Does Playwright Set navigator.webdriver to True?

Short version: yes. A browser launched by stock Playwright reports navigator.webdriver === true, and it has done so for years. That single boolean is one of the oldest and cheapest automation tells there is, which is exactly why so many people try to patch it and why the patch usually makes things worse.

This page is what the flag is, why setting it to false from JavaScript is itself a signal, what a native false looks like to a detector, and the honest limit of what clearing it buys you.

The one-line answer, and the check that proves it

The flag is specified. The WebDriver standard says a browser under automation control must expose navigator.webdriver as true, so Playwright, Selenium and any other WebDriver-based driver all set it. Read it yourself:

from invisible_playwright import InvisiblePlaywright

with InvisiblePlaywright(seed=42) as browser:
    page = browser.new_page()
    page.goto("https://example.com")
    print(page.evaluate("() => navigator.webdriver"))

On a plain Playwright launch of the same engine that prints True. The value above comes back False, and the rest of this page is about why the difference between those two lines is not one character of JavaScript.

A clean, human-driven browser reports false here. Note that this is different from the Chromium story, where an untouched browser reports undefined and a hand-set false is itself an anomaly. On Firefox the honest value is a plain false, which is what this engine returns.

Why that boolean is such an old tell

It costs a detector nothing. One property read, no timing, no fingerprint math, no iframe. It is the first line of almost every detection script ever written, and it catches the enormous number of scrapers that never touch it at all.

Because it is so cheap, it is also the first thing every serious tool fixes, which means that by itself it separates almost nobody from almost nobody. It matters for the reverse reason: leaving it true marks you instantly, so it is table stakes rather than an advantage. The interesting question is not whether the flag reads false, it is how it came to read false, because that is where the second tell lives.

Why a JavaScript patch of the flag is itself detectable

The obvious fix is to overwrite the property from a script that runs before the page:

// the classic, and the trap
Object.defineProperty(navigator, "webdriver", { get: () => false });

This does make navigator.webdriver return false. It also leaves two fresh fingerprints that are easier to check than the original flag.

The getter is no longer native code. In a real browser the property is backed by an internal getter implemented in C++. Ask that getter to describe itself and it says so:

Object.getOwnPropertyDescriptor(Navigator.prototype, "webdriver")
  .get.toString();
// real browser:  "function get webdriver() { [native code] }"
// JS override:   "() => false"       <-- the whole source, in plain sight

Every function in JavaScript carries its own source, and Function.prototype.toString hands it back. A native builtin answers with the [native code] placeholder; your arrow function answers with its actual body. So the override that was supposed to hide automation has instead written the word false into a place a detector can read directly. This is the general ceiling on page-level stealth, and it is worth understanding on its own terms in the note on toString and the native-code check.

The property moved to the wrong object. On a real Firefox the webdriver accessor lives on Navigator.prototype, not on the navigator instance. defineProperty on navigator puts an own property on the instance, so navigator.hasOwnProperty("webdriver") flips from false to true and the descriptor now sits one rung lower on the prototype chain than it should. A detector that walks the chain, which the tampering-focused suites do by design, sees the shape change even when the value is correct.

The pattern is the one that runs through the whole checklist for being flagged on a single site: a value fixed in isolation contradicts a value nobody thought to fix. Setting the boolean right while leaving the getter’s source and its location wrong is two new tells in exchange for one old one.

What a native false looks like to a detector

The way out is not a better shim, it is to not have a shim. This engine is a Firefox patched at the C++ level, so navigator.webdriver is false because the browser’s own internals return false, the same code path a normal Firefox uses. There is no JavaScript getter to inspect and no instance property to trip over:

from invisible_playwright import InvisiblePlaywright

with InvisiblePlaywright(seed=42) as browser:
    page = browser.new_page()
    page.goto("https://example.com")
    report = page.evaluate("""() => {
        const desc = Object.getOwnPropertyDescriptor(
            Navigator.prototype, "webdriver");
        return {
            value: navigator.webdriver,
            ownProp: navigator.hasOwnProperty("webdriver"),
            getterSource: desc && desc.get.toString(),
        };
    }""")
    print(report)
    # {'value': False,
    #  'ownProp': False,
    #  'getterSource': 'function get webdriver() { [native code] }'}

All three answers match a browser nobody automated: the value is false, the property is not an own property of the instance, and the getter still reports [native code]. There is nothing here for a descriptor walk or a toString probe to catch, because the thing they look for was never added. The same reasoning, at more length and with the history of the flag, is in why navigator.webdriver is not the tell you think it is.

The honest caveat: necessary, not sufficient

Clearing this flag correctly is table stakes, and table stakes are not a win. A false navigator.webdriver is one boolean among dozens of signals, and it says nothing about the surfaces that actually get datacenter automation caught:

The IP. A perfectly clean browser on a known or datacenter address still loses, and no browser property changes that. You supply the exit.
Rate and volume. Per-account quotas, request velocity and how many sessions share one address are counted server-side, where the driver flag is invisible.
Behaviour and timing. A pointer that teleports, a form filled in eighty milliseconds, keystrokes at a metronome interval. This engine draws mouse motion on a Bezier curve, but the pacing of your actions is yours to make human.
The rest of the fingerprint. GPU, fonts, audio device, screen, the TLS handshake. These read as a genuine Firefox here, which is the whole design, but they are a system you keep consistent, not a switch you flip once.

invisible_playwright is built to look like a real browser driven by a real person, which is why it passes most detection checks at the fingerprint, TLS and driver layers. It does not, and cannot, fix a bad IP, a blown quota or robotic timing. The method for actually testing which layer is failing is worth more than any single flag reading, because it tells you whether the thing you fixed was ever the thing catching you.

Conclusion

Stock Playwright sets navigator.webdriver to true by specification. Patching it from JavaScript trades one obvious tell for two subtler ones: a getter whose source reads false under toString, and a property that has migrated onto the wrong object. A native false, returned by the browser’s own C++ path, leaves neither, which is why this engine returns false without a shim to catch. Treat that as necessary and never as sufficient: it clears the driver layer, and the IP, the pacing and the behaviour are still yours to get right.

Short answers to the questions that lead here

Does Playwright set navigator.webdriver to true? Yes, on every WebDriver-based launch, because the standard requires an automated browser to report true.

Can I just set navigator.webdriver to false in a script? You can, and the value will read false, but the override’s getter no longer reports [native code] and the property lands on the wrong object, so you have added two tells to remove one.

How do detectors catch the JavaScript override? By reading the getter’s source with Function.prototype.toString, and by walking the prototype chain to see that the property moved onto the instance.

What should a clean Firefox report? A plain false. That is different from Chromium, where an untouched browser reports undefined and a hand-set false is itself odd.

Does invisible_playwright fake the value? No. It returns false from the browser’s own native code path, so there is no JavaScript getter to inspect and no own property to find.

If navigator.webdriver is false, am I undetectable? No. It is one boolean among many signals. It does nothing for your IP reputation, your rate limits or your behaviour, all of which you still have to supply.

Sources

The WebDriver specification, which defines navigator.webdriver and requires an automated user agent to expose it as true.
Function.prototype.toString, whose [native code] behaviour for builtins is the standard mechanism behind the getter-source check.
This project’s own driver-layer checks, comparing a native false against a JavaScript override on the same engine.

Written while maintaining invisible_playwright, a Firefox patched at the C++ level driven by stock Playwright. The flag reads false because the browser’s own code returns false, not because a script overwrote it.