WebRTC IP leak detection comes down to one comparison: the address your server sees on the HTTP request, and the address the browser exposes through its real-time communication stack. When those two disagree, you have a signal — one that is easy to collect and easy to misread.
The HTTP IP tells you where a request came from. It says nothing about whether the same browser is also reachable on another path. A proxy or VPN can cover the HTTP request while leaving UDP traffic — the transport WebRTC uses — on a different egress. That gap is what makes a session look like one thing in your logs and behave like another.
The useful move is not to block on that gap. It is to classify it: which mismatches reflect normal network design, which ones point to a proxy covering only part of the traffic, and how much weight the difference deserves in your risk score.
What WebRTC actually exposes
When a page creates an RTCPeerConnection, the browser gathers ICE candidates to find a path for media. Those candidates come in a few flavors: host candidates (local interfaces), server-reflexive candidates (the public address as a STUN server sees it), and relay candidates (a TURN server).
The server-reflexive candidate matters most for fraud work. It is the browser’s public address from the perspective of an independent server, gathered over UDP. Browsers hide host candidates behind mDNS hostnames and may skip gathering entirely, but when a server-reflexive candidate appears you learn something the HTTP request did not tell you.
Map the mismatch types before writing rules
Same address, same story
HTTP IP and server-reflexive IP match. This is the baseline for most traffic, VPN or not: a full-tunnel VPN produces matching addresses because both paths leave through the same egress.
Public mismatch: proxied HTTP, direct WebRTC
The HTTP IP belongs to a hosting provider, the WebRTC address belongs to a residential ISP. A proxy that terminates HTTP but not UDP produces exactly this. It is one of the clearest tells that the request path and the browser path are not the same.
Geographic disagreement
HTTP resolves to one country, the WebRTC address to another. That often means a split tunnel, a misconfigured proxy chain, or a browser routing media outside the tunnel. It is a real inconsistency, not automatically fraud, but it does contradict a claim of a single location.
Address family split
HTTP arrives over IPv6 while WebRTC reports IPv4. Two different strings, one host. Normalize before comparing.
No candidates at all
Gathering blocked by extension, policy, or browser settings. Absence is not a leak; it is a missing input. Treat it as unknown.
Which mismatches are noise
Carrier-grade NAT on mobile networks can rotate the public address between requests, so a mismatch may simply be two NAT egresses in the same carrier pool. Corporate split tunneling intentionally routes only some traffic through the VPN. Privacy relays on some platforms exit traffic in places you cannot map cleanly to a residential or hosting ASN.
Normalize to ASN and country instead of raw address strings, and check IPv4 and IPv6 separately before you compare. Most of the noise then collapses into “same network, different address.”
Turn the gap into a score
Design the rule around a small set of reason codes, then let your policy decide:
ip_country_mismatch HTTP country != WebRTC country
ip_asn_type_mismatch hosting HTTP + residential WebRTC, or inverse
rtc_gathering_absent no server-reflexive candidate
ip_family_split IPv4/IPv6 comparison, informational only
Give each code a weight you can defend, and enforce in stages. Log only, for a week or two: does the code fire on your QA scripts, on support staff behind a corporate VPN, on mobile users? Then feed the score into the decision you already make — allow, challenge, block — next to accountsOnDevice, bot signals, and linkedId matches. A country mismatch on a signup with three accounts on the same device deserves a different response than the same mismatch on a returning customer with a long session history.
For the challenge path, prefer step-up over rejection: email verification, a proof-of-work challenge, or Form Shield on the signup form. That keeps the false-positive cost low and still raises the price of automated multi-accounting.
One more boundary: WebRTC data arrives from the browser, so it is client-supplied. Verify it server-side, treat sealed results and passToken values as evidence rather than truth, and re-derive the address comparison on your own infrastructure.
Store the conclusion, not the address
Keep the reason code, the country pair, and the ASN type. You rarely need the full address for the decision, and a smaller footprint makes retention and erasure requests easier to answer. If you operate in jurisdictions where IP addresses count as personal data, have counsel review what you keep and for how long.
FAQ
Is a WebRTC IP mismatch proof of a VPN?
No. It is proof that two network paths disagree. Split tunneling, mobile NAT, dual-stack addressing, and privacy relays all produce mismatches. The signal gains value only when combined with the rest of the request.
Why does the WebRTC address change between two requests?
Mobile carriers and some VPN pools rotate public addresses. Compare country and ASN rather than exact addresses, and check whether the rotation stays inside one provider range.
Should I block when the two addresses disagree?
Only after you have measured the mismatch against your own traffic. Start with logging and a challenge path, then tighten to block for the combinations that never appear in legitimate sessions.
Try it free
Run the comparison on your own signup flow before you enforce anything: collect the HTTP address and the WebRTC address, tag the mismatch with a reason code, and watch how it behaves next to your existing signals. You can start on the free tier — 20,000 identifications per month without Smart Signals — and add policy once the codes are stable. Get started at pryntid.com.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.