Maskbreak vs Castle (castle.io)

On this page: what Castle is, how the Castle token works, the feature table, Castle pricing, one fake signup scored by both and which to choose. Surveying the wider field instead? Read our Castle alternatives or Fingerprint alternatives guide.

What is Castle?

Castle (legally Castle Intelligence, Inc.) is a bot and account-abuse prevention platform based in San Francisco. Johan Brissmyr and Sebastian Wallin founded it in 2015, it went through Y Combinator in Winter 2016, and Index Ventures led its $9.2M Series A in 2019. We found no acquisition or later round, and its about page calls the company profitable. The homepage promise is “Stop bots and account abuse at scale”, built on device, email, IP and behavioural intelligence, and the use cases reach well past the login form: fake accounts, account takeover, multi-accounting, content abuse, SMS pumping, account sharing, transaction abuse and API abuse. Atlassian, Canva, Rockstar Games, Rakuten and YNAB are among the names on its customer page.

Castle moved fast in 2026, which is why older comparisons (an earlier version of this page included) undersell it. Its older proxy_ip signal already covered datacenter proxies, VPNs and other known proxy ranges. On 28 May it added residential_proxy_access, on 18 June vpn_access, with the operator in ip.tunnels when known, and on 6 August a standalone IP Intelligence API that reports recently seen proxy, VPN, relay and Tor tunnels behind any address, with no browser integration at all. Its mobile SDKs gained phone-call detection in May, aimed at scams where a fraudster talks the victim through a login or a transaction while on the phone.

The Castle token and device fingerprint, and what Maskbreak sends instead

Castle. The browser SDK is Castle.js. Castle’s docs describe version 3.0, whose public surface is two calls, configure() and createRequestToken(); the older client-side page, form and custom event tracking is gone, so tutorials written for earlier versions will not match.

The request token carries the device and behavioural signals Castle.js collected in the browser. It expires after 120 seconds, has to be created fresh for every backend request, and goes in the request body, not a header. Your server sends it to the Filter API while the visitor is anonymous (a registration, login or password-reset attempt) or to the Risk API once they are signed in. The answer holds three scores from 0 to 1 (Bot, Account Takeover and Account Abuse), the signals that fired, and a policy action: allow, challenge or deny.

One trap when you install it: on 24 September 2026 the latest @castleio/castle-js on the public npm registry was 2.8.5, not a 3.x release. Check which version your bundle actually ships before you follow the 3.0 docs; Castle’s dashboard has had an SDKs page since 1 September that shows which versions your traffic runs.

Maskbreak. The browser snippet collects two values, a network token and a device event ID from Maskbreak’s device check, and adds them to your form or hands them to your code. Your server sends both to POST /v1/evaluate with the API key. Back comes a decision (allow, review or block), a risk_score from 0 to 100, machine-readable reasons, a network block, a device block when the browser check completed, and evaluated_in_ms, the server’s own processing time for that call. The quickstart shows the whole call.

The actions line up: Castle’s allow, challenge and deny match Maskbreak’s allow, review and block, and both come back on the first call, before you write a single rule. What drives the default differs. Castle ships default policies that act on its scores (on the dashboard’s 0–100 scale, 90 to 100 denies and 60 to 100 challenges), and fitting them to your traffic means editing Policies and lists. Maskbreak’s default acts on named signals: a proxy, Tor exit, automation, emulator or fake browser blocks, a VPN on its own goes to review, and a cloud-server address alone stays allow. You change it one signal at a time on the console’s Rules tab.

Can the Castle token be bypassed or replayed? Castle plans for both. Beyond the 120-second expiry, its signals reference lists three signals for tampered evidence: missing_device_data when the device data arrives empty (possibly an attempt to skip the browser analysis, possibly a misconfigured integration), invalid_device_data when it cannot be interpreted, “most likely due to a replay attack or request tampering”, and replayed_device_data when the same device data is used again. Sandbox environments accept mock tokens, so real users belong in a production environment. Maskbreak’s integration guide sets the same rule for enforce mode: a missing token, a missing device result or degraded evidence is not a clean check, so the action is held, not waved through.

Maskbreak vs Castle, feature by feature

Castle.io pricing (September 2026)

Castle publishes its prices on castle.io/pricing. Free and Pro are a monthly API budget: a successful Risk or Filter request costs $0.005, each valid IP returned by a lookup costs $0.001 (so a full 1,000-address batch costs up to $1), and the plan fee buys that much usage. Enterprise is custom, priced by monthly tracked users or by volume. Unused budget expires at the end of each billing cycle, and you can cancel at any time.

In practice, protecting real users on Castle starts at Pro, $200 a month. Castle’s sandbox docs explain why the Free environment is not for production: a sandbox accepts mock tokens that let the client set the returned action, and Castle says it “must never protect real users”.

What a month of risk checks costs at list prices. The Maskbreak column is the smallest plan that covers the volume (prices exclude VAT). The Castle column is our arithmetic from the published rates on Pro, not a quote; Enterprise terms are negotiated, so large volumes may price differently.

Two costs the table leaves out. First, history: Free and Pro keep 3 and 7 days of data, so a fraud team that investigates last month’s incidents inside Castle is really shopping for Enterprise. Second, IP lookups are not the same product on both sides. Castle’s IP Intelligence API reports recently seen tunnels behind any address you send, with the operator when known; Maskbreak’s bare-IP lookup comes with every plan (100,000 a month on Free) but covers only Tor exits and published cloud ranges, and its VPN and proxy verdicts need a live visit.

One fake signup, scored by both

A common account-farm setup: one fake (antidetect) browser profile per account, each profile routed through a residential proxy so the address looks like a home broadband line. The check runs on the signup form, before the account exists.

Castle sees it in the Filter API call on a $registration attempt. If Castle has seen the address on a residential proxy provider’s network in the last seven days, residential_proxy_access fires and ip.tunnels gives the proxy type and, when known, the operator. Device parameters the profile invented may raise spoofed_device; the signals reference has no signal named for antidetect browsers. Castle’s default policies then act on the resulting scores, and a Policy you add can deny, challenge, or put the device on a list for someone to review.

Maskbreak returns one decision with the reasons attached. When the address is a known proxy and the browser check names a fake browser, the default policy blocks. Example response, selected fields:

{
  "decision": "block",
  "risk_score": 100,
  "reasons": ["proxy_detected", "antidetect_browser"],
  "network": {
    "vpn": false,
    "proxy": true,
    "tor": false,
    "service": "<operator name when known>"
  },
  "device": {
    "antidetect": true,
    "automation": false,
    "tampering_score": 0.9
  }
}

Swap the proxy for a consumer VPN on a clean browser and the answer drops to review, not block, because plenty of real people sign up through a VPN. Example response, selected fields:

{
  "decision": "review",
  "risk_score": 35,
  "reasons": ["vpn_detected"],
  "network": {
    "vpn": true,
    "proxy": false,
    "datacenter": false,
    "tor": false,
    "residential": true,
    "service": "<operator name when known>"
  }
}

Three caveats apply to both vendors. First, no address check can promise to recognise every residential exit. Castle’s residential_proxy_access needs the address to have turned up on a provider’s network in the last seven days (its signals reference says proxy_ip, which also weighs timezone and timing anomalies, does not cover residential proxies), and Maskbreak publishes no detection rate at all.

Second, a rotating pool changes the address on every attempt, but not the machine. That is where the device side helps. Send your accountId and Maskbreak’s device block counts how many of your accounts the device has been linked to (linked_accounts); past one, the reason multi_account_device appears. Castle’s equivalent is multiple_accounts_per_device. The multi-accounting playbook covers what to do with it.

Third, missing browser evidence is not a clean result: if the token never arrived, hold the signup rather than wave it through. The fake signup playbook covers what to do with each answer; for background, read our residential proxy detection guide and what an antidetect browser is.

Where Castle is the better choice, and what it costs you

What neither product does for you

Castle returns challenge; it does not run the CAPTCHA or send the one-time code. Its docs advise against CAPTCHAs in favour of contextual steps, such as restricting sensitive actions or trusting a device for a while after email or SMS verification, and a separate guide shows how to trigger your own CAPTCHA on a challenge. Maskbreak’s review is the same kind of instruction: hold the signup, ask for a code, or queue it for a person on your side.

Both let you start quietly. Castle Policies can run as Log Only; Maskbreak’s recommended first integration is watch mode, which logs every decision and always lets the visitor through; you switch to enforce once the Events log looks right (the staged rollout guide walks through it). Maskbreak does not replace authentication, MFA, KYC or payment-fraud tooling, and it has no manual review team, no chargeback guarantee, no cross-customer identity network and no email or phone reputation beyond the disposable-domain signal. For logins, pair either one with the controls in account takeover prevention that holds up and the account takeover playbook.

Castle or Maskbreak: which one to choose