IPv6 Leaks: The Second Address Your Proxy Forgot
Run ipconfig on a Windows machine at home, or ip addr on Linux, and there’s a fair chance you’ll see an address starting with a 2 that’s around 39 characters long. That’s a public IPv6 address. Your router hides your IPv4 address behind one shared public one, and this address gets no such cover. It’s globally reachable, it’s assigned to your machine specifically, and the first half of it identifies your home connection. Now point a browser profile at an IPv4 proxy in another city and load a leak test. The page shows the proxy’s IPv4 address, which is what you wanted, and then a second line underneath it with that long address, which is what you didn’t. That’s an IPv6 leak. It gets one line in most leak checklists. It deserves more, because the way IPv6 addresses work makes the leak more identifying than an IPv4 leak would be, and the ways it escapes aren’t the ones people expect.
Two internets running side by side
Most of the internet now runs two address systems at once. IPv4 is the old one, with about four billion addresses, long since exhausted. IPv6 is the newer one, with so many addresses that every device can have its own public one without anyone sharing.
A machine that has both is called dual stack. Most home broadband connections in many countries now hand out IPv6 alongside IPv4, and many mobile networks do too. Your operating system has had IPv6 switched on by default for years. So a lot of people running proxies are on dual stack machines without having chosen it or noticed it.
When a dual stack machine connects to a site that supports both, it has to pick one. Modern operating systems and browsers use an approach called Happy Eyeballs, which tries both and uses whichever answers quickly, with a preference for IPv6. For big platforms, which almost all support IPv6, that often means the connection goes out over IPv6.
That matters for proxies because many proxy setups only deal with IPv4. The proxy address itself is IPv4, the tunnel runs over IPv4, and whatever route the tool configures on your machine covers IPv4. The IPv6 side of your machine is left exactly as it was, with its own route straight to the internet through your ISP.
If anything in your setup sends traffic without going through the proxy tunnel, IPv6 is the easiest path out, because nothing was ever set up to stop it.
Why an IPv6 leak says more about you
An IPv4 leak exposes your home IP, which is bad enough. An IPv6 leak can expose more, for two structural reasons.
The first is that there’s no NAT hiding the device. With IPv4, your router has one public address and every device behind it shares that address. Your laptop’s own address is a private one, like something starting with 192.168, which is meaningless outside your home. With IPv6, your laptop usually has its own public address. A leak shows the world an address that belongs to that one machine, not just to your household.
The second is the prefix. An IPv6 address splits into two halves. The first 64 bits are the network prefix, assigned by your ISP to your connection. The last 64 bits identify the device on that network. Every device in your home shares the same prefix. And ISPs typically assign a whole block of prefixes to each customer, which stays the same for long periods on many connections, though that varies by provider.
So an IPv6 address tells a site which household it came from, via the prefix, and often which device, via the second half. Platforms know this. The usual practice for IPv6 rate limiting and reputation is to treat a whole /64, or a larger block, as one unit, the way a single IPv4 address is treated. If two of your accounts ever leak IPv6 addresses from the same prefix, they’re tied to the same connection.
Privacy addresses don’t solve it
Operating systems know that a fixed device identifier in every IPv6 address is a tracking problem, so they use privacy extensions. Windows, macOS, iOS and Android all generate temporary IPv6 addresses whose second half is random and changes regularly, often daily. Outgoing connections use the temporary address.
That helps against a site following one device over months. It doesn’t help much with multi account linking, because only the second half changes. The prefix stays the same. Every temporary address your laptop generates, and every address every other device in your home generates, sits under the same household prefix.
A platform that groups by /64 sees all of those addresses as one connection. If account A leaks one temporary address on Monday and account B leaks a different temporary address on Friday, both carry your prefix. The randomised half made them look different to a person reading them. It didn’t make them different to a system grouping by prefix.
There’s also a second kind of stable address many systems keep alongside the temporary ones, used for incoming connections. Some systems generate it from the hardware address, some generate it in a way that’s stable per network. Outgoing browsing rarely uses it. It still shows up in some leak paths, including local interface lists that WebRTC can expose.
The practical point is simple. You can’t rely on IPv6 privacy settings to keep your real connection out of a leak. If an IPv6 address escapes, assume it carries your household with it.
The WebRTC path
The best known way an IPv6 address escapes is WebRTC, the browser feature that powers video calls and peer to peer connections.
To set up a call, WebRTC gathers candidate addresses that another party could use to reach you. It lists local interface addresses, called host candidates, and addresses learned by asking a STUN server what your public address looks like from outside. A page can trigger this process and read the candidates without starting any call.
With IPv4 behind NAT, the host candidates are private addresses, and browsers now hide them behind random local names by default. The public IPv4 only shows up through the STUN query, which travels over UDP.
IPv6 changes things. Your IPv6 host address is already public. If the browser lists it, the page has your real public IPv6 without needing STUN at all. And the STUN query itself can go out over IPv6 directly, since an HTTP or SOCKS proxy for web traffic usually doesn’t carry that UDP.
Browsers have tightened this over the years, and the exact behaviour depends on the browser version and its WebRTC policy setting. Chrome has a policy with options that range from exposing all interfaces to disabling any UDP that doesn’t go through the proxy. Antidetect tools usually have their own WebRTC setting that either blocks it, replaces the addresses with the proxy’s, or passes through the real ones.
The only way to know what your profile does is to test it, which comes later in this piece.
The DNS path
The second path out is name resolution.
Before a browser connects to a site, something has to turn the domain name into an address. For a dual stack machine, that means asking for both an A record, the IPv4 address, and an AAAA record, the IPv6 address. Where that question gets sent decides who sees it.
If the profile’s proxy handles DNS on the far side, the question goes through the proxy and your machine never asks anyone directly. SOCKS5 can do this when the client sends the hostname rather than a resolved address. HTTP proxies do it naturally for tunnelled connections, since the browser gives the proxy a hostname.
If resolution happens locally, the question goes to whatever DNS server your machine is set to use, usually your ISP’s or your router’s, and that request can travel over IPv6. The DNS server then knows your real connection asked for that platform’s domain at that moment. Leak tests show this as a list of resolvers, and a resolver belonging to your home ISP next to a proxy in another country is the tell.
DNS leaks don’t hand the platform your IPv6 address directly, since the platform isn’t usually the resolver. But large platforms and CDNs run their own authoritative DNS, and some techniques let a site match a DNS lookup to the page load that followed it. Leak test pages use exactly that technique to show you your resolvers. A platform can do the same.
Direct requests that skip the proxy
The third path is traffic that never touches the proxy at all.
A browser profile with a proxy configured sends its web requests through that proxy. But not everything a machine does goes through the browser. Other apps, background services and system updates use the machine’s normal routes. None of that reaches the platform as the account, so most of it doesn’t matter.
It matters when something related to the account does go direct. A desktop app for the same platform, running on the same machine, with no proxy set. A browser extension that makes its own network calls in a way the proxy doesn’t cover. A helper process some tools use for sync or updates. A UDP based protocol like QUIC that the browser tries alongside the tunnel, if nothing blocks it.
System wide proxy tools add another variety. Some tools capture traffic at the operating system level instead of per browser, and some of them only install IPv4 routes. Everything IPv4 goes through the tunnel. Everything IPv6 follows the machine’s normal IPv6 default route, straight out through your ISP. For a platform that supports IPv6, which most do, the browser may simply connect over IPv6 and never touch the tunnel.
That last case is the most complete leak there is. The page itself loads over your real address. The leak test shows your real IPv6 as the main address, not as a side note.
A worked example with two profiles
Here’s a hypothetical to make the prefix problem concrete.
An operator runs two browser profiles on one Windows laptop at home. Profile A goes out through an IPv4 proxy in one city. Profile B goes out through a different IPv4 proxy in another. The home connection is dual stack. Nobody has touched the WebRTC setting in the tool, which is left on a mode that passes real addresses through.
On Monday, a page in profile A runs a WebRTC check. It collects the laptop’s temporary IPv6 address. Call it an address under prefix X, ending in random digits.
On Thursday, the laptop has generated a new temporary address. A page in profile B runs the same check. It collects the new address. Different ending. Same prefix X.
To the operator, those look like two different addresses. To a system that groups IPv6 by /64, they’re the same connection. Profile A and profile B both touched household X. Their IPv4 proxies were different. Their canvas hashes were different. None of that matters for this particular edge.
Now add a third fact. Both proxies say one country. Prefix X geolocates to another.
So one small setting produced two problems at once. A link between the accounts. A location contradiction on each.
The fix is one setting and one test. Switch WebRTC to the mode that replaces or blocks non proxied addresses. Reload the leak test in both profiles. Confirm prefix X appears in neither.
Short work. It just has to be done before the accounts exist, not after.
What a platform actually sees
It helps to picture the leak from the platform’s side, because that tells you how much each kind matters.
In the worst case the account’s page loads come over your real IPv6. The platform sees your household prefix as the account’s connection. Every account that leaks the same way sits on the same prefix. That’s as direct a link as sharing one IPv4 address.
In a WebRTC leak, the page loads come through the proxy, but a script on the page collects a second address. Whether the platform runs such a script, and what it does with the result, varies. Where it does, it has a proxy address and a real address attached to the same session. A proxy address in one country and a home ISP address in another looks like a proxied user, which is a common thing to flag.
In a DNS leak, the platform may or may not see anything, depending on how it resolves and whether it correlates lookups.
Across all three, the location story is the other problem. The proxy says one city. The IPv6 geolocates to wherever you actually are. Geolocation databases are generally less precise for IPv6 than IPv4, but they usually get the country and ISP right, and that’s enough for a mismatch.
I can’t tell you how heavily any platform weighs each of these today. The mechanism is what’s certain. Your real connection was visible where it didn’t need to be.
Checking whether you have IPv6 at all
Start by finding out whether the machine has a public IPv6 address. Plenty of people worry about IPv6 leaks on connections that don’t have IPv6.
On Windows, run ipconfig in a command prompt and look for lines labelled IPv6 Address or Temporary IPv6 Address. On macOS or Linux, run ifconfig or ip addr and look for inet6 entries.
Then read them properly. An address starting with fe80 is link local. Every IPv6 capable machine has one, it only works on the local network, and it isn’t a leak. Addresses starting with fc or fd are private addresses used inside some networks, similar to the 192.168 range, and they aren’t public either. A global address usually starts with a 2 or a 3. If you see one of those, your machine is dual stack and IPv6 leaks are possible.
A quick browser check confirms it. Sites like test-ipv6.com tell you whether your connection can reach the IPv6 internet at all. Load it without any proxy first. If it reports no IPv6 connectivity, IPv6 leaks from your home connection aren’t your problem today, though that can change when your ISP or router updates.
If it does report IPv6, note the prefix you see, the first four groups of the address. That’s the thing you’re looking for when you test profiles next.
Testing a profile for the leak
Now test the actual profile with its proxy.
Open the profile and load a leak test page that shows IPv4, IPv6, WebRTC and DNS together. BrowserLeaks and ipleak.net both do this. Look at each line.
The IPv4 line should show the proxy’s address. The IPv6 line should show either nothing, meaning no IPv6 connection was made, or an IPv6 address that belongs to the proxy’s exit and geolocates to the same place. It should never show an address under your home prefix.
The WebRTC section should show no public address other than the proxy’s. If it lists an IPv6 address under your home prefix, the WebRTC setting isn’t covering IPv6. If it shows local names ending in .local, those are the browser’s obfuscated host candidates and aren’t a leak by themselves.
The DNS section should show resolvers that belong to the proxy’s side, or a public resolver that fits the proxy’s location. Your home ISP’s resolver means local resolution.
Then load test-ipv6.com in the profile. If it reports IPv6 connectivity and the address is yours, the browser has an IPv6 route outside the proxy.
Do this on every host machine, not just one, and repeat after any change to the tool, the operating system or your router.
Why the leak shows up later
A lot of IPv6 leaks aren’t there on day one. They appear months into a setup that was tested and clean, and the cause is almost always outside the profile.
ISPs roll IPv6 out to existing customers gradually. A connection that was IPv4 only when you set up your fleet can quietly become dual stack after a network upgrade, and nobody sends you an email about it.
Router firmware updates can switch IPv6 on by default. So can replacing the router, which the ISP sometimes does for you.
Laptops move. A machine that’s IPv4 only on your home network can pick up IPv6 on a coworking space’s Wi-Fi, a hotel network or a phone hotspot. Tethering to a phone on a mobile network that supports IPv6 is an easy way to go dual stack without realising.
Operating system updates occasionally change how addresses are chosen or re-enable adapters you’d adjusted. Antidetect tool updates can reset a WebRTC setting to a new default.
None of these are exotic. They’re the normal churn of a machine that’s in daily use. That’s why a leak test is a routine, not a one time box to tick. Run it again whenever the network, the router, the operating system or the tool changes. Otherwise run it once a month on each host.
If a profile that tested clean in spring leaks in autumn, check the list above before you blame the proxy.
Mistakes people make when fixing it
A few fixes look right and aren’t, or fix one path and leave another.
Disabling IPv6 in the browser only. There isn’t a clean browser level switch for this in most browsers, and the settings that exist don’t stop other routes on the machine. The adapter or the tunnel is where the decision lives.
Treating fe80 as a leak. Link local addresses are on every machine and never leave the local network. Chasing them wastes time and sometimes leads people to break things that were fine.
Blocking IPv6 traffic but leaving DNS local. The browser stops connecting over IPv6, but lookups still go to your home resolver over IPv4. The DNS leak remains.
Relying on a system tunnel’s name. Some tools advertise IPv6 support that covers blocking rather than routing, or only covers one direction. Test the actual behaviour.
Testing only one profile. The WebRTC setting can differ between profiles, especially if some were created before a tool update and some after.
Testing on the wrong machine. If profiles run on three hosts, a clean result on one says nothing about the other two.
Closing the leak
There are several ways to close it, and which you choose depends on your setup.
The cleanest is to make sure every browser request goes through the proxy tunnel, whatever the IP version. For per profile proxies in antidetect tools, that’s usually already true for page loads, because the browser hands hostnames to the proxy and never connects directly. What’s left is WebRTC and DNS.
For WebRTC, use the tool’s setting that replaces candidates with the proxy address or disables non proxied UDP. Then test again and confirm no home prefix appears.
For DNS, make sure resolution happens on the proxy side. With SOCKS5, that means the tool sends hostnames to the proxy instead of resolving locally. Most antidetect tools do this by default. Check anyway.
For system wide tunnels that only capture IPv4, either switch to a tool that captures IPv6 too or block IPv6 outbound while the tunnel is running.
The blunt option is turning off IPv6 on the machine’s network adapter, or on the router. It works. It also changes how that machine behaves for everything, and some software expects IPv6 to exist. If you do it, do it on a machine used only for fleet work, and remember it when you troubleshoot something unrelated months later.
Whatever you choose, the final step is the same. Test it and write down the result.
When the proxy has IPv6 too
The other way to handle IPv6 is to have the proxy provide it.
Some proxies are dual stack at the exit. The site can reach them over IPv4 or IPv6, and the proxy connects onward using whichever the site prefers. In that case an IPv6 address showing up on a leak test is fine, as long as it’s the proxy’s and it fits the same location.
This is actually the more natural picture for some kinds of traffic. A real phone on a mobile network that supports IPv6 connects to IPv6 capable platforms over IPv6. A real home broadband user on a dual stack line does the same. A profile that claims to be one of those people and never, ever uses IPv6 isn’t a red flag, because plenty of real networks are still IPv4 only. But a profile with a dual stack exit that fits its location looks fully ordinary.
With mobile proxies, whether IPv6 is available at the exit is decided per modem and per SIM, by what the carrier assigns on that connection. The host machine running the modems doesn’t tell you. On the farms I run, the only reliable way to know is to test through each specific modem, not through the server it’s plugged into.
If your proxy provider says it supports IPv6, test that too. Confirm the IPv6 exit geolocates where the IPv4 exit does, and that it isn’t shared with a different location’s pool.
Cloud phones and apps
On mobile the leak paths are different, because there’s usually no browser proxy setting in the picture.
A phone app uses the phone’s own network connection. If the phone is on a mobile data connection with a SIM, the app’s traffic goes out through the carrier, over IPv4 or IPv6 depending on the network. That’s the phone’s real connection, and for a real device that’s the point. Nothing is hidden, so nothing can leak around it.
Where it gets complicated is when a phone or emulator is pointed at a proxy through a VPN style app or a system proxy setting. Those have exactly the same IPv4 versus IPv6 question as a desktop tunnel. If the app only captures IPv4 and the phone has an IPv6 route through Wi-Fi or its own mobile data, apps can connect over IPv6 directly.
Android’s own VPN interface can capture both versions, but whether a given app configures it that way is up to the app. So the same testing rule applies. Load an IPv6 test inside the phone’s browser, and ideally check what address the target app’s own traffic uses.
For a hosted real device on a SIM, the cleanest arrangement is no proxy on the phone at all. The phone uses its own mobile connection, which is the connection that fits its SIM and its location.
What to do on Monday
If you’ve never checked IPv6 on your fleet, this is the short list.
Find out whether each host machine has a global IPv6 address. If none of them do, note that and recheck when your ISP or router changes.
For each host that does, open two profiles and run the combined leak test. Look at the IPv6 line, the WebRTC addresses and the DNS resolvers. Compare anything IPv6 against your home prefix.
Fix WebRTC in the tool settings first, since that’s the most common path and the cheapest fix. Then fix DNS if it’s local. Then decide whether to disable IPv6 on the host or leave it and rely on the tool.
If you use a system wide tunnel anywhere, check it specifically for IPv6 coverage, because that’s where the full leak happens.
Log the results, with dates, next to the other checks you keep for each profile.
The honest limit
Closing IPv6 leaks removes one route by which your real connection can get attached to an account. It doesn’t protect accounts that are linked some other way, and a clean leak test doesn’t mean a platform has nothing on you. It means that one particular door is shut.
It’s also a moving target. Browsers change their WebRTC defaults, operating systems change how they pick addresses, ISPs switch customers onto IPv6 with no announcement, and a router firmware update can turn it on overnight. A check you ran six months ago says little about today.
What makes the IPv6 leak worth special attention is the prefix. An IPv4 leak shows your household’s shared address. An IPv6 leak can show a prefix every device in your home shares, stable for long stretches on many connections, grouped by platforms the same way they group a single address. Two accounts that leak it are tied together whether they meant to be or not.
I run Singapore mobile proxy and cloud phone farms myself, so testing what each modem actually exits over is part of the routine rather than an afterthought. If you want the rest of the leak and network picture, start on the Multi Account Ops home page.
Get new guides and videos first — join the Telegram channel.