VPN Detection False Positives: Who Gets Wrongly Flagged, and What to Do
Keep real customers: why VPN detection false positives hit privacy users, corporate VPNs and mobile CGNAT, and how evidence-based scoring still catches fraud.

For a decade the fraud rule was simple: hidden IP, higher risk. Flag the VPNs, block the proxies, move on. In 2026 that rule quietly bleeds revenue, because the population hiding behind anonymized IPs has changed. A growing share of it is your best customers: iPhone users with iCloud Private Relay turned on, privacy-conscious shoppers on a VPN, and mobile buyers sharing one carrier address with thousands of neighbors. 47% of merchants report that false declines cost them sales, and Apple tells network operators the same thing from the other side: "Traditional fraud detection that relies on IP addresses might need to be adjusted to ensure legitimate users aren't impacted" (Apple, Prepare your network for iCloud Private Relay).
Meanwhile the traffic you actually need to stop has gone the other way: the most dangerous attackers no longer come from obvious datacenter ranges at all. They come through residential IPs that look exactly like a real customer. So the binary "block anonymized traffic" instinct now fails twice: it punishes real users and misses the real threats. Here is how to tell them apart, starting from what VPN and proxy detection can actually see.

If you landed here as the person being blocked, not the site doing it, read what a wrongly flagged visitor can do.
Where VPN detection false positives come from
Most VPN detection false positives come from legitimate people whose IP is hidden or shared for ordinary reasons. A large slice of anonymized traffic is paying, non-fraud users:
- Consumer VPNs are mainstream. In Security.org's August 2026 survey of 1,008 US adults, 42% said they use a VPN, and the top reasons are general privacy (56%) and security (51%), with not a single fraud-related motive in the list. Block all VPN traffic and you are turning away privacy-conscious customers by design.
- Corporate VPNs and travelers. Employees routed through a company VPN, or customers connecting from abroad, legitimately surface on "foreign" or datacenter-adjacent IPs every day.
- Mobile users behind CGNAT. Carriers ran out of IPv4 long ago, so mobile networks place huge numbers of real subscribers behind a handful of shared addresses via Carrier-Grade NAT. One mobile IP can front thousands of genuine customers. Block it on one bad session and you block a city block's worth of real buyers. That same shared address space is exactly what makes mobile proxies so hard to detect.
- Cloudflare WARP. A free consumer privacy service, not a fraud proxy; here is how to recognise WARP traffic.
And then there is the case that breaks naive detection completely: Apple iCloud Private Relay.
iCloud Private Relay is not a VPN, so don't score it like one
Private Relay is a privacy relay, not a VPN: per Apple's own overview, it splits each connection across two hops so no single party sees both the user and the destination, and it does not let users appear to connect from a different country or region. Apple validates every connection at the device and account level and limits the service to Apple devices with a paid iCloud+ subscription, so flagging a relay IP as an anonymous proxy declines verified paying customers. Our guide to detecting iCloud Private Relay and WARP covers how to recognise the traffic and let it through.
The malicious side: residential proxies are the real threat
So what should set off alarms? The category that was built to defeat exactly the "trust residential IPs" instinct: residential proxy networks.
A residential proxy routes an attacker's traffic through an IP owned by an ISP and assigned to a real home or small business, usually harvested by trojanized apps and SDKs silently enrolling consumer devices as exit nodes. To your server it looks like an ordinary customer on a consumer ISP. That is the whole point, and it is why Google's Threat Intelligence Group calls them a serious challenge for defenders: when GTIG disrupted the IPIDEA network in January 2026, it found over 550 distinct tracked threat groups (including state-linked actors from China, North Korea, Iran, and Russia) using IPIDEA exit nodes in a single seven-day window for SaaS intrusions, on-prem access, and password-spray attacks.
This is the inversion that breaks legacy rules: the datacenter VPN is mostly your customer; the "residential" IP may be the attacker. And the volume pressure is real: automated bots now exceed human web traffic, at more than 53% of all requests in 2025, with AI agents emerging as a new category that blurs legitimate automation and abuse even further. Our guide to detecting bot traffic by IP sorts that automation into classes, from verified crawlers to scanners, and shows which ones an IP lookup can identify.
Anonymized traffic, ranked by real risk
Not every anonymizer carries the same risk, so the type of anonymizer should set the default action. A useful first cut:
Apple iCloud Private Relay
- Typical risk: Low (verified Apple customers)
- Default action: Allow; treat location as approximate
Cloudflare WARP / consumer privacy
- Typical risk: Low to medium
- Default action: Allow, score on behavior
Mobile / CGNAT shared IPs
- Typical risk: Low to medium (shared)
- Default action: Never hard-block the IP; use user-level signals
Corporate / consumer VPN
- Typical risk: Medium, mixed-use
- Default action: Score, challenge if other signals stack
Residential proxy
- Typical risk: High (built to evade)
- Default action: Challenge or block on strong evidence
Abusive datacenter proxy / known botnet
- Typical risk: High
- Default action: Block
The lesson isn't "trust everything" or "block everything." It's that the type of anonymizer, plus corroborating evidence, should drive the decision.
What a false decline costs you
A false positive at checkout is a false decline: a real customer turned away because their IP looked anonymized. 47% of merchants report that false declines cost them sales. Every wrongly blocked Private Relay user, mobile shopper, or VPN-using buyer is lost revenue, a support ticket, and often a customer who doesn't come back. Fraud you can measure; the sale you never made because you blocked a real person is invisible, which is exactly why blanket blocking feels safe and quietly isn't.
The playbook: score on evidence, not on a single flag
The way out is detection that answers "what does the evidence say about this IP, and how confident are we?" rather than "is any anonymizer flag set?" In practice:
- Identify the anonymizer type, not just 'anonymous': Private Relay, consumer VPN, corporate VPN, residential proxy, or hosting. They carry very different risk.
- Use exonerating signals: a residential ISP or a verified crawler lowers the score in our model, and a known privacy service such as Private Relay is capped so it can never reach block. None of them should raise risk on their own.
- Cap shared infrastructure: never hard-block a CGNAT or mobile carrier range on one bad session, because thousands of real users sit behind it.
- Weight detection recency: an IP last seen in a proxy pool an hour ago is very different from one whose only evidence is six months stale.
- Respond gradually: allow low risk, step up verification in the middle, and block only high-confidence fraud.
This is exactly how GeoIPHub's 0 to 100 fraud score is built. Exonerating signals are first-class: a residential ISP subtracts points, a verified crawler subtracts more, and CGNAT ranges are capped, so legitimate privacy traffic stays low. Every lookup returns the connection type, provider attribution, an is_residential_proxy flag (a coarse, ASN-based inference, so treat it as a hint rather than proof), and the detection_methods that fired, so you can act and explain the decision. A blocked checkout should never be a black box.
A graduated response, not a wall
Here are three real lookups we ran on 12 October 2026, trimmed. First, 172.224.226.1, an Apple iCloud Private Relay egress address:
{
"ip": "172.224.226.1",
"asn": { "org": "AKAMAI-AS - Akamai Technologies, Inc.", "connection_type": "datacenter" },
"geo": { "country_code": "GB", "city": "London" },
"detection": { "is_vpn": false, "is_proxy": false, "is_tor": false, "is_relay": true,
"is_anonymous": false, "relay_provider": "iCloud Private Relay" },
"scoring": { "fraud_score": 24, "recommended_action": "allow",
"detection_methods": ["datacenter_ip", "no_ptr_datacenter", "relay_ip"] }
}
Second, 185.220.101.1, a Tor exit node:
{
"ip": "185.220.101.1",
"asn": { "org": "TORSERVERS-NET - Stiftung Erneuerbare Freiheit", "connection_type": "datacenter" },
"geo": { "country_code": "DE", "city": "Hamburg (Hamburg-Nord)" },
"detection": { "is_vpn": false, "is_proxy": false, "is_tor": true, "is_relay": false,
"is_anonymous": true, "anonymity_level": "high" },
"threat": { "is_spammer": true, "threat_types": ["spammer", "tor"] },
"scoring": { "fraud_score": 100, "recommended_action": "block",
"detection_methods": ["tor_exit", "datacenter_blocklisted", "rdns_tor", "known_spammer"] }
}
Both addresses sit on datacenter networks and both hide the user's home IP. The Private Relay address is named as a relay, not marked anonymous, and scores 24: allow. The Tor exit is also on a spam blocklist and scores 100: block. Same lookup, opposite outcomes, because the evidence is different.
Third, 45.83.220.7, a Mullvad VPN server in Malmö, which is the case most VPN rules get wrong:
{
"ip": "45.83.220.7",
"asn": { "org": "ESAB-AS - 31173 Services AB", "connection_type": "datacenter" },
"geo": { "country_code": "SE", "city": "Malmö" },
"detection": { "is_vpn": true, "is_hosting": true, "is_anonymous": true,
"vpn_provider": "Mullvad (31173 Services)", "vpn_confidence": 85 },
"threat": { "is_abusive": false, "is_spammer": false, "threat_types": ["vpn"] },
"scoring": { "fraud_score": 46, "recommended_action": "review",
"detection_methods": ["known_vpn_server", "datacenter_ip"] }
}
The VPN is identified by name with high confidence, yet the score stops at 46 and the action is review, not block, because nothing else in the record points to abuse. A binary "is_vpn means block" rule would have declined this user. Since our 10 November 2026 scoring change, a consumer VPN with no abuse, spam, botnet, Tor or proxy evidence stops at 70, so VPN evidence alone never reaches block; on 11 November this address returned 70, step_up. A company VPN gateway gets its own, lighter reason, corporate_vpn_gateway, worth 12 points. A graduated policy lets them through on low-risk actions and asks for a step-up check only where the stakes justify it. A residential-proxy flag deserves the same caution: because GeoIPHub only infers residential-proxy status coarsely from ASN data, it should earn a challenge or block only when other recent evidence backs it up.
You can see how any address scores, including Private Relay and VPN endpoints, with the free VPN & proxy detection test and the IP fraud score checker, or read exactly how the score is built in the detection methodology and full API reference.
Catch the residential proxy, let the iPhone customer check out
The anonymized internet is no longer a fraud signal by itself. Apple Private Relay, WARP, corporate VPNs, and mobile CGNAT put real customers behind "hidden" IPs, while the actual attackers hide inside residential ranges designed to look legitimate. Detection that still works in 2026 doesn't ask whether an IP is anonymized. It asks which kind, how recently, and with what corroboration, then responds in proportion. Catch the residential-proxy fraud. Let the iPhone customer check out.
Frequently Asked Questions
What causes VPN detection false positives?
A rule that treats any hidden or shared IP as fraud. It flags privacy-conscious shoppers on consumer VPNs, employees on corporate VPNs, travelers, mobile users who share one carrier IP through CGNAT, and Apple iCloud Private Relay users. Most of them are legitimate customers, so a block on the IP alone turns them away.
Does blocking iCloud Private Relay cause false positives?
Yes. Private Relay is not a VPN: Apple routes each connection through two hops, limits the service to Apple devices with a paid iCloud+ subscription, and does not let users appear to connect from another country. Apple's guidance for network operators warns that traditional fraud detection that relies on IP addresses might need to be adjusted to ensure legitimate users aren't impacted. Treat a relay IP as low risk and enforce fraud rules with account and session signals.
What anonymized traffic is actually dangerous?
Residential proxy networks are the genuinely high-risk category. They route attacker traffic through real ISP-owned home IPs harvested via trojanized apps, so it blends in with legitimate users. Google's January 2026 disruption of the IPIDEA network found 550+ tracked threat groups using its exit nodes in a single week. Commercial VPNs and Private Relay are mostly legitimate; residential proxies rarely are.
Why is blocking all VPN and proxy traffic a bad idea?
Because a large share of it is real customers. Roughly a third of U.S. adults use a VPN, overwhelmingly for privacy and security rather than fraud, and mobile users routinely share one carrier IP via CGNAT. Hard-blocking those IPs produces false declines, and 47% of merchants report that false declines cost them sales. The goal is graduated risk scoring, not a blanket block.
How does GeoIPHub avoid false positives?
By scoring on evidence, not on the mere presence of a flag. The 0 to 100 fraud score includes exonerating signals (a residential ISP subtracts points, a verified crawler subtracts more, and CGNAT ranges are capped), and every lookup returns the connection type, provider attribution, detection recency (last_seen), and the detection_methods that fired, so you can challenge the genuinely risky and let real customers through.
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.


