Fingerprint Hunting: How Bot Services Collect and Sell Your Data

You visit a captcha solver, disposable email site, or fingerprint checker — and you give this site your real browser fingerprint. By 2026, this is no longer a side effect but a business model: real fingerprints are collected, sorted by anti-bot protection vendors, and sold for about 5 dollars per thousand.

What Happened

The Castle research group published a paper titled “Fingerprint harvesting in the bot ecosystem” (April 22, 2026, last revision — September 7, 2026), which examined 811 websites servicing the bot and fraud ecosystem. A total of 678 domains were accessible and analyzable; 133 were offline or closed.

The methodology is simple and verifiable: sites were opened via Puppeteer with instrumented JavaScript APIs (navigator, screen, Canvas, WebGL), and a script was considered “fingerprint-related” if it accessed at least four navigator attributes and two screen properties. The next step involved analyzing stack traces and manual code reverse engineering.

The result: 12.5% of the analyzed sites (85 domains) loaded fingerprint collection scripts. This is not analytics or local anti-fraud — the structure of the payload indicates that data is collected for subsequent reproduction.

Who Collects: Not “Bad Sites,” But Your Work Tools

The categories where collectors were found are exactly the set of services used by someone engaged in automation:

• proxy providers;
• disposable email services;
• captcha solvers;
• retail and sneaker bots;
• activity boosting services;
• temporary number providers.

The cases analyzed by Castle provide insight into technical maturity:

• cybertemp[.]xyz — a disposable email service with dual-layer collection: the fingerprint is encrypted and sent within the HTTP header X-Client-Meta. The service operator, as observed by researchers, directly stated in Telegram that they collect fingerprints for future use.
• Ez-CAPTCHA — collection occurs in three environments simultaneously: the main page context, an embedded iframe, and a Web Worker. This is not developer paranoia but a replication of the method used by modern anti-bots to verify signal consistency across contexts.
• Cybersole — a sneaker bot where logic for hashing Canvas/WebGL/fonts was found with a characteristic seed string matching the reverse of the client script from the payment provider Adyen.
• StellarAIO — retail automation for over 70 platforms. Most notably: the payload is distributed across different endpoints — /px/c, /incapsula/c, /td/c. The code contained PerimeterX markers like PX11982, PX12423, PX11474.

Thus, fingerprints are not just piled up — they are segmented by protection vendors: PerimeterX (HUMAN Security), Imperva Incapsula, Akamai, Adyen, hCaptcha, Arkose Labs. The collected set of signals is tailored to be reproducible specifically against certain protections.

Economics: 5 Dollars for a Thousand Fingerprints

Researchers noted demand in closed chats: operators discussed their willingness to pay about 5 dollars for 1000 real fingerprints, suggested “monetizing traffic through fingerprint collection,” and offered access to a “fingerprint dashboard” where one could connect their own collector. A separate piece of advice mentioned there was to drive paid advertising to the collector to increase volume.

Why pay for something that can be generated? Because a generated fingerprint is a statistical outlier. A real device provides a set of values that cannot be neatly fabricated: consistent Canvas and WebGL, a real set of fonts, plausible timings of native functions, permission states, capabilities of the WebGPU adapter, and the presence of sensors. It is this consistency that commands a price.

What Exactly They Collect From You

The set of signals in the analyzed scripts repeats from service to service — it is essentially a checklist of what holds value in the market:

• device attributes — screen resolution, hardware concurrency, deviceMemory;
• rendering results — Canvas, WebGL, enumeration of installed fonts;
• signs of automation — navigator.webdriver, traces of extensions;
• timings of native primitives and integrity checks of native functions (whether they have been tampered with);
• sensors — accelerometer, gyroscope, light sensor;
• permission states and capabilities of the WebGPU adapter.

Pay attention to the penultimate two points: timings and integrity checks are needed not to describe the device but to distinguish a live machine from a fake. They are collected so that one can convincingly fake them later.

Some services send all this in encrypted form, while the code retains partially unobfuscated variable names — a sign that the logic was not written from scratch but inserted as a ready-made piece taken from commercial anti-bot scripts.

This Is Not a New Idea — Only the Industry Is New

The mechanics of reproducing someone else's fingerprint have been well established for a long time. The toolkit Bablosoft includes PerfectCanvas (rendering Canvas on real hardware and inserting the result into an automated browser) and FingerprintSwitcher (substituting a set of attributes with a saved profile). As of February 2026, Castle counted over 1500 indexed scans referencing Bablosoft's fingerprint infrastructure.

How this looks in an attacking execution was demonstrated by Group-IB in their study “Fingerprint Heists”: the group ScreamedJungle from May to October 2024 deployed the script clientsafe.js on more than 115 hacked e-commerce sites. The script pulled encoded instructions from the server (including a request for PerfectCanvas), executed them via eval in the browser of an ordinary shopper, and sent the result to a private database through the CustomServers mechanism. Only nine Italian stores from this sample accounted for about 130,000 unique visitors per month — an estimate of 200,000+ collected fingerprints monthly.

The endpoint of this chain is well known from Genesis Market: there, not passwords were sold, but “bots” — snapshots of the victim's browser environment along with cookies, which were loaded into a special browser build and allowed access to the account as if it were the owner themselves. By March 21, 2023, the catalog contained over 450,000 such snapshots; the platform was seized on April 4, 2023, as part of Operation Cookie Monster, resulting in the arrest of 120 individuals.

Why This Hits Particularly Hard for Those Working with Proxies

It is important to distinguish two things that often get conflated in people's minds.

IP — rented and changeable. Fingerprint — not. Address rotation solves a network-level issue: avoiding limits, accessing the necessary geo, not exposing one address across a hundred sessions. But if your real device has already been entered into someone else's fingerprint database and someone reproduces it in front of Akamai or PerimeterX, changing the IP does nothing. You cannot “revoke” the consistency of your Canvas and font list.

Then the unpleasantness begins. Your real fingerprint, reproduced by someone else's bot on someone else's address, starts to gain a reputation to which you have no connection. Next, two simultaneous sessions with the same fingerprint from opposite ends of the world send a signal not of “bot,” but of “device hijacking,” which anti-fraud systems respond to more harshly.

The flip side of the coin is the market trend of 2026: anti-detect browsers are increasingly selling not “generated” but “real” fingerprints — templates taken from live devices. This in itself is not a crime: profiles can be collected legally, with consent, on one's own hardware. However, the origin of the database is rarely disclosed, and the question “where did your real fingerprints come from” is now precisely the question to ask the provider before payment. If you are working with someone else's fingerprint, you are also working with its history: reputation, past bans, and possibly a living owner who is currently using the same profile.

What to Do Practically

1. Separate your hardware. Open captcha solver sites, disposable email services, bot stores, forums, and checkers not from the machine or browser profile where your work and personal accounts are stored. A disposable virtual machine or a separate profile is minimal hygiene.
2. Remember that the checker sees everything. Fingerprint checking is a voluntary disclosure. If you are testing a work configuration, test it specifically from the environment you will be working in, not from your personal Chrome “just to look.” We have a review of CreepJS, Pixelscan, Iphey, and BrowserScan regarding which checkers are truthful.
3. Cut third-party scripts during “reconnaissance.” uBlock Origin or a browser with anti-fingerprint protection significantly reduces the quality of what can be collected from you. A fingerprint collected through a blocker is nearly useless for resale.
4. Do not install their software. Extensions, “branded browser builds,” and desktop clients from such services are code with full access to your sessions. This is how the Genesis scheme operated: buyers were given a browser with a plugin that substituted someone else's attributes.
5. Keep captcha solvers at a distance. Ez-CAPTCHA in the study collected signals in three contexts simultaneously. If an external solver is needed for a task — call its API from an isolated node, not from your working profile. We discussed why relying on solvers in 2026 is fundamentally a losing strategy in our article about captcha and why solvers are failing.
6. If the site is yours — check the foreign code. A script that reads navigator, screen, Canvas, and WebGL and sends data to an external domain has no place on your store's front end. Inventorying third-party tags and enforcing a strict CSP closes exactly the vector used by ScreamedJungle.
7. Separate identities entirely, not just by IP. For each work context — its own profile, its own session, and its own address. For scenarios requiring a stable “domestic” network, use residential proxies, and for mobile platforms — mobile ones; but remember that this is half the task: the other half is covered by the environment, not the address.

Conclusion

The main shift of 2026 is not that fingerprints are being stolen — that has been happening for a long time. It is that collection has become structural: 12.5% of the bot ecosystem sites collect signals, sort them by specific protection vendors, and sell them at a price. The ecosystem that teaches how to bypass detection has itself become the largest source of fresh material for detection — and for hijacking.

The takeaway for practice is simple: consider any “automation” service a potential collector and access it from a disposable environment. You can always change the IP. Your real device fingerprint — no longer.