Entropy and why rare configurations are easy to track
What entropy actually measures
Entropy, in the fingerprinting sense, is a measure of how much a single piece of information narrows down who you are. If a data point is shared by half the internet, it barely narrows anything down. If a data point is shared by one person in ten million, it narrows things down almost completely. Trackers don’t need a name attached to your browser to follow you around the web. They just need enough low-probability details stacked together that the combination points at one machine.
This is the same math behind the old “how many people share your birthday and shoe size” party trick. Birthday alone tells you almost nothing, since it’s shared by roughly 1 in 365 people. Add shoe size, add city, add the car you drive, and pretty quickly you’re the only match. Browser fingerprinting works the same way, except instead of birthdays it’s collecting screen resolution, installed fonts, GPU rendering quirks, timezone, audio stack behavior, and a dozen other signals that on their own look harmless.
I run proxy farms and cloud-phone setups for multi-accounting work, and the fastest way I’ve seen operators get flagged isn’t a leaked IP. It’s a browser that’s technically “unique” in a way that stands out from real traffic.
Why rare is worse than common
A lot of people assume that if a fingerprint is unusual, it’s harder to track because there’s less data about it in a training set. It’s the opposite. Detection systems don’t need historical data on a specific rare fingerprint to flag it. They need to know that fingerprint is rare relative to the population of real users hitting that platform.
Think of it as a distribution problem. Most real users on a given site cluster around a small number of common configurations: mainstream OS versions, mainstream GPUs, mainstream screen sizes, fonts that come preinstalled with the OS. When a browser shows up with a font list nobody else has, a canvas hash that’s never been seen before, or a WebGL renderer string that doesn’t match any real GPU on the market, it sits out on the tail of that distribution. Tails are exactly where anomaly detection looks first, because that’s where the signal-to-noise ratio is best. A detector doesn’t need to know who you are to flag “this profile is statistically unlike the other 50,000 sessions we saw today.”
This is why a spoofed fingerprint that’s too clean can be a worse tell than an honest, slightly-inconsistent real one. Real machines have noise. A stock Windows laptop has fonts installed by whatever software the user has on it, a GPU driver that reports specific quirks, a screen resolution tied to a specific panel. Spoofing tools that generate a fingerprint from a template often produce something too tidy: a font list that’s a clean subset of a known OS default, a hardware concurrency value that doesn’t match the reported GPU, a canvas render that’s suspiciously smooth. Consistency between fields matters as much as any individual field.
The Panopticlick lesson
The EFF’s Panopticlick project (now folded into Cover Your Tracks) made this concrete years ago by scoring how many bits of identifying information a browser leaks. The finding that mattered wasn’t that fingerprinting is possible. It’s that most browsers, run with default settings, are unique or near-unique among the sample the project collected. Plugins, fonts, and screen properties alone were often enough to narrow a session down to one in tens of thousands.
The practical takeaway for anti-detect browser users is that “blending in” isn’t about hiding data, it’s about matching a distribution. A profile that reports zero fonts, a blank plugin list, and a generic user agent isn’t invisible. It’s a different kind of rare, one that says “this session has been stripped down,” which is its own fingerprint. Detection teams at ad platforms and marketplaces have seen enough automation traffic to build detectors specifically for over-sanitized browsers, not just over-unique ones.
Where the entropy actually comes from
When I test anti-detect browsers like Multilogin, GoLogin, Kameleo, AdsPower, and Dolphin Anty, the fields that move the needle most in terms of entropy are usually:
Canvas and WebGL rendering. These generate a hash based on how the actual GPU and driver render a hidden image or 3D scene. Different hardware and driver versions produce measurably different output, and the hash is stable per machine, which makes it a strong identifier if it isn’t varied consistently across sessions.
Font enumeration. The list of fonts a browser can see depends on the OS, the language packs installed, and any software that bundled its own fonts (design tools, PDF readers, some games). A font list is one of the higher-entropy signals precisely because it reflects a machine’s history, not just its spec sheet.
Audio context fingerprinting. Similar idea to canvas, but through the Web Audio API. Small differences in how audio is processed and output create a consistent, hashable signature.
Hardware concurrency, device memory, and screen metrics. Individually low entropy, since these cluster around common values. But combined with everything else, they help confirm or contradict the rest of the profile.
Timezone and language versus IP geolocation. This one isn’t really about entropy in the “rare configuration” sense, it’s a consistency check. A profile claiming to be in one country while the timezone, language, and IP all disagree adds a different kind of anomaly signal on top of whatever the raw fingerprint entropy is.
The tools that hold up better in my testing are the ones that treat these fields as a set that has to stay internally consistent, not a checklist of individual values to randomize. Kameleo and Multilogin both put real engineering into keeping GPU-reported values, canvas output, and hardware specs aligned with a device template rather than randomized independently. AdsPower and Dolphin Anty lean more on volume and profile isolation, which helps with account separation but doesn’t by itself solve the “does this look like a real, common device” problem.
None of that makes any of these tools undetectable. Platforms that care about fingerprinting, and the bigger ones do, keep updating their detection models against exactly the profile-generation logic anti-detect vendors ship. What changes between tools is how much they think about entropy as a distribution problem versus a randomization problem.
What this means in practice
If you’re testing anti-detect browsers for real multi-accounting work, the useful question isn’t “is this fingerprint unique.” Almost every browser is unique on some axis. The useful question is “does this fingerprint look like it belongs to the population of real devices that normally show up here.” A profile that mimics a common, mainstream device configuration, with the internal consistency a real device would have, blends into the noise floor better than one that’s been maximally randomized to look different from every other session you’ve run.
That also means testing your own setup matters more than trusting a vendor’s marketing claim. Cover Your Tracks and similar entropy-testing tools will tell you roughly how identifiable a given profile is relative to their sample. It’s not a guarantee about how any specific platform’s detection will react, since every platform builds its own models on its own traffic, but it’s a useful sanity check before you trust a profile with real account work.
No anti-detect browser removes risk entirely, and no vendor’s claim about detection resistance is something to take at face value without testing it yourself. Accounts still get reviewed, flagged, and banned on platforms that invest in fingerprinting defenses, regardless of which tool generated the profile. Understanding entropy is about making informed decisions with realistic expectations, not finding a shortcut around detection.
If you want to see how these tools actually behave side by side, with real entropy and consistency checks instead of vendor claims, that’s what we test on the channel and write up here at Anti-Detect Review.
Get new guides and videos first — join the Telegram channel.