Skip to content
IP Intelligence

How VPN and Proxy Detection Actually Works, Signal by Signal

How VPN detection works in practice: the signals behind modern VPN and proxy detection, from provider lists and ASN analysis to protocol probing.

ByGeoIPHub TeamPublishedReading time10 min read
Glowing tunnel of rings with probe beams touching it from outside, representing how VPN and proxy detection works

Every fraud, abuse, and compliance team eventually asks the same question: is this IP address hiding something? (If you first want the basics, start with what a VPN is and how to detect VPN traffic. This post goes deep on the machinery.) The honest answer is that no single check can tell you. A commercial VPN, a datacenter SOCKS proxy, a residential proxy rented from a botnet, and a Tor exit node all look different on the wire (the breakdown of proxy types covers each family), and each one defeats a different naive detection method.

This guide walks through the signals that production-grade detection systems draw on, including two (latency checks and behavioral modeling) that GeoIPHub does not use. GeoIPHub builds its own verdicts from seven verification method families that cover the same ground. For each layer you will see what it can and cannot see, and how the results get merged into the response fields you actually use: is_vpn, is_proxy, is_residential_proxy, and is_tor.

Slender light beams striking a sphere with some bouncing back, representing protocol handshake probes
A handshake that answers is evidence; a list entry is a claim.

Provider Server Lists

The most reliable evidence is the most direct: VPN providers publish their server infrastructure. Apps need endpoint lists to connect, and many providers expose official APIs or config bundles listing every server hostname and IP.

Detection systems continuously crawl these sources. When an IP appears in NordVPN's or Mullvad's published server list, the match names the provider, but only when that exact IP is on a list we hold. A server missing from the lists can still be caught by the signals below and stay unnamed. This is why a good API can return not just is_vpn: true but the provider name itself, and why it sometimes returns no name for a server it has flagged.

The limitation is coverage. Published lists only catch providers that publish, which means this layer is strongest for the large commercial VPNs that account for most consumer VPN traffic, and weakest for private or self-hosted VPNs.

ASN and Hosting Infrastructure

Every IP belongs to an Autonomous System (AS), the network that announces it to the internet. The ASN tells you who operates the network, and that is a powerful prior:

  • An IP announced by AS396982 (Google Cloud) is almost certainly a server, not a person on a couch.
  • An IP announced by a residential ISP (Comcast, Deutsche Telekom, Jio) is probably an eyeball connection.
  • A long-haul hosting or colocation ASN with a history of VPN endpoints raises the prior for every IP in its ranges.

Datacenter detection alone catches a huge share of VPNs and proxies, because most anonymizing infrastructure rents cloud servers. But it cannot distinguish a VPN endpoint from an ordinary web server. That requires the next layer.

Active Protocol Probing

This is the layer most lookup databases skip, and it is the difference between guessing and verifying. A VPN server has to answer VPN protocol handshakes; that is its job. So a detection system can knock on the door and see who answers:

  • OpenVPN responds to a characteristic TLS or UDP handshake on its listening port.
  • IKEv2/IPsec responds on UDP 500/4500 with identifiable payloads.
  • PPTP and L2TP answer their control-channel requests.
  • SOCKS5 and HTTP proxies complete their respective greeting sequences.

WireGuard is the exception: a server stays silent unless the first message comes from a key it already knows, so no outside probe can confirm it. GeoIPHub's wireguard_open method comes from the provider's published server list marking that server as WireGuard, not from a handshake.

If an IP completes an OpenVPN handshake, it is running OpenVPN. No statistical inference required. GeoIPHub probes across 8 protocols and records exactly which handshake succeeded, so the evidence behind is_vpn is named and auditable. This lookup of a commercial VPN address returned the response below on 5 October 2026:

Terminal
curl -H "x-api-key: YOUR_API_KEY" \ https://api.geoiphub.com/v1/lookup/102.129.235.17
JSON
{ "ip": "102.129.235.17", "asn": { "asn": 834, "org": "IPXO - IPXO LLC", "asn_type": "vpn", "connection_type": "datacenter" }, "detection": { "is_vpn": true, "is_proxy": false, "is_tor": false, "is_anonymous": true, "vpn_provider": "none", "vpn_confidence": 100 }, "scoring": { "fraud_score": 92, "recommended_action": "block", "detection_methods": ["known_vpn_server", "openvpn_open", "ipsec_open"] } }

On that date openvpn_open and ipsec_open in detection_methods recorded that this address answered OpenVPN and IPsec probes. vpn_provider was "none", so no brand was attributed; the verdict rested on protocol evidence rather than a provider label. Two things have changed since. The same address returned 0, allow, with no evidence on record when we checked it on 11 November 2026. And since our 10 November scoring change, a VPN found only by our handshake gets vpn_service_unattributed instead of known_vpn_server, with vpn_confidence 60 rather than 100. openvpn_open and ipsec_open can also come from a provider's server list that advertises those protocols, so read them as "this server offers the protocol", not always as "our probe got an answer".

Active probing has one cost: it takes infrastructure and time, which is why it runs continuously in the background rather than per-request. The probe results are cached and refreshed so the API can answer in milliseconds.

Port and Service Fingerprints

Even when a full handshake is not possible, the shape of an IP's open services is informative. A residential connection typically exposes nothing. A VPN endpoint exposes a small, characteristic set of ports. A proxy farm exposes management panels, SSH, and proxy ports across whole subnets.

Service fingerprinting also catches misconfigured or unbranded anonymizers: a server answering on TCP 1080 with a SOCKS banner is a proxy regardless of whether any list says so.

Latency and Location Consistency

Geography leaves fingerprints in physics. If an IP claims (via registration data) to be in São Paulo but its round-trip times from global vantage points triangulate to Amsterdam, something is relaying traffic. Latency-based checks are especially useful against:

  • Geo-spoofed VPN servers: providers marketing "Brazil" servers that physically sit in Miami.
  • GRE/tunnel relays that forward traffic to a hidden real location.

This layer rarely fires alone, but it is a strong corroborator when combined with hosting signals.

Residential Proxy Behavior

Residential proxies are the hardest class, because the exit IP genuinely belongs to a home ISP. List checks pass. ASN checks pass. Probing finds nothing listening, since the "server" is an SDK buried in someone's free mobile app.

Detection therefore shifts from infrastructure to behavior over time:

  • The same residential IP appearing as an exit across known proxy marketplaces.
  • Improbable session diversity: one household IP presenting dozens of device fingerprints per hour.
  • Churn patterns matching proxy-network rotation rather than DHCP renewal.
  • ISP-type mismatch: mobile-carrier IPs producing sustained datacenter-style request volumes (the guide to spotting mobile proxies covers why carrier IPs are a special case).

This is why residential exits get a separate field from is_proxy: the evidence class is different, and so is the appropriate action. Be clear about what an IP lookup can see here. GeoIPHub's is_residential_proxy is a coarse, ASN-based inference, not a direct observation of proxy-network membership, so treat it as a hint rather than proof. If home-ISP traffic suddenly looks suspicious in your logs, why residential proxies keep showing up in your traffic walks through the usual causes. A residential proxy hit on a checkout flow is a much stronger fraud signal than a consumer VPN hit.

Tor Directory Data

Tor is the easy one, by design. The Tor network publishes its consensus of relays and exit nodes, refreshed continually. Cross-referencing the public exit list yields near-perfect is_tor accuracy with one caveat: timing. Exit nodes join and leave, so the directory needs to be re-ingested on a tight schedule, not weekly.

Privacy relays are the opposite case. Apple publishes every iCloud Private Relay egress address, so a relay can be matched exactly too, but on its own a relay belongs in allow, not on a blocklist. We cover both relays, and why Cloudflare publishes no WARP egress list, in iCloud Private Relay and Cloudflare WARP detection.

Cross-Signal Confidence

The final layer is the merge. Each preceding layer produces evidence with a different reliability profile:

Provider lists

  • Catches: Commercial VPNs
  • Confidence when it fires: Very high (attributed)
  • Blind spot: Private VPNs

ASN / hosting

  • Catches: Datacenter anonymizers
  • Confidence when it fires: Medium (prior, not proof)
  • Blind spot: Residential exits

Active probing

  • Catches: Live VPN/proxy daemons
  • Confidence when it fires: Very high (verified)
  • Blind spot: SDK-based proxies

Port fingerprints

  • Catches: Unbranded proxies
  • Confidence when it fires: High
  • Blind spot: Firewalled endpoints

Latency checks

  • Catches: Geo-spoofing relays
  • Confidence when it fires: Medium
  • Blind spot: Nearby relays

Behavioral models

  • Catches: Residential proxies
  • Confidence when it fires: Medium-high
  • Blind spot: Brand-new IPs

Tor directory

  • Catches: Exit nodes
  • Confidence when it fires: Near-certain
  • Blind spot: Bridges

A defensible verdict weighs these together. One medium-confidence signal should not flip a boolean; two independent high-confidence signals should. The output you want from a detection API is therefore three things: the booleans for routing decisions, the named detection methods for auditability, and a risk score that expresses how much the signals corroborate. The named methods are also what a wrongly flagged user needs: our walk-through of why an IP gets flagged as a VPN or proxy reads them on real addresses.

Putting VPN detection to work: route on combinations

Three practical takeaways:

  1. Do not treat is_vpn as a block list by itself. Plenty of paying customers use consumer VPNs (see how to flag anonymized traffic without turning away real customers). Route on combinations: VPN + datacenter + abuse history is a very different situation from VPN alone.
  2. Weight residential-proxy signals heavily, knowing they are inferred. Nobody rents residential exits for privacy; the legitimate-use base rate is tiny compared to VPNs. An IP-level residential flag is a coarse inference, though, so pair it with your own session signals.
  3. Prefer verified detection over static lists. Ask any vendor how they know an IP is a VPN. If the answer is "a database we license," the data is already decaying. Active probing and continuously crawled provider lists keep pace with infrastructure churn.

You can see every layer's output yourself: run any IP through the free IP lookup tool or test your own VPN connection, no signup required.

Frequently Asked Questions

Can VPN detection ever be 100% accurate?

No. Detection is probabilistic. Commercial VPNs with published infrastructure are detected with very high confidence, while small private VPNs on fresh cloud IPs are harder. That is why good APIs return per-signal evidence and a confidence level instead of a bare boolean.

What is the difference between a proxy and a VPN at the detection level?

A VPN tunnels all traffic at the network layer and usually answers specific protocol handshakes (OpenVPN, WireGuard, IKEv2) on its server. A proxy relays application-level traffic (HTTP, SOCKS) and exposes different ports and fingerprints. Detection systems probe for both families separately.

Why are residential proxies so hard to detect?

Because the exit IP genuinely belongs to a home ISP, list-based and ASN-based checks pass. Detection relies on behavioral evidence instead: the same IP appearing in proxy networks, abnormal reuse patterns, and mismatches between the ISP type and the observed traffic.

Does blocking every VPN user make sense for fraud prevention?

Usually not. Many legitimate users browse through VPNs for privacy. A better pattern is to treat VPN use as one weighted input to a risk score and only escalate when it co-occurs with other signals such as datacenter hosting, abuse history, or impossible travel.

How fresh does VPN detection data need to be?

Very fresh. VPN providers rotate servers daily and proxy networks churn IPs by the hour. Detection backed by continuous re-probing and re-crawled provider lists stays accurate; static datasets decay within weeks.

Written by

GeoIPHub Team

The team behind the GeoIPHub API

Written by the GeoIPHub team, the people who build the GeoIPHub IP intelligence API and its scanner. Drafts are prepared with AI assistance and edited by the team; API examples are real lookups against the live API unless labelled illustrative.

Share this article
More articles

Related posts