Skip to content

DNS and Ai agents: What Enterprises Need to Prepare For

DNS can support AI agents without a fundamental protocol redesign, but enterprises need resilient resolvers, query visibility and enforceable policies. DNS security should complement application, identity and database controls rather than replace them.

November 17, 2026 | Written by: Paul Vixie | ,

Summary

Dns and Ai Agents What Enterprises Need to Prepare For Featuring Andreas Taudte and Paul Vixie

Key Takeaways

  • AI agents, by their complexity and demands, increase the importance of DNS reliability, visibility and policy enforcement. Actual capacity needs depend on their workloads.
  • Useful protection combines, as before, query context, threat intelligence and enforceable controls at the resolver.
  • DNS contributes to AI agent security alongside authentication and authorization at the application, tool and database layers.

What stood out from the conversation

The DNS protocol has demonstrated its ability to accommodate successive generations of applications. AI agents are another kind of client, although their discovery patterns and automation can change the workload. An individual server still needs the capacity, headroom, and configuration to handle changing demands.

Questions about DNS proficiency produced an equally useful distinction. There is substantial expertise among specialists, coupled with a loss of basic understanding among people who can now consume DNS through services and APIs. The practical takeaway is to match knowledge to responsibility: administrators need to understand resolution, forwarding, caching and failure behavior well enough to diagnose problems even when the management interface looks healthy.

The discussion also covered organizational ownership and the difficulty of introducing DNS security when security teams wanted protection but DNS teams saw another operational responsibility. Effective policy can only result when both teams agree on enforcement, exceptions and troubleshooting.

Q&A Highlights

Does AI require new DNS record types?

AI agents can use existing DNS records to reach services. Discovery proposals address a more specific problem: publishing where agents are available, which protocols they support and how clients find their metadata.

Two approaches illustrate the discussion. The DNS-AID Internet-Draft uses structured names and existing Service Binding (SVCB) records. The DNS-Based Agent Naming proposal introduces AIDISCA and AIINDEX record types. Both are proposals, rather than adopted IETF standards.

One question raised was whether introducing new record types could keep pace with agent development. For operators, the immediate task is to understand the approach their applications actually implement, including publication, caching and validation requirements. Discovering an agent does not by itself establish permission to use its tools.

How can organizations protect agents when request volumes are unpredictable?

First and foremost, give agent workloads an approved resolution path and define the destinations they need. Establish baselines for query rates, failures and destination diversity, then investigate deviations in the context of the workload.

Next, control retry loops, concurrency and execution budgets in the agent platform as well. DNS-side limits should be tested against legitimate bursts so that protective measures do not create avoidable outages.

Finally, measure resolver latency, cache behavior and connection capacity during representative workloads. Cached answers and reused connections mean DNS query counts cannot serve as a complete count of agent actions.

How can teams distinguish malicious domains from legitimate AI traffic?

A domain generation algorithm (DGA) generates candidate domain names algorithmically. Malware can use those names to locate command-and-control infrastructure. Neither an unusual name nor a high query rate establishes malicious intent by itself. Many DGA systems generate high-entropy names which don’t look like natural language words or abbreviations and may be detectable that way, but on the flip side many country-code top-level domains have wildcards which suppress negative answers (RCODE NXDOMAIN) that can otherwise help with DGA detection. This is a cat-and-mouse game having an infinite number of turns.

In response to the question about cheap, disposable domains, the discussion emphasized local protection through Response Policy Zones (RPZ), which express policies that a recursive resolver applies to DNS responses. Local enforcement gives defenders a response without waiting for a registrar or hosting provider to act.

The discussion also covered newly observed domains: names first detected by a particular observation system, which are not necessarily newly registered. Andreas described activity where queries returned NXDOMAIN before supporting infrastructure appeared. A name’s registration (WHOIS) age alone is therefore an insufficient trust signal.

The broader opportunity here is to correlate domain history, query patterns, threat intelligence and workload identity with endpoint or application evidence. DNS Guardian supports DNS traffic inspection and behavioral analysis as part of that investigation.

Can EfficientIP protect database schemas from rogue agents?

DNS controls can restrict resolution of database and service names. DNS Client Query Filtering supports granular policies at that layer, but schema, table and operation permissions must be enforced by the database, application or tool service.

Each agent can be given an appropriately scoped credential, allowing us to separate read access from write access and enforce authorization on every operation. For Model Context Protocol (MCP) integrations, follow the protocol’s security guidance on scope minimization and token handling.

An agent reaching an authorized endpoint must still be prevented from performing unauthorized actions there. DNS cannot repair a bypass of application or database role-based access control (RBAC).

Can encrypted DNS coexist with enterprise and parental controls?

Perhaps. DNS over HTTPS (DoH) transports DNS queries and responses over an encrypted HTTPS connection. The policy question is which resolver receives those requests.

An approved encrypted resolver can apply local policy. Applications using an external resolver can bypass that policy and remove queries from the organization’s DNS visibility. Such bypasses highlight the limits of approaches that depend entirely on client cooperation.

We must configure managed operating systems and browsers to use approved services, and combine this with appropriate endpoint and outbound connection controls. Blocking a list of known public DoH services cannot cover every private endpoint. The discussion also covered research on controlling protocols over encrypted channels. A related patent is US 11,438,309.

What does the Quad9 discussion show about DNS control?

It shows why resolver selection has operational and governance consequences. In its October 22, 2026 announcement, Quad9 scheduled its withdrawal from serving users in France for November 24, 2026, following a court ruling requiring it to perform domain blocking.

The discussion acknowledged both national sovereignty and a provider’s ability to choose where it operates, while questioning approaches that create obvious workarounds. Andreas’s concern was whether the resolver is always the appropriate place to address harmful or unlawful content. Paul noted that motivated citizens can choose to use upstream DNS servers in other jurisdictions, or to operate their own DNS server (such as Pi-hole with Unbound).

For enterprises, the practical lesson is to understand the policies, jurisdictions and continuity arrangements attached to their chosen resolution services.

Conclusion

The main takeaway from the conversation is that AI makes familiar DNS responsibilities more consequential. We must each review where agents resolve names, what we can observe, how policies are enforced and who owns failures. Then we must test those controls alongside the authorization boundaries protecting the tools and data agents can access. Confidence is achievable but not without effort.

FAQ

Watch the full conversation on DNS and agentic AI

Paul Vixie and Andreas Taudte discuss how DNS must evolve for agentic AI, covering encrypted resolution, visibility, security and operational readiness.

Talk to an expert