can you run playwright without being detected

Can You Run Playwright Without Being Detected?

Not in an absolute sense, and any tool that promises otherwise is lying to you.

What is achievable is narrower and more useful: you can remove the browser-level tells so the fingerprint and the automation flags read like a real person's browser, and that alone clears most JavaScript-based detection. The rest of what gets a session flagged is not a browser problem at all, and no amount of browser work touches it.

This page draws that line precisely. It shows what "undetected" can honestly mean, what a real-browser fingerprint clears, and the three signal classes that are detected independently of the browser and that you, not the tool, have to supply an answer for.

The honest answer: "looks like a real browser", not "invisible"

The useful reframing is to stop asking "can I be undetected" and start asking "does this look like a real browser driven by a real person". That is a question you can actually answer yes to, and it is the one that most detection is really asking.

A detector almost never has a single "is this a bot" bit. It has a pile of signals and a score. Some of those signals live in the page, in JavaScript: the fingerprint, the automation flags, whether any built-in has been tampered with. Some live below the page: the network handshake, the exit address. Some live in what the session does over time: pacing, motion, request velocity.

A tool that drives a genuine, unmodified-looking browser can make the first pile read clean. It cannot make the other two piles read clean, because they are not properties of the browser. Being honest about that split is the whole point of this page, and it is why passing every public suite still does not mean the session passes.

What "detected" actually means: three layers of signal

Before touching code, it helps to name the three layers, because a fix for one does nothing for the others.

1. The browser layer. The fingerprint (GPU, canvas, audio, fonts, screen), the automation flags (navigator.webdriver and friends), and internal consistency (does the platform you claim agree with the fonts you have). This is in JavaScript's reach, and it is the layer a stealth engine addresses.
2. The network layer. The TLS handshake, the HTTP/2 settings frame, header order, and the reputation of the exit address. Decided before a line of page script runs, and invisible to every in-page test.
3. The behaviour layer. Pointer motion, typing rhythm, how fast a form is filled, how many requests per minute from one account, and the timing shape of an automated agent.

invisible_playwright is built to solve layer 1 well, and it drives a real Firefox binary, which means layer 1's network handshake is a genuine Firefox handshake rather than an impersonation. What it deliberately does not do is manufacture a clean IP, a per-account quota, or human pacing. Those are yours to bring.

What invisible_playwright removes: the browser layer

The product is a Firefox patched at the C++ level, driven by stock Playwright, with a fingerprint derived from a seed so it is real-browser-shaped and reproducible. Every session gets a coherent identity (roughly 400 fields that agree with each other) instead of the default automation profile that detectors are trained on.

Switching from plain Playwright is a two-line change, and the object you get back is a real Playwright Browser, so every method works exactly as documented upstream:

from invisible_playwright import InvisiblePlaywright

# seed=42 makes the identity reproducible; drop it for a fresh one each run
with InvisiblePlaywright(seed=42) as browser:
    page = browser.new_page()
    page.goto("https://example.com")
    # navigator.webdriver reads undefined, the fingerprint is internally
    # consistent, and the TLS handshake is a real Firefox handshake
    print(page.evaluate("navigator.webdriver"))   # -> None

Add a proxy and let the browser timezone follow the exit automatically, which is how you keep the browser and the network from telling two different stories:

proxy = {"server": "socks5://gate.example.com:1080", "username": "u", "password": "p"}

with InvisiblePlaywright(seed=42, proxy=proxy) as browser:
    page = browser.new_page()
    page.goto("https://example.com/login")
    page.fill("#user", "someone")
    page.click("#submit")   # the pointer arcs to the button on a curve

The measurable result of layer 1 being clean: open the same fingerprinting page in this browser and in a stock Firefox on the same machine, diff the reports field by field, and the difference collapses to per-session noise (the IP and the random canvas jitter a real browser also has). The JavaScript detectors read it as a genuine Firefox because, at the layer they can see, it is one. That is why it passes most of them. It is also the ceiling of what the browser layer can do.

The three signal classes it deliberately does not touch

This is the part most "undetectable" claims skip. These three are detected independently of the browser, the tool does not address them by design, and a clean fingerprint on top of any one of them still loses.

• IP reputation. A perfect browser on a datacenter range, or on a residential exit a thousand other people are sharing this minute, is still on a flagged address, and that is decided before your script runs. Around 90 percent of cheap proxies are public and already known. The fix is a clean exit, which is your input, not the browser's output. WebRTC through a proxy is the related browser-visible surface, but the reputation itself is not a browser property.
• Per-account rate limits and quotas. If one account does in an hour what a person does in a week, no fingerprint hides it, because the signal is the account's own history, not the browser's identity. Spreading work across accounts and staying under human volume is an application-design decision the tool cannot make for you.
• Behaviour and timing. Uniform keystroke intervals, a form filled in eighty milliseconds, a pointer that teleports, a request cadence with no human jitter, or the pause shaped exactly like model latency in an agent loop. The engine gives you curved mouse motion for clicks, but the overall rhythm of the session is driven by your code.

None of these is a gap or a bug. They are outside the browser, so they are outside what a browser-fingerprint tool can honestly claim. Pairing a real-browser fingerprint with a clean exit and human pacing is the actual recipe; the tool is one of those three parts.

How to check what is left over

Because the browser layer is handled, your testing effort should go to the two layers that are not. The method that catches the most:

• Prove it is the browser before blaming it. Open the target by hand from the same machine and network. If the manual visit also gets the wrong page, this is the IP or the account, and no browser change helps. This ordering is the whole point of the detected-on-one-site checklist.
• Compare against a stock browser, do not read a score. Diff the fingerprint fields against a real Firefox on the same machine. Anything that matches is not your problem whatever the verdict says; anything that differs, other than the address, is a candidate.
• Assert presence, not absence. A blocked or empty signal is itself a tell and a failure, not a pass. Why a green verdict can mean the feature is broken is the trap here.
• Watch the score, not just the pass line. A trust or confidence score tells you more than a binary verdict, and it is where residual behaviour and IP signals show up even when the fingerprint is clean.

Run any of this at least ten times, through the proxy you actually deploy with, on the machine you actually deploy on.

Conclusion

Can you run Playwright without being detected? You can make it look like a real browser driven by a real person, and that clears most JavaScript-based detection, which is what invisible_playwright is built to do: a patched Firefox, a seed-consistent real-browser fingerprint, and a genuine handshake because the browser is genuine. What it does not do, on purpose, is fix your IP reputation, your per-account quotas, or your behaviour and timing. Those are detected independently, and you supply the answers: a clean exit and human pacing. Treat "undetectable" as a slogan and you will overclaim; treat it as "looks like a real browser, plus the two layers I own", and you will actually pass.

Short answers to the questions that lead here

Can Playwright be truly undetectable? No, and any tool that says so is lying. You can remove the browser-level tells so the fingerprint and flags look like a real person's browser, which clears most JavaScript detection. IP, account limits and behaviour are detected separately.

Why does invisible_playwright pass most checks then? Because at the layer those checks can see, it is a real Firefox: real fingerprint, no automation flags, a genuine handshake. Most public detection is a JavaScript check, and that is exactly the layer it addresses.

If my fingerprint is clean, why am I still blocked? Almost always the IP, the account's own request history, or the behaviour. Those are not browser properties, so a clean browser does not touch them.

Does it fix my proxy or my IP reputation? No. It deliberately does not touch the exit address. A perfect browser on a known-bad IP still loses, and you supply the clean exit.

Does it make my automation behave like a human? It gives you curved mouse motion for clicks, but the pacing, typing rhythm and request volume come from your code. That is the behaviour layer, and it is yours.

Is a passing sannysoft or CreepJS result enough to be safe? No. It means the browser layer is internally consistent and does not announce automation. The network and behaviour layers are not in those tests at all.

Sources

• This project's release gates, which compare the browser's fingerprint field by field against a stock Firefox on the same machine and treat a suppressed signal as a failure rather than a pass.
• The public detection suites named across this documentation set (sannysoft, CreepJS, BotD, FingerprintJS, BrowserLeaks), each read from its own source rather than its rendered verdict.

See also: how to test whether your browser is detected, the checklist for being detected on one site, and what a browser trust score actually measures.

---

Written while maintaining invisible_playwright, a Firefox patched at the C++ level driven by stock Playwright. It is built to look like a real browser, which is why it passes most checks and why it will never claim to make you invisible.