Linux Foundation Wants to Give AI Agents a DNS Passport, but the Spec Is Still a Draft

Every AI agent that books, buys or signs something on your behalf raises the same awkward question: who exactly is on the other end of the call? On Tuesday, from KubeCon in Salt Lake City, the Linux Foundation put its name behind one answer. It announced that Agent Name Service, or ANS, has launched as a Linux Foundation project, an open standard that uses DNS and the certificates websites already rely on to name, find and verify AI agents.

The pitch is deliberately boring, and that is the point. Rather than a new registry owned by one company, ANS ties an agent to a domain its operator controls. The foundation says the seven-document specification covers naming, registration, DNS publication, monitoring for unauthorized changes and agent-to-agent verification.

"Autonomous AI agents are transacting and operating across global systems, but they can't succeed without verifiable identity." Jim Zemlin, CEO of the Linux Foundation

The supporter list is real, but it is not the whole industry. According to the foundation, 11 organizations have contributed to or adopted the standard since its intent to launch was announced in June: Cisco, Cloudflare, DigiCert, DistributedApps.ai, GoDaddy, HOL, Infoblox, Radix, Salesforce, OWASP and Trustmodel.ai. That list leans hard toward registrars, certificate sellers and network security vendors, companies that already make money from trust on the web.

Look at who steers it, too. Unite.AI reports that the initial steering committee has five voting members, and three of them, Scott Courtney, Keith Perry and Connor Snitker, are from GoDaddy. The other two are Ken Huang of DistributedApps.ai, who says in the release that he led the initial development of ANS, and Michael Kantor of HOL. For a project selling neutral governance, that seating chart is worth watching.

Formalized, or still a draft

The foundation's release says the specification "is now formalized." The project's own site is more modest, stating that "the published specifications are drafts under active development." Unite.AI spotted the same gap, noting that the authentication document, ANS-6, is marked DRAFT v1.0, version 0.2.0, dated November 3, 2026. Launched and unfinished can both be true.

Unite.AI also does the useful work of spelling out what ANS does not do. Finding an agent is not the same as trusting it. In its words, "discovery alone does not authenticate the caller." Authentication, the outlet explains, requires three checks: the certificate is bound to the registered identity, the registration is still valid, and the caller holds the matching private key.

Permissions are another limit. Back in June, the foundation's intent announcement said the framework would let people verify "what permissions it has." The project site now puts it plainly: "ANS establishes identity. Applications still decide what a peer may do." Unite.AI adds that authorization and trust scoring sit outside the authentication draft. ANS can tell you which agent showed up. It cannot tell you whether to let it in.

Early adoption is still thin. The foundation points to two signs: AGNTCY's Agent Directory Service, another Linux Foundation project, supports verification of ANS identities, and a tool called ANS Finder implements the Agentic Resource Discovery specification and passes its registry conformance tests. Unite.AI cautions that the local quickstart uses development keys and a no-op DNS verifier, so passing it proves nothing about a live deployment.

None of this makes ANS the default way agents prove who they are. It is a governed project with a draft spec, working code and a supporter list heavy on companies that sell domains and certificates. Identity is also a separate question from the commerce rules in our look at the shopping bot rulebook from Meta, Walmart and Stripe. If ANS sticks, it will be because deployments actually run all three checks, not because a foundation said the word formalized.