The Internet Is Getting Standards for AI Agents. DAWN Is What That Looks Like.

The internet has standards. HTTP. DNS. TCP/IP. The protocols that make the web work were built by the IETF — the Internet Engineering Task Force — through a slow, deliberate process of published requirements, debate, and iteration. This week the IETF published the first requirements for AI agents: DAWN — Discovery of Agents, Workloads, and Named Entities.

This is early. DAWN is a requirements document, not a finished protocol. But the fact that the IETF is working on it at all is significant.

What DAWN Is Trying to Solve

Right now, AI agents are islands. A Claude agent and a GPT agent and a Gemini agent cannot reliably discover each other, verify each other’s identity, or establish trusted communication channels. Every multi-agent system is a custom engineering project built on top of unstandardized protocols.

DAWN is attempting to define how agents announce their existence, what they can do, how they authenticate, and how they route requests to each other. Think of it as DNS for agents — a way to find and verify who you’re talking to before you trust them with anything.

Why Standards Matter for AI Safety

One of the underappreciated AI security risks is that agents currently have no reliable way to verify they’re talking to who they think they’re talking to. Prompt injection attacks, adversarial agent impersonation, and supply chain attacks on agent infrastructure all exploit the absence of authentication standards.

DAWN won’t solve all of these problems. But standardized identity and discovery is a necessary foundation for solving any of them.

The Buccaneer Take

The internet we have today was built on standards that nobody thought would matter much when they were first proposed. DAWN feels early and academic right now. In five years it may be infrastructure. Watch it. 🏴‍☠️