The internet still runs on two addressing systems that don't really get along: IPv4 is the one practically everyone still has to understand, with its familiar four-number addresses, and it ran out of available addresses to hand out years ago. IPv6 is the official successor, first standardized in 1998 and updated since, and despite nearly three decades of deployment work, it's widely used in some places and still absent or only partially implemented in many others. That leaves the internet stuck in a messy halfway house: IPv4-only in some places, IPv6-first in others, and dual-stack across the entire middle. And that's exactly the situation nobody wanted from a replacement to IPv4.

Every so often, though, someone drops a proposal claiming to have solved it. It's typically positioned as a third way that fixes addressing without the migration pain, often given a confident name and presented alongside lofty claims. Most of those go nowhere, even if a few become infamous. China's so-called IPv9, which was long, complicated, and never recognized by anyone who actually governs the internet, is an example of this, and it still managed to generate years of confused headlines. In April 2026, a new one turned up called IPv8, and it followed a strangely familiar arc: a single document, formatted like an official standard, promising to fix the internet's hardest problems all at once.

IPv8 is a real file sitting on a real IETF server, but it's not a standard; there's no working-group adoption, and there's no IETF endorsement of any kind. It's an individual Internet-Draft, which is about as close to an open upload as the standards world gets, and anyone can submit one. The man behind it, Jamie Thain, has openly admitted he used AI chatbots to help write it. And despite all of that, IPv8 drew a 100-plus comment pile-on on Hacker News, an interview in The Register, and quite an impressive amount of coverage across the internet as a whole. Unfortunately, it also carried enough surface-level legitimacy that those with only a casual interest in networking could easily mistake it for something far further along than it was.

IPv8 isn't a "hoax," but Thain isn't an anonymous troll lobbing a joke over the fence, either. He's a long-time IT entrepreneur who appears to be sincerely trying to build the thing, crowdfunding and all. While the documents look official, engineers have said that it's unworkable.

What IPv8 actually proposes

Zone servers, authentication, and more

Strip away the formatting and the core idea is simple enough to explain in a sentence. Thain's IPv8 uses 64-bit addresses written as eight decimal numbers, r.r.r.r.n.n.n.n, where the first four numbers are a routing prefix and the last four are a host address with, in the draft's own words, "identical semantics to IPv4." Thain frames it as something that can be likened to an area code bolted onto the front of a normal IPv4 address. The routing prefix is derived from an Autonomous System Number, the identifier that every network operator already uses to announce its routes to the rest of the internet.

The biggest consequence of that proposal is that every ASN holder would get its own full IPv4-sized address space, roughly 4.3 billion host addresses, all to itself. That would, in theory, remove the address-pressure argument for carrier-grade NAT, the address-sharing hack ISPs lean on because IPv4 ran dry. And the draft makes a much bigger claim on top of that: "IPv4 is a proper subset of IPv8." An IPv8 address with a routing prefix of 0.0.0.0 simply is an IPv4 address, which is how Thain arrives at his boldest claim, that IPv8 is "100% backward compatible" and can be rolled out as a software update with no flag day and none of the dual-stack misery he blames for IPv6's failure.

It doesn't stop there, as Thain's IPv8 bundles in a whole parallel suite of management protocols, most of them named by stapling an "8" onto something that already exists. There's BGP8, OSPF8, and IS-IS8 for routing, DHCP8 for address assignment, DNS8 for names, NTP8 for time, WHOIS8 for ownership lookups, ICMPv8, ARP8, plus a few entirely new ones like XLATE8 for translation and Route8 for, well, routing. By the author's own count, it cascades into around ten coordinated specifications.

Holding all of that together is the "Zone Server," and it's the most ambitious piece of the whole design. The draft describes it as a paired active/active platform that bundles DHCP8, DNS8, NTP8, an OAuth8 caching layer, WHOIS8 validation, access-control enforcement, logging, and address translation into a single management plane for the network. Security is baked directly into the third layer of the stack: every manageable element has to be authorized with OAuth2 JWT tokens pulled from a local cache, traffic between systems is gated by access-control rules, and traffic leaving the network gets validated against DNS8 and WHOIS8 lookups before it's allowed out.

There's even a routing metric called the Cost Factor, and a rule that no prefix more specific than a /16 can be injected across network boundaries, which Thain argues would cap the global routing table at roughly 175,000 entries instead of letting it balloon the way today's table has. One entry per ASN, in principle, makes it a structurally bounded table rather than one that grows forever.

The pitch essentially amounts to saying that IPv6 lost as an implementation; it's technically fine, but not worth the cost and effort of migrating to. That's why IPv8 tries to solve addressing, routing-table growth, network management, and security as a single package you can deploy on the hardware you already own. It's a rather bold vision, but the moment you look at how it would actually move packets, it starts coming apart.

Networking engineers say the core promise contradicts itself

It can't be backward compatible

The most important claim in the draft is also the one that raised the most eyebrows: "100% backward compatible." The problem is that IPv8 sets the IP version field to 8 and uses 64-bit source and destination addresses instead of IPv4's 32-bit ones, which makes for a longer header. Existing IPv4 routers, switches, and firewalls are built to parse version 4, so a native packet stamped version 8 isn't something they can just transparently pass along, even if the draft tries to route around this. It says IPv8-capable devices should send ordinary IPv4 packets to IPv4-only neighbors, lean on an IPv4 "compatibility mode," and translate at network boundaries through a mechanism it calls XLATE8. Thain put it rather bluntly himself in a LinkedIn comment: "ipv8 sends ipv4 and bury it in the datagram."

That's a workable transition, but it's translation and encapsulation, not native backward compatibility, and it's hard to call it "100% backward compatible" under those conditions. A seamless rollout depending on capability detection, boundary translation, and upgraded edge devices can't really claim that "no existing device, application, or network requires modification." Cloud network architect and Microsoft MVP Simon Painter also pointed out that a lot of networking gear parses packets in a hardcoded manner. The forwarding logic is burned into the chip, and "you can't ship a firmware update that teaches a fixed-function pipeline to parse a longer header." Okay, so complete backwards compatibility is out, but what about the three-layer design?

Well, it turns out that attaching OAuth wasn't exactly popular, either. The first issue was fairly obvious: if you need a token to make a request, how do you make a request to get the token in the first place? Not only that, but putting authentication that deep in the stack also breaks the layering the internet is built on. Tying the routing prefix to an ASN, meanwhile, quietly breaks multihoming, the extremely common practice of a network reaching the internet through more than one provider, which the current system handles precisely because addresses aren't welded to a single autonomous system. The proposed Zone Server may not be a literal single box, since the draft pairs it active/active, but it still concentrates DNS, DHCP, time, authentication, logging, and translation into one tightly coupled control plane.

Then there's the name itself, which several people pointed out was never Thain's to take. In IANA's official registry of IP version numbers, value 8 is already listed as Reserved (Historic), pointing back through RFC 1621 to an early-1990s proposal called Pip. "IPv8" is also the long-standing name of an unrelated peer-to-peer overlay networking library from the Tribler project at Delft University, which has existed for well over a decade and has nothing to do with replacing the internet layer. Picking an already-occupied, slightly cursed version number for a brand-new protocol comes across as if it's being proposed by someone who doesn't know the space.

I'll be honest; none of this means every idea in the draft is worthless. In fact, there were some critics of the draft who actually praised some parts of it. But the contradictions and problematic proposals land right in the center of the overall pitch, and it makes any good suggestion look like an accidental one rather than one that's been well thought out.

Jamie Thain isn't an anonymous nobody

This took a lot of digging

Many people have viewed this debacle as a guy with too much free time and a subscription to ChatGPT, but that's actually an unfair reading of the situation. Jamie Thain, who also goes by James, appears to be a fairly seasoned technology figure with a track record that runs through both Bermuda and Canada. He submitted the draft proposal with a location in Bermuda, under the company name of "One Limited." As well, from what I could find, he appears to be the same Jamie Thain who was named as chief technical officer of Bluewave Bermuda, a fixed-wireless ISP, when it launched in 2017, though no single source directly bridges that role to his current ventures, so it's a strong likelihood rather than a certainty.

Thain's recent history is easier to track: he and his business partner Jack Benaim went through the DMZ startup accelerator at Toronto Metropolitan University. There, they co-founded InnoFund, a Bermuda startup incubator that launched in late 2021. Thain was also chief technology officer of FanDemand, a Toronto company, and was introduced in that role on a Bermuda government 5G advisory panel in 2020. Today he's the founder and CTO of Quadlii, a Canadian startup pitching AI-driven "logic-first, composable" software infrastructure. From everything I can find, this is someone who's been building and running technology companies across two countries for years, with a trail stretching back to at least the mid-2010s.

Unfortunately, what he is not, though, is a network-standards person. IPv8 is his first-ever Internet-Draft. His IETF profile shows no prior drafts and no RFCs, alongside no track record in the standards process at all. That doesn't make him clueless, but as we'll get to, it's hard to figure out where the line is between AI-assisted Thain and just, well, Thain. In one response on LinkedIn, he appeared to reason fluently about real networking concepts, BGP, anycast, 802.1x, ASICs, all the while cheerfully calling himself "an EIGRP bigot." Charitably, you could say that he's someone with hands-on experience and an entrepreneur with self-taught networking depth. However, he's also someone who seems to have never been through the slow, consensus-driven machinery that turns an idea into an internet standard.

His online presence muddies things further, which is part of what makes all of this so difficult to concretely pin on him. There are two LinkedIn accounts. One belongs to a credentialed "James Thain" in Ontario, with a real photo, real connections, and his current Quadlii role. The other is a near-empty "Jamie Thain" listed in Hamilton, Bermuda, with a generic avatar, a handful of connections, and barely any activity, and it's that sparse Bermuda-presenting account the IPv8 writing is bylined to. The threads that tie the Bermuda-based author back to the actual person are numerous, and we'll refer to the credentialed account as "Primary" and the near-empty account as "Secondary" for ease of understanding.

• Primary published a post with a link to the bylined article by Secondary.
• Primary also responded to someone commenting on the link to the article by Secondary as if it were his own creation.
• Secondary, a few months ago, said "This is why we are building Quadlii AI" at the start of a post.

As well, his own crowdfunding page gives us additional clues: The Register confirmed the GoFundMe for IPv8 is his, and it lists him in Oakville, Ontario... not Bermuda. Between the GoFundMe, the Register interview, the IETF author record, and the overlapping Quadlii identity, all signs indicate that the Bermuda-based "Jamie Thain" and the Canadian AI-focused "James Thain" are likely to be the same person.

When it comes to AI, Thain has been rather upfront on his usage of it. He told The Register outright that he'd used chatbots while working on the draft and considers that contemporary practice. For what it's worth, AI-detection tools did flag large chunks of the document as likely machine-written, but so-called "AI detectors" are notoriously unreliable and shouldn't be used as a means to judge the quality of written content. Thain's own admission lines up with everything else about him: Quadlii's entire pitch is that AI tooling can handle what he describes as trillions of dollars of technical debt. Him being someone who believes a large language model can draft a sprawling protocol suite to upend most of the modern web would appear to be entirely on brand.

His view on this comes from an interesting place, though. From what I could find, one of Thain's prior ventures was contracted to build digital, mobile applications for Bermuda government services delivered through a single portal: they were identity-linked, approval-gated systems residents and travellers both used. IPv8's focus on collapsing a sprawl of independently-evolved services into one authenticated system is that same concept scaled from a government portal to the entire internet, and it's achieved by folding DHCP, DNS, time, authentication, and telemetry into one management plane. I couldn't verify that those government systems used the same OAuth/JWT authentication model IPv8 imagines, but there's a pattern of thinking here regardless. Funnily enough, Quadlii makes the same pitch for enterprise software: it's one composable system in place of a mess of bespoke ones. His own framing of the protocol leans hard on identity, pitching the ASN baked into every address as a "caller ID." It's the same unify-and-authenticate approach, just taken to a different target.

Granted, it's certainly a stretch to read too much into all this, but it's a consistent enough pattern that helps explain why this particular person produced this particular over-unified design. Oh, and all of this is without getting into the privacy aspects, as tying every address to an ASN-derived organisational prefix can make it significantly easier to track down individuals on the internet.

Thain has also been open about his motivations for publishing this proposal. Asked on a networking mailing list whether there was a commercial play, he said there were "no patent apps it's all free," and, by his own account, the whole thing started as a downtime project, writing that "recently during a bit of Hiatis [sic], I invented IPv8." He then filed three revisions in four days and launched a crowdfunder with a goal of 100,000 Canadian dollars to build an open-source testbed. Publicly, it looks more like a sincere long-shot than a straightforward grift.

The internet's networking community took it apart

A united front against IPv8

By the normal rules, an individual draft like this just kind of... disappears, but this one didn't. Granted, not all of the interactions this project received were merely dunking on it, but quite a few of them were. The most interesting and useful responses, though, came from the people who took it seriously on purpose.

Jason Rahm was one such person, a senior solutions architect at F5, who wrote a full review titled "IPv8 Would Fix My Routing Tables. It Will Never Ship," and it's actually a good read. Rahm liked a few parts of it, such as the backward compatibility, as flawed as it may be. He said that it was "the one thing this gets right that IPv6 got wrong." As well, the bounded routing table, that /16 floor dragging the global table down from hundreds of thousands of routes toward a couple hundred thousand, addresses "a problem I spent half my career explaining to customers." And the Cost Factor metric, he reckoned, is "the routing metric OSPF and EIGRP always wanted to be."

His verdict, though, was that none of it matters. "IPv6 didn't struggle because it solved the wrong problem," he wrote. "It struggled because it solved only one problem in a system where operators needed several to be solved concurrently." IPv8's habit of solving all of them at once becomes its own death sentence, because "monolithic suites rarely" get adopted, "just ask OSI." And the real roadblock isn't ambition, it's incentives: "IPv8 won't fail because it's too ambitious. It will fail because no one with budget authority is experiencing enough pain to justify replacing the system."

Simon Painter was less generous (but significantly more entertaining), referring to the whole thing as "an April Fool that ran a couple of weeks late," helped along by a companion spec called "WiFi8" that collides head-on with 802.11bn, the actual next generation of Wi-Fi. He still made room to defend IPv6's real-world record, noting it carries around half of Google's traffic these days. What was interesting, though, was that the more thoughtful responses didn't simply treat Thain as an idiot. They recognized that he was treating routing, addressing, management, and security as one connected problem, an instinct some people quietly admired, before trying to solve all of it in a single unilateral push, which none of them thought could possibly work.

A draft like this is easy to file and hard to live up to

Filing a draft looks hard, even if it isn't

Very little stood between Thain and the appearance of authority. Filing an Internet-Draft doesn't need a working-group sponsor, nor does it need prior credentials or even something as simple as a technical endorsement. The format required to file a draft used to be a rough proxy for seriousness, because producing it took a lot of effort and knowledge that acted as a natural filter for casual submissions. It's similar to what's happening to vulnerability research in a sense; using AI to "discover" a vulnerability and draft a write-up overcomes the natural barrier that prevented outright bogus reports.

That filter was hugely important, and the proliferation of AI has been the perfect catalyst to classic Dunning-Kruger traps. The person behind the report is often just knowledgeable enough to ask the AI to build something, but not knowledgeable enough to evaluate the output. What Thain demonstrates isn't as egregious an example as you would find elsewhere, given that he clearly knows some real networking. However, knowing BGP and EIGRP is a long way from understanding why the global routing table, multihoming, fixed-function silicon, and the social economics of migration all conspire to make a "just replace it" protocol impossible, and the polish of the output makes that gap harder to spot on an initial read. Readers can be fooled on a first pass, but stepping away and coming back makes the cracks easier to see. Thain is on the wrong side of that loop permanently, though, because evaluating the output requires exactly the knowledge the AI was standing in for.

His frustration, for what it's worth, is real and widely shared. IPv6 adoption is happening, even if the migration is still painfully incomplete. Thain's -02 revision of IPv8 is set to expire on October 19, 2026, like every unattended draft does, and absent a working group adopting it, that's almost certainly where it ends. The protocol won't fix the internet, but it did show how convincingly one person and a chatbot can now imitate the thing that might.