Skip to content
iplogs.com
Guide · Last updated 2026-05-24

Ethics and privacy in IP intelligence

An IP-intelligence service makes judgements about people's network behaviour, usually for a fraud, compliance or abuse team. Whether those judgements help or hurt comes down to two things: what data the service collects and exposes, and how its customers apply the verdicts. Below is what we do and refuse to do, followed by a standard you can hold any vendor to, us included.

By Mehmood Ali ·

What an IP-intelligence service should and shouldn't do

ShouldShouldn't
Surface every signal that contributed to a verdictReturn an opaque score with no per-signal provenance
Publish a measurable false-positive rate against a labelled corpusClaim accuracy without a falsifiable benchmark
Treat Apple Private Relay, Cloudflare WARP, and CGNAT as separate categoriesLump them in with commercial VPNs to inflate detection rates
Keep request logs transient and minimalBuild a persistent database of who looked up what
Document what each signal means and how it is weightedClaim “proprietary signals” that can't be audited
Be honest about what IP intelligence cannot verifyPosition IP detection as identity, age, or location proof

What IPLogs concretely does

  • No signup, no API key. The detection endpoint needs no authentication at all. Since nobody has an account, a check can't be tied back to a customer.
  • Short-lived logs only. Access logs are kept for abuse prevention, rate limiting and aggregate visitor counts, then deleted after 30 days. There is no persistent database of individual lookups, and the site runs without tracking cookies, advertising integrations or third-party analytics. The full statement is in /privacy.
  • Per-signal provenance in every response. Every response includes a typed Signal[] array listing each feed and probe that matched, with its weight and a readable detail string, so the score can always be checked against what produced it.
  • Published false-positive rate. /accuracy documents the ground-truth methodology and the error rates we measured. If a vendor won't tell you its error rate on a labelled corpus, you have no way to check its accuracy claims.
  • Apple Private Relay and Cloudflare WARP are separate categories. Both are mainstream privacy products that ordinary consumers use. Counting them as commercial VPN exits would punish real users, so the API reports each as its own signal type and your policy can treat them differently.
  • CC-BY 4.0 datasets at /tools. The aggregated VPN-provider, datacenter, and Tor-exit datasets are public under CC-BY 4.0, so other detection projects can work from the same ground truth and check the data for themselves.

The legitimate privacy critique

Privacy-rights groups including the Electronic Frontier Foundation, Access Now, and Article 19 have criticised the use of IP detection for compliance, especially where laws require sites to verify a user's physical location even when that user is on a VPN. On the substance, they are right:

  • IP detection is probabilistic. Treating its output as a legal determination is a category error.
  • Blanket VPN blocks fall hardest on privacy-conscious users, journalists, activists, people in censored regions and ordinary travellers, and the rule was probably not written with any of them in mind.
  • Data that helps prevent fraud can do real harm once a state actor, a stalker or an employer exceeding its authority gets hold of it.

Where we draw the line: detection is for fraud prevention, payment risk and signup abuse. It shouldn't be used for identity verification, blanket country bans or watching specific people. Tooling that shows its working makes it easier for the operator and the user alike to see which side of that line a given use falls on.

Practical rules for fraud teams

  • Block writes, allow reads. On a confirmed VPN, block the high-risk actions: signup, payment, KYC, withdrawal, admin. Weaker signals get step-up auth, and read traffic stays open. /guides/block-vpn-traffic has the full pattern.
  • Always provide a dispute path. Tell users their connection looked anonymized and give them a way to reach support. You will get false positives, and each one costs far less when the user can ask a person to take a look.
  • Log decisions, not just inputs. Record which verdict led to which action, so an auditor (or you, months later) can work through the edge cases. A log line that only says “blocked” tells you nothing. Keep the raw response next to the action and you can reconstruct what happened.
  • Don't use IP detection as identity or age verification. It can't do either job. Identity verification (document-based or credit-bureau-based) is a different product category, and the regulations overview explains where the line sits.

FAQ

Does IPLogs store the IP addresses it checks?

Only in the web server's access logs (IP, page, user agent, time). Those are kept for 30 days for abuse prevention, rate limiting and aggregate visitor counts, then deleted. There is no persistent database of individual lookups, and the site uses no tracking cookies, advertising integrations or third-party analytics. The full privacy and storage statement is at /privacy.

Is there a public API key or signup that would let me identify users?

No. /v1/check is fully unauthenticated, with no signup, API key or token. Since there are no accounts, we have no way to attribute a check to a particular customer. That is deliberate: we don't collect the data that would make it possible.

How transparent is the detection methodology?

Each signal that feeds a verdict comes back in the API response as a typed Signal with a weight, a matched flag and a human-readable detail string. The signal types are documented at /docs, and the false-positive rate is published at /accuracy. A closed proprietary score is the industry default, and we chose not to ship one.

Is blocking VPN users ethical?

It depends on the action and the user. Hard-blocking read traffic is rarely justified, because that is where Apple Private Relay, Cloudflare WARP, CGNAT mobile users, corporate VPNs and privacy-conscious customers all turn up. A per-action policy (block writes, step up auth on risk, allow reads) does much less collateral damage. /guides/why-vpn-blocking-fails makes the technical and product case in more detail.

How does IPLogs handle private and CGNAT addresses?

Private (RFC 1918), loopback, link-local, multicast, IPv6 ULA and IPv4-mapped IPv6 addresses are recognised at the routing layer. For those, the per-IP page explains what the visitor is looking at rather than returning a verdict that would mean nothing. CGNAT space (100.64.0.0/10) is labelled as carrier-grade NAT, since the same public IP may be shared by many subscribers.

Related: privacy policy, published accuracy benchmark, why complete VPN blocking can't be a compliance solution, and the API docs for the unauthenticated endpoint surface.