Domain-anchored agent identities, built from parts the internet already runs
The problem ANS addresses is concrete. An agent calling another agent over a protocol such as MCP or A2A learns a hostname and an endpoint URL. It does not learn, in any interoperable way, who answers for that endpoint, which registered version of the agent it is dealing with, or whether that version's registration history can be audited. A TLS certificate names the hostname. It says nothing about the registered release.
ANS v2, as set out in its IETF Internet-Draft, answers with infrastructure that already exists. The name is anchored in a DNS domain the operator controls; the design builds on DNSSEC and the Web PKI. Certificates authenticate the endpoint, and a Registration Authority checks domain control and can issue an Identity Certificate, when the registrant requests one, that binds one versioned name to one key. Registrations, renewals and revocations are recorded as lifecycle events in an append-only transparency log that anyone can audit. A new version of an agent is a new registration, so a verifier can check trust against a specific registered release rather than a name that silently changed underneath it.
One distinction matters here. ANS lets a verifier check a registered, versioned identity and its associated key and registration history. That does not by itself attest to the code executing behind the endpoint. A registered version is a claim on record, not a runtime guarantee.
That design choice is the strongest part of the launch. ANS builds on existing DNS and certificate infrastructure, alongside its registration and transparency-log mechanisms, instead of asking the world to adopt a proprietary registry. The foundation's supporter list reflects where the pieces come from: registrars, certificate authorities and DNS operators (GoDaddy, DigiCert, Infoblox, Radix, Cloudflare) sit alongside Cisco, Salesforce and OWASP.
Launched, and still a draft
The foundation's release says the specification "is now formalized." Directly under its introduction, the project's own homepage, ansinfo.ai, states: "The published specifications are drafts under active development." The release calls the specification "formalized," while the project site calls the published specifications drafts under active development. These statements establish a project launch without establishing a final, stable specification: the release also says the project's technical steering committee will govern the specification from here, and the IETF version is an Internet-Draft, a work in progress that expires on April 29, 2027, whose own boilerplate says such drafts should be cited only as work in progress. Builders should plan against the second statement. An identity layer whose authentication document is still moving is a direction to align with, not a finished interface to hard-code against.
The reference implementation narrows the claim further, usefully. The project offers a runnable Registration Authority and Transparency Log that can be run locally, with an offline verifier. That is a way to test the lifecycle. It is not, by itself, evidence of a live public registration ecosystem operating at internet scale, and the launch announcement does not publish a count of registered agents in production.
Identity is the first question, not the answer
The project states its own boundary in one sentence: "ANS establishes identity. Applications still decide what a peer may do."
The IETF draft makes the same point at length. ANS records registrations and their lifecycle, not the actions an agent takes. It does not bind the keys an agent uses to sign payment authorizations. Trust scoring is left to separate services that read ANS records. So an ANS-verified agent is an agent whose operator, domain, registered version and registration history you can check. It is not an agent that has been judged safe, honest or competent, and a verified name should never be presented to users as a safety mark.
That distinction decides how much weight this launch carries. If agent traffic grows across company boundaries, a shared identity mechanism could support cross-organization authorization, auditing and accountability. ANS is a credible candidate for that layer. It launched this week as a governed project with draft specifications and two early integrations. That is a real start, and it is also all it is.
Sources
- [1] Linux Foundation via PR Newswire, "Linux Foundation Welcomes Agent Name Service to Bring Verifiable Identity and Discovery to AI Agents" (primary announcement, November 10, 2026)Read source
- [2] Agent Name Service project site, ansinfo.ai (primary; specification status and scope wording checked November 11, 2026)Read source
- [3] Courtney et al., "Agent Name Service v2 (ANS): A Domain-Anchored Identity Layer for Autonomous AI Agents," IETF Internet-Draft draft-narajala-courtney-ansv2-02 (primary specification document, October 26, 2026; expires April 29, 2027)Read source
- [4] ANS open-source organization (specifications, reference registry, transparency log, verifier, SDKs and CLI, linked from the foundation announcement)Read source