“A rule catches the known-bad value. Signal depth across device characteristics, identity, integrity, network, location, remote access and virtualized environments catches the infrastructure behind complex, automated fraud.”
Daniel Gerhat, Senior Fraud Consultant at SEON
quick summary
What Is Proxy Detection?
Proxy detection tells you whether a connection is hiding its route or its location, so you can decide whether the environment behind a session can be trusted. It reads the network path a request took, not only the address it arrived from.
A proxy on its own is not proof of fraud. Staff on a corporate network, travelers on a consumer VPN, people on public Wi-Fi with security software switched on and privacy-minded customers all arrive through a masked route every day.
The flag still matters, because the same cover lets a fraudster pass a geo rule, open accounts in bulk, test stolen cards or sign in to an account that is not theirs. A fraud team needs to know whether the user is connecting from where they claim and whether the route, read alongside everything else in the session, raises the risk.
What Each Proxy Detection Layer Catches and Misses
Proxy detection runs on two layers, and each one lets a team do something the other cannot.
An IP check gives fast context on the first request. It identifies many known VPNs, datacenter ranges, web and public proxies and Tor exits, and it can flag some listed residential proxy addresses. It cannot keep up with residential or mobile proxies that use ordinary ISP addresses, because reputation only knows an address once someone has reported it.
A live connection check on the web closes that gap. Instead of looking up the address, it reads the TCP/IP and TLS fingerprints of the connection itself, alongside related browser signals, which is how residential and mobile proxies surface without a reputation lookup. Accuracy depends on the connection. Proxied sessions run slower, and a check cut short by a tight timeout can come back incomplete, so give it enough time to finish.
The result also needs a rule before it does anything. Teams write a custom rule on the VPN and proxy flags, or set a score on the default Proxy/VPN rule, which ships at 0 so its hits can be watched before they count.
You need both layers. The IP path continues to work when device collection is blocked, but it will not close the residential and mobile gap on its own. Connection-fingerprint detection, and the vpn1 to vpn4 and proxy1 to proxy4 reason flags that come with it, run on the web through the JavaScript Agent.
Types of Proxies and How Hard Each Is to Detect
Not all proxies carry the same risk, and they are not equally easy to spot. Datacenter and public proxy exits announce themselves at the address level. Residential and mobile proxies look like a customer at home or on a phone plan, which is why they are the hardest to catch and the most common cover for multi-accounting.
The table compares what fraudsters use each type for, what SEON shows you and the legitimate traffic you should expect on the same route. VPNs get one line here because they deserve their own treatment, covered in our guide to VPN detection tests.
| Type | What fraudsters use it for | What SEON’s proxy detection shows you | Legitimate traffic to allow for |
|---|---|---|---|
| Datacenter or hosting proxies | Bots and scripted sign-ups from cloud and hosting ranges | When a consumer session comes from a hosting or cloud network | Developers, QA teams and many VPN users |
| Web proxies | Opening accounts or filing claims from a chosen country through a dedicated browser proxy | That the session is on a web proxy route, scored as its own class | Privacy and access tools, including some company browser setups |
| Public or open proxies | Credential stuffing and new accounts on cheap, shared exits | Public proxy exits, especially when the same route appears across new accounts | Rare for ordinary customers. Still check the rest of the session |
| Residential proxies | Multi-accounting and promo abuse that looks like many households | On the web, the live TCP/IP and TLS fingerprint of a residential proxy, for your rules to act on | Real customers on a home internet provider |
| Mobile or carrier proxies | Device farms hidden behind shared carrier IPs | On the web, the same live connection fingerprint, plus when an address belongs to a mobile network | Ordinary customers on mobile data |
| Commercial VPNs | Location masking to reach geo-restricted offers and bonuses | VPN use, and the provider when it is known so that you can compare the exit with other location signals | Travel and privacy on public Wi-Fi |
| Enterprise private networks | Stolen employee accounts signing in through the company network | That the session is on an enterprise private network, so staff can keep moving | Remote work on a company-controlled route |
| Consumer privacy networks | Appearing as a local consumer to reach a geo-restricted offer or bonus | That a consumer privacy network is in use, so a privacy tool is not treated like a public proxy | Everyday privacy-minded traffic |
| Tor | Location masking for geo-restricted offers, or accounts that should not tie back to a home IP | When the address is a known Tor node | Rare in payments and account opening. Some privacy and safety use |
| Location mismatch | A nearby VPN or proxy exit used to pass a country rule while the device still sits elsewhere | Disagreement across IP, device IP, WebRTC, DNS IP, time zone, language and consented GPS | Travelers and staff on a company network |
How to Verify a Datacenter Proxy
In a consumer sign-up or payment, exiting a hosting range is unusual enough to create friction early. Check whether the IP belongs to a cloud or hosting provider rather than a household or mobile ISP. A datacenter type also appears for many VPN users and for developers and QA teams working from the cloud, so it is not automatically a proxy.
Residential Proxies
A residential proxy is built to look like an ordinary home user, which is why a clean IP is not the same as a clean route. A live, web-based connection check can detect residential proxies without relying on IP reputation, and the result needs a rule before it scores or blocks anyone.
Mobile or Carrier Proxies
Mobile proxies exit through ordinary cellular addresses, so reputation is a weak control, and shared carrier IPs also flag good customers. Our Hidden Risk Files case, The Fraud Ring That Looked Like Two Hundred Strangers, shows the pattern at scale. Nearly 200 accounts shared no device ID. What tied them together was the mobile proxy infrastructure their sessions were routed through, and the same live connection fingerprint that exposes residential exits is what reads it.
How Proxy Detection Works Across IP, Connection and Device Checks
Proxy detection works by asking a session whether the IP itself is risky, whether the location adds up and whether the device is hiding something. Each technique answers part of one of those questions and misses the rest, so each one below comes with what it lets you decide and the honest traffic it also catches.
Checks on the IP
IP checks give you early context on known infrastructure. A datacenter range lookup tells you whether the address belongs to a hosting provider rather than a consumer ISP. Blacklist checks tell you whether the address has been reported for abuse. An open port scan can show services running on the address that a home connection would not expose. An IP fraud score rolls these into one number for a first decision.
One distinction matters more than any single check. IP type describes who owns the address, whether residential, mobile or datacenter. It does not say whether the route is a proxy. For that, use the proxy and VPN fields. A mobile IP type is context, not a proxy flag, because shared carrier addresses carry thousands of ordinary customers at once.
Checks on the Connection
Connection checks are how you move past address reputation and read the live route. On the web, TCP/IP and TLS fingerprints describe the network stack a request came through, and a proxied connection leaves a different fingerprint than a direct one. HTTP traffic analysis adds how the request itself behaves. Comparing the DNS resolver’s IP and its country or ISP with the connection IP shows whether name resolution is happening somewhere other than where the address claims to be.
Latency comparison, which measures a round trip against what the claimed location would allow, is a general method rather than a product feature. It also flags slow networks, so it belongs low in the weighting.
Checks on the Device
Device checks ask whether the browser agrees with the network. WebRTC can reveal IP addresses other than the one the site sees, so comparing them with the connection IP directly tests the route. The browser’s time zone set against the IP’s time zone, and its language and region settings set against the IP’s country, indicate whether the person is where the address says they are. Tor is identified when the IP is a known Tor node, not by scanning the browser.
Travelers and remote staff produce some of these disagreements for ordinary reasons, which is why no single check is enough on its own.
Where Proxy Signals Catch Fraud
Proxy signals pay off where a masked route is the whole point of the attack. Five patterns account for most of what fraud teams see.
Banned Users Coming Back and Multi-Accounting
A blocked user returns via a new VPN with a new identity, or by rotating residential IPs, making each new claim look like a different household. Because residential IPs look like ordinary home connections, this kind of fraud often passes standard IP checks. The giveaway is a route that repeats where identities do not, which is the core of multi-accounting detection.
Geo-Restriction and Licensing Evasion
Users connect through a VPN exit in an allowed state or country to become eligible. Sweepstakes, skill gaming and prop trading operators all restrict jurisdictions, and all of them see it. The VPN flag, read with a location mismatch between the exit and the device, is what catches it.
Account Takeover and Session Changes
A clean login, then the network changes. A VPN switches on, or the IP jumps to a different country, minutes after onboarding. Check for this during the session, not only at the start. Session Monitoring can flag an IP change or a VPN during the journey on web, Android and iOS. A switch from Wi-Fi to mobile data is a documented false positive, so pair the change with the action being attempted, such as a payout or a new payment method, before treating it as account takeover.
Payment and Promo Fraud
A fraudster in Portugal holds stolen Australian card details. The first order is flagged because the IP and the card do not agree. The fraudster routes through an Australian IP to look local and tries again. A proxy flag on the retry, plus a mismatch between the card country and the device’s country, time zone or language, is what stops the second attempt.
Mule Accounts and Device Farms
Two separate stories hide under one label, and they need different rules.
Account handover. A known device returns, sometimes via a residential or mobile proxy, and attempts a risky action, such as a new payout. The device looks familiar. The route and the action may not. A stable device ID does not prove the same person is in control, which is why handed-over mule accounts pass device checks that look only for change.
Scripted or farm sign-ups. Each account gets a fresh device ID, so nothing looks shared on the surface. The link is in the infrastructure. Mobile proxy routes repeat across accounts, and hardware characteristics that should vary come back identical. Our Two Hundred Strangers case file documents nearly 200 accounts with no shared device ID, tied together by the mobile proxy infrastructure behind them.
Location Mismatch: When the IP Says One Place and the Device Another
Location spoofing detection comes down to comparing signals that should agree. A fraud team can set IP location against device IP, WebRTC IP, DNS resolver IP, time zone, language and, with the user’s permission, GPS. When the IP says one country and the device says another, something in the route is being managed.
The comparison does the most work where location is a rule. We have seen cases like:
- Gaming and sweepstakes operators restrict states.
- Lenders and banks need to know that a customer sits within the jurisdiction in which they are licensed.
- A rebate processing platform rejects claims filed from outside the country of purchase, and it has caught claims entered from one continent for products bought on another on exactly this comparison.
SEON’s solutions engineers describe the same approach. Compare location signals against each other, identify the real country behind a proxy, and fall back to IP-inferred location when GPS is not allowed. One of them runs it as a rule that adds risk whenever the device region and the IP region disagree.
Location Permission
The comparison does the most work where location is a rule. Gaming and sweepstakes operators restrict states. Lenders and banks need to know that a customer sits within the jurisdiction in which they are licensed. Rebate and claims platforms often reject claims filed from outside the country of purchase, and this comparison is what exposes a claim entered from one continent for a product bought on another.
How to Set Up Proxy Rules Without Drowning in Reviews
Proxy rules go wrong in one of two ways. Either they block good customers, who then call support, or they fill a review queue faster than the team can clear it. The decisions below, taken in order, keep both in check.
1. Score the IP and the Device Separately
Send the client IP from your server as well as the device session. The IP API and device fingerprinting are separate detection paths, so the IP still scores when a privacy browser or security software blocks device collection. One SEON customer found exactly that on good users’ machines. The IP path is what kept those sessions scoreable.
2. Check at Every Sensitive Event, Not Only Sign-Up
Login, payout, payment detail changes and, for banks, every transaction. A proxy that appears at payout after a clean sign-up is a stronger signal than one at sign-up. NIST SP 800-63B lists IP address characteristics and geolocation among the session signals for continuous authentication. It is written for U.S. federal agencies, but it is a useful reference point for how often to look.
3. Use Flag Types and Provider Names, Not One Proxy: Yes or No
Specialist VPNs marketed to fraudsters and public proxies carry more risk than mainstream consumer VPNs or enterprise networks. A gaming operator and a payments company both told us the provider matters more than the flag. Score the flag type and, when it is available, the provider name. SEON’s vpn1 to vpn4 and proxy1 to proxy4 flags name the reason the signal fired, for example HTTP traffic analysis or WebRTC IP characteristics, so that a rule can weight a public proxy and a corporate route differently.
4. Add History: Has This User Ever Used a Residential Proxy?
Velocity rules on proxy use over time, and on the same IP across accounts, turn a one-off flag into a pattern. A customer who has never been seen on a residential proxy and appears on one at payout is a different case from one who always arrives that way.
5. Run New Rules at Zero Score First
Watch hits and false positives before a rule can decline. SEON’s default device-level Proxy/VPN rule, which covers residential proxies, ships with a score of 0 so teams can monitor it from day one. A gaming operator and a prop trading firm both described this as standard practice, and both raised the score only after review mode showed them which customers they would have lost.
6. Give Network Checks Time to Finish
Proxies slow connections, so short timeouts can leave the network fingerprint incomplete. Plan for your page-speed budget and treat “not enough data” as a different outcome from “no proxy.” SEON exposes a max_proxy_delay setting in the JavaScript SDK to extend the window.
7. Check Consent and Data Residency
Precise location needs explicit consent, and some regulators restrict where IP and device data can be processed. Lenders and identity verification providers raised data residency early on calls with us, before any discussion of rules. NIST SP 800-63B also requires IP and geolocation session signals to be covered in a privacy risk assessment. Get your legal team’s answer before you turn on the GPS prompt.
Proxy Detection vs. Bot Protection vs. Device Intelligence
The three overlap less than vendors imply. A rebate processing platform runs bot protection at the network edge and still saw fraud arrive through residential IPs because edge protection lacks device profiling and residential exits pass IP reputation checks. Each layer answers a different question.
| Approach | What it answers | What it catches | Where it is blind |
|---|---|---|---|
| IP-based proxy detection | Is this address known infrastructure? | Datacenter, VPN, Tor, web and public proxy exits, and listed residential addresses | Residential and mobile proxies on clean addresses. Cannot tell a company network from an attacker on it |
| Edge bot protection (WAF or CDN) | Is this traffic automated? | Scripted volume, known bot signatures and rate abuse | Human-driven fraud at normal speed. Residential proxy traffic that looks like customers. No device history |
| Device intelligence | Have we seen this environment, and does it agree with itself? | Residential and mobile proxies on web via connection fingerprint, location mismatch, emulators and repeat devices behind new identities | Needs the script to load. Blocked collection falls back to the IP path |
How SEON Detects Proxies
SEON runs both layers in one Fraud API call: IP reputation for known infrastructure, plus live connection checks on the web that catch residential and mobile proxies. Every flag explains why it fired and names the provider when known, so your rules can treat a public proxy and a corporate VPN differently. New proxy rules start at zero score, so you can review them before they block anyone.
FAQs From Fraud Teams
How accurate is proxy detection?
Accuracy depends on which layer is doing the work. IP reputation is reliable for datacenter, VPN, Tor and public proxy exits that have been cataloged and weak for residential and mobile proxies on ordinary ISP addresses. A live connection check on the web reads the route itself, so it catches those, provided the check has time to complete. No single test is accurate enough to decline on alone.
How accurate is residential proxy detection?
Residential proxy detection is accurate when it reads the connection rather than the address. The TCP/IP and TLS fingerprints of a proxied session differ from those of a direct one, which is what SEON’s web-based check uses. Accuracy drops when a slow proxied connection is cut off by a short timeout, which is why the check needs enough time and why “no data” should be treated differently from “no proxy.”
Can you detect a proxy from the IP address alone?
Partly. An IP lookup can tell you the address belongs to a data center, a known VPN provider, a Tor node or a listed public proxy. It cannot tell you whether a residential or mobile address is being used as a proxy because the address looks like any other customer’s. For those, you need a web connection check and a comparison of the IP’s location with what the device reports.
How accurate are proxy detection tools?
A tool that relies only on IP lists is accurate for known infrastructure and blind to residential and mobile proxies. A tool that reads the live connection on the web, compares location signals against each other and gives you the reason a flag fired is accurate enough to build rules on. Judge a tool by whether it explains its flags and whether it lets you run a new rule at zero score before it blocks anyone.
Can you detect residential proxies?
Yes, on the web, by reading the connection rather than the address. SEON identifies residential proxies from the session’s TCP/IP and TLS fingerprint, and the finding feeds the default Proxy/VPN rule, which ships at score 0 so you can watch it before you act on it. Pair the flag with device and account history, because a real customer on a home ISP looks identical at the address level.
What are the types of proxies?
Proxies range from datacenter, web and public proxies to residential and mobile proxies, VPNs, enterprise and consumer privacy networks, and Tor. Residential and mobile proxies are the hardest to detect because they use ordinary customer addresses. A proxy flag is only one signal: read against device, email, phone and account history, it lets a good customer on a VPN keep moving while stopping a fraud ring of nearly 200 accounts hiding behind mobile proxies.
