What privacy.resistFingerprinting actually does

Firefox ships an anti-fingerprinting mode, and turning it on is one preference away. This is what privacy.resistFingerprinting actually changes, what it breaks, and why an automation tool is better off leaving it switched off.

privacy.resistFingerprinting = true

If you are automating a browser and trying not to be fingerprinted, turning it on looks like the obvious first move. It is usually the wrong one, and the reason is worth understanding even if you never touch the setting, because it is the clearest example of two goals that sound identical and are opposites.

This project sets it to false. On purpose. Here is the reasoning.

What it changes

resistFingerprinting, RFP, came out of the Tor Uplift project: work done for Tor Browser and merged into Firefox. Its job is to make every user of it report the same values, so that no individual can be told apart from the others.

Roughly, with the caveat that the exact list moves between Firefox versions:

Timezone is reported as UTC, whatever your machine says.
Screen and window dimensions are rounded, and letterboxing pads the viewport to a standard size so the window itself carries no entropy.
The user agent and navigator.platform are generalised, and the minor version is dropped.
hardwareConcurrency reports a fixed low number instead of your core count.
Timer precision is reduced, which blunts timing side channels.
Canvas reads are blocked or prompt, so a canvas hash cannot be taken silently.
Locale is forced toward en-US for the APIs that expose it.
Various smaller surfaces: media queries like prefers-color-scheme, some WebGL parameters, keyboard layout, and more each release.

That is a real and well-engineered privacy feature. The complaints about it breaking sites are also real, and they follow directly from the list: a calendar that believes you are in UTC shows the wrong times, a layout that reads the viewport gets bars around it, and anything that needs a canvas read either prompts or fails.

Why it is the wrong tool for automation

Because of what “not fingerprintable” means here.

RFP does not make you look like a typical user. It makes you look identical to every other RFP user. That is exactly right when the population is large and the goal is to disappear into it: this is the Tor Browser model, where uniformity is the whole point and the anonymity set is everybody else running the same build.

Now count the population on the open web. Almost nobody has RFP on. So the signature it produces, UTC timezone regardless of IP, a rounded screen, a suspiciously low core count, a generalised user agent, is not a hiding place. It is a small, sharply defined, easily recognised group, and a site that wants to treat that group differently can do so with one check.

Worse for an automated session: the RFP signature actively contradicts the rest of what you are doing. You are coming from a residential IP in one country and reporting UTC. Your proxy says one locale and your Intl says another. Uniformity with a tiny population plus a plausible IP is not two protections, it is a contradiction, and contradiction is the thing detectors are built to find.

The distinction in one line

RFP optimises for anonymity within a set. An automation session needs plausibility as an individual.

Those are different objectives and they want opposite implementations. Anonymity wants everyone identical; plausibility wants you to be one specific, unremarkable, internally consistent machine, different from the next session, and consistent with the network you appear to be on.

If your threat model is a website building a profile of a human across visits, RFP is a good answer and you should use it. If your problem is a site deciding whether this session is a person, RFP hands it an easier question, not a harder one.

What to do instead

Pick a specific machine and be it, completely and consistently.

That means every surface derives from the same source rather than being set independently: the user agent, the platform, the screen, the GPU strings, the font set, the audio stack, the timezone, the language list. And it means the timezone and locale follow the IP you are actually leaving from, and not either your host’s values or a fixed UTC.

It also means not layering. Turning RFP on and setting a user agent and running a stealth plugin gives three components independently answering the same questions. They will not agree, and the disagreement is louder than any single wrong value.

Concretely, in this project: the five preferences above are all set to false, and the identity comes from a seeded profile instead. Same seed, same machine, every run. Different seed, a different machine that is equally unremarkable.

When RFP is right and this is not

If you are a person who wants privacy from ad networks, use Firefox with RFP on, or Tor Browser, and accept the breakage. That is the tool doing its job, and no automation library is a substitute for it.

The advice here is narrow: it applies when the thing on the other side is asking whether you are a browser and a person, and where being unusual costs you more than being identifiable.

Short answers to the questions that lead here

Should I turn on privacy.resistFingerprinting for automation? No. It makes you uniform with other users of that mode, which is a small and recognisable population, and detectors identify the mode itself.

What does it actually change? Timezone forced to UTC, a rounded viewport, a reduced timer precision, a fixed hardwareConcurrency, canvas and WebGL readings altered, and a restricted font set, among others.

So it is useless? Not at all. It is the right tool for a person who wants privacy from ad networks, and Tor Browser builds on it. It is the wrong tool for looking like an ordinary user.

Why does it break sites? Because it lies to the page about things the page uses for layout and timing. That is the trade it exists to make.

What does this project do instead? Sets the values to a plausible, consistent machine derived from a seed, and leaves the mode off.

See also: how CreepJS decides you are lying, whose resistance module exists to identify this mode, and the three levels.

From the notes of invisible_playwright. The preference values quoted are the ones it actually writes, which you can print yourself with get_default_stealth_prefs.