me from a Web Worker? Yes. WorkerNavigator exposes the user agent, platform, languages, core count and memory, and OffscreenCanvas allows canvas fingerprinting from inside one.

Do page-level stealth patches cover workers? Not unless they explicitly intercept worker creation and inject into each one. Most do not.

Does OffscreenCanvas bypass canvas protection? It bypasses protection that hooks the HTML element’s methods, which is most of it.

Why does my worker report a different value than the page? Because your override was applied to the document’s realm and the worker has its own.

Is this a real technique or a theoretical one? Real. Engines have shipped fixes for their own gaps in worker canvas protection, which is evidence that it was being used.

How do I test a stealth tool for this? Read the same value in both places and compare. Two lines.

See also: three ways to make Playwright undetected, Function.prototype.toString and the native code check, canvas fingerprint noise, and hardwareConcurrency, deviceMemory and storage quota, whose worker check is this idea applied to one value.

Sources

• MDN for OffscreenCanvas, WorkerNavigator and the worker types.
• Public reports of canvas protections not applying inside shared and service workers in one engine, and Firefox’s own tracked OffscreenCanvas gap, for the claim that this is actively used.
• This project sets the worker-side navigator values in the engine’s worker navigator code, and applies canvas substitution at readback rather than at the element.

---

From the notes of invisible_playwright, a Firefox patched at the C++ level. The two-line worker comparison is the test we would run first on anything claiming to be undetectable, including ours.