Anti-detect browser leaderboard

Nothing on the fingerprint layer: just the two deductions every row here carries regardless of the browser under test, which is the provisional ceiling for a profile that was not put through the behavioural challenge. That is a genuine contrast with the community report below, which is about a different check on a different site.

One critical finding, and the automation flag itself is not it: navigator.webdriver reads false, because this drives the real installed Chrome rather than a bundled build. What is left is a window sized past its own claimed screen, which is a launcher default rather than anything specific to being automated, plus the unanswered challenge every unverified row here carries.

No critical findings, but two independent tells at the same weight: the generic font families collapse the way a spoofed font list does, and the default speech-synthesis voice is zh-CN while the profile locale claims en-US - two different subsystems leaking the same fact about the machine underneath from two different angles.

No critical findings, and the UA-vs-host mismatches that sank GoLogin do not show up here. What is left is a real but second-order fingerprint tell - two of the three generic font families measuring identically, which a normally provisioned machine does not do - plus the same baseline penalties every unscanned-for-a-hand browser on this board carries.

WebRTC came back with zero ICE candidates, the same cut-off-rather-than-private state GoLogin showed. On top of that the GPU is a real but ancient AMD Radeon HD 7900 reporting a texture size ceiling no current Chrome ships with - internally consistent, but a rare enough pairing to be its own signal.

Marketed as premium, and the two findings here are both the kind that come from emulating a device rather than being one: a window too large for its claimed screen, and a touch point count with none of the pointer-media signals a real touchscreen carries alongside it.

One critical finding, and it is the interesting kind: a cross-interface identity mismatch rather than a stale value. WebGL and WebGPU are two different ways of asking the same GPU who it is, and here they give two different answers - the profile spoofed one without touching the other.

A cross-OS persona that the machine underneath contradicts three separate ways. The profile claims macOS, but the font set is Windows, the GPU story disagrees with itself between WebGL and WebGPU, and WebRTC is switched off hard enough to be its own signal. Each of the three is a different subsystem failing to follow the same lie, which is exactly the shape a persona is supposed to prevent.

Dropping headless removes most of what sinks the headless run below, but the automation flag itself is not one of them: Playwright launches with navigator.webdriver=true and does not clear it, headed or not. On top of that its screen emulation pushes the window past the display it claims to sit on - a Playwright default, not a host artefact.

The independent CreepJS comparison rated this the strongest of seven anti-detect browsers. Scanned directly, it is the worst score on this board: four critical findings that all point the same direction. The UA claims macOS while the font metrics, the installed font set and the fallback width all say Windows; the reported screen is smaller than the window it supposedly contains; and WebRTC produced zero ICE candidates instead of the mDNS host candidate Chrome always emits. None of these are subtle - they are the kind of cross-check a single spoofed UA string cannot survive.

The floor of the board, and the reason the board has a floor. Eleven flags, four critical, on a stack that was not trying to hide: the automation flag is set, the User-Agent still says HeadlessChrome, WebGL runs on a software rasteriser and the window reports no browser UI at all. Zero is not clamping for its own sake - the deductions add up past 160.