Securing the agentic frontier: Using DNS AID as a trust foundation for autonomous AI

As autonomous and multi-agent AI systems move from experiments into production, a new question is emerging: how do we let agents safely find and trust each other across organisational and national boundaries? Without a neutral, open discovery layer, the "agentic web" risks inheriting the worst of today’s internet—centralization, opaque control, and fragile trust—at far greater speed and scale.

This talk argues that DNS-AID (DNS-based Agent Identification and Discovery), built on the existing global Domain Name System (DNS) and its security extensions, is a practical candidate for a decentralized root of trust for autonomous AI. Rather than creating yet another proprietary registry, DNS-AID anchors agent identities and discovery in domains enterprises and governments already own, and in security controls they already operate.

We will:

• Show why agent discovery and identity have become first-class security and governance issues, using concrete risk scenarios such as agent impersonation, registry poisoning, and cross-domain abuse.
• Introduce the DNS-AID pattern: a lightweight way to publish where agents live, which protocols they speak, and where to find their policies and capabilities, using familiar DNS record types and DNSSEC/DANE for integrity.
• Position DNS-AID alongside emerging proposals such as Agent Name Service (ANS), AgentDNS, DNS-Native Agent Naming and Resolution, and DID/VC-based identity, explaining how a DNS-anchored layer can give them a common, interoperable substrate.
• Outline a pragmatic roadmap toward a vendor-neutral IETF standard that aligns with zero-trust architectures, Protective DNS initiatives, and national AI assurance frameworks.

Attendees—security leaders, architects, DNS operators and policymakers—will leave with a clear threat model for agent discovery, an actionable reference pattern they can pilot today, and a view of how near-term standards decisions will determine whether autonomous agents remain open and interoperable, or are mediated by a handful of private discovery providers.