Infoblox published guidance this week framing AI agent security as a two-layer DNS problem: one layer for finding agents, another for blocking the ones you don't trust. The company's DNS-AID standard, now advancing as an IETF draft alongside an open-source reference implementation, lets an organization publish an agent's identity directly under its own domain, and the recommended pattern uses existing record types to make that identity checkable rather than asserted. The core move is anchoring discovery to DNSSEC. Infoblox's guidance calls DNS-AID a form of "Positive Control," describing it as a way to name the agent, validate the operator's domain, and verify endpoint provenance via DNSSEC. Paired with what Infoblox calls "Protective DNS" as a "Negative Control" — restricting outbound traffic to unauthorized or malicious domains before a connection is even attempted — the pitch is that discovery and blocking need to work together: you can't police what you can't uniquely name, and naming alone doesn't stop bad actors from squatting on lookalike identities. What's notable isn't the cryptography — DNSSEC and DANE are decades old — it's the reuse. Rather than invent a new PKI or agent-specific trust root, DNS-AID uses existing DNS record types, including SVCB service bindings, DNS-SD, DNSSEC, and DANE, to let agents publish discoverable metadata. Infoblox's own framing leans on that lineage explicitly, with its legal chief describing the goal as avoiding the mistake of building another centralized directory and instead using an open, federated protocol anyone can join, the same argument DNS itself won decades ago against centralized name lists. The company recommends organizations start now: enforce DNSSEC signing on every DNS record tied to an agent, publish agents under a real enterprise subdomain like billing-agent.example.com so provenance is unambiguous, and configure internal DNS policy to log and isolate unapproved agent traffic. None of this requires waiting on IETF ratification — the repository is open-source today. The catch is adoption asymmetry. DNS-AID only tells you something true about agents whose operators bothered to publish signed records. It does nothing for the agent that never registers, or the one squatting on a domain that never enabled DNSSEC in the first place — which, given how much of the web still doesn't sign its zones, may be most of them for a while. The standard makes verification cheap for good actors; it doesn't yet make non-participation costly for bad ones.