The Linux Foundation Wants DNS to Be the Phone Book for AI Agents
DNS-AID proposes using existing DNS infrastructure for agent discovery — no new directories, no vendor lock-in, no new protocols to trust.
Agent sprawl is a real problem. As autonomous AI systems multiply and begin talking to each other, the question of how they find each other has quietly become load-bearing.
The Linux Foundation is backing an answer: DNS-AID, a draft standard that uses the Domain Name System — the same infrastructure resolving every hostname you've ever typed — as a global, vendor-neutral registry for AI agents and Model Context Protocol (MCP) servers.
The core mechanic is straightforward. Domain owners publish a well-known DNS record at _index._agents.{domain}. An agent looking for counterparts queries that subdomain and gets a starting point. No proprietary registry. No new trust anchor. Just DNS, which the internet has relied on since RFC 1034.
That's either elegant or optimistic, depending on your threat model.
DNS was not designed with agent authentication in mind. The system tells you where something is; it doesn't prove what that something is. The distinction between discovery (finding) and verification (trusting) matters enormously when autonomous agents are executing actions on behalf of users. DNSSEC can sign records and prevent tampering, but it's still widely under-deployed, and it doesn't solve the problem of a legitimately registered domain hosting a malicious agent.
The proposal acknowledges verification as an open question. Several details remain to be worked out, per the Linux Foundation's own framing.
DNS-AID originated at Infoblox. The current internet draft includes contributions from Deutsche Telekom and Amazon staff, which gives it some cross-organizational weight. The Linux Foundation is now soliciting broader contributions and has committed to keeping the standard vendor-neutral — a credible promise given the Foundation's track record, though one worth watching as larger cloud providers deepen their involvement.
Jim Zemlin, CEO of the Linux Foundation, framed the stakes plainly: without secure, open discovery infrastructure, agent-to-agent connectivity becomes a liability rather than a feature.
He's not wrong. Proprietary agent registries already exist, and a fragmented ecosystem of siloed directories is a predictably bad outcome. Standardizing on DNS at least anchors discovery in infrastructure with decades of operational experience, distributed ownership, and existing tooling.
The harder work — defining how agents authenticate to each other after discovery, what scopes they advertise, how consent flows between them — sits downstream of DNS-AID and remains largely unsolved across the industry. Discovery is step one. It's a necessary step. But solving the phone-book problem doesn't automatically solve the identity problem, and in agentic systems, conflating the two could get expensive fast.
