Guide, sources citedLast checked 16 October 2026 with AI-assisted research. Editor: Akshay Dubey.Corrections logEditorial policy
On this page
  1. Why a registry is foundational
  2. What a registry must record
  3. The hard part is discovery
  4. Agents you did not deploy
  5. Keeping it true
  6. FAQ

TL;DR

  • Unique agent identity is the common thread in the MAESTRO threat model published on the CSA blog, the IETF WIMSE draft on AI agent authentication and the NIST NCCoE concept paper. None of them prescribes a registry by name. Treating the registry as the place where that identity is recorded is this site's view.
  • A registry is not a list of names. It records identity, owner, purpose, reach, credential type, expiry and lifecycle state.
  • Storage is trivial; discovery is the problem. The agents that matter are the ones nobody registered.
  • Discover agents where they authenticate, not where they run.
  • A registry that is not reconciled against reality on a schedule becomes a source of false assurance.

Why a registry is foundational

CSA's MAESTRO threat model (published 6 February 2025) lists agent impersonation among its agent ecosystem threats: malicious actors deceiving users or other agents by impersonating legitimate AI agents. It also lists a compromised agent registry as a threat, so a registry is itself something to protect. The IETF WIMSE working group draft on AI agent authentication and authorization (draft-ietf-wimse-aims-00, 15 September 2026, an Internet-Draft and not an RFC, which replaced the individual draft draft-klrc-aiagent-auth) says agents must be uniquely identified, that agent credentials should be short-lived, and that static API keys are an antipattern. It does not define an agent registry. The NIST NCCoE concept paper on software and AI agent identity and authorization (a draft, open for public comment from 5 February to 2 April 2026) asks how identity standards can be applied to agents so that access management systems can distinguish agent identities from human ones. It does not mention a registry.

None of the three prescribes a registry. All three assume that each agent is uniquely identified, and an organisation cannot own, review or revoke an identity it has no record of. That inference is ours, and it is the reason we treat the registry as foundational and not as a tooling preference.

What a registry must record

FieldWhy it is there
Stable identifierSurvives redeployment and version change. Without it, an agent's history restarts at every release.
Accountable ownerA named individual. A team name means nobody is accountable and offboarding never triggers.
PurposeIn one sentence a non-engineer can evaluate. If it cannot be written, the scope is not understood.
ReachWhich systems and data classes it can access, read versus write. This is the risk classification input.
Credential type and expiryAttested workload credential or static secret, and lifetime. Static plus no expiry is the finding.
Delegation modelDoes it act autonomously, for a specific user, or both? Determines what the logs must carry.
Human-in-the-loop conditionsWhich actions require approval. Empty is an answer, and one worth reviewing.
Lifecycle state and review dateProposed, active, suspended, retired. Without state, retirement never happens.
ProvenanceBuilt in-house, vendor-supplied, or embedded in a SaaS product. Determines who can actually change its behaviour.

Nine fields, most of which a team can complete in minutes at registration and almost none of which can be recovered accurately a year later. Capture at creation is worth more than any retrospective exercise.

The hard part is discovery

Any organisation can stand up a registry. It will become incomplete, because agents are created by people solving immediate problems and registration is a step they can skip.

The reliable approach is to discover agents where they authenticate rather than where they run:

  • OAuth clients registered against your authorization server, particularly ones created for a proof of concept.
  • Third-party app grants in SaaS admin consoles. Created by business users through consent screens and easy to miss in a security inventory.
  • API gateway consumers that do not map to a known service.
  • Model provider API keys and their billing records. Spend is a useful agent detector: finance may know about agents that security does not.
  • MCP servers reachable from your environment. See MCP security.
  • Egress to model endpoints from workloads with no declared AI purpose.

Every one of these is a control point you already operate. None requires a new product.

Agents you did not deploy

Three categories are easy to leave out of a registry:

  1. Vendor agents inside SaaS products. Your CRM ships an assistant that reads customer records. You did not deploy it, you cannot inspect it, and it operates under a grant somebody accepted. The Salesloft/Drift incident shows the related risk of a third-party integration holding OAuth tokens to customer data.
  2. Developer-adopted tooling. Coding assistants and local MCP servers with repository access, adopted individually and invisible to procurement.
  3. Embedded automation. Workflow tools whose AI steps were enabled inside an existing integration, so no new grant appears.

None can be prevented by policy alone. Registering what you find, with reach and provenance recorded, is more achievable than preventing adoption and produces better decisions.

Keeping it true

A registry decays quickly because agents are cheap to create and invisible when idle. Three practices keep it honest:

  • Reconcile against discovery on a schedule. Compare registered agents to authenticating agents and treat both directions as findings: unregistered agents, and registered agents that have not authenticated in ninety days.
  • Expire registrations, not just credentials. A registration that requires renewal forces someone to confirm the agent should still exist.
  • Publish the delta, not the total. "Fourteen agents registered" tells a governance forum nothing. "Six new, two retired, three unregistered found in discovery" is a decision-grade report.

Frequently asked questions

What is an AI agent registry?

An authoritative record of which agents exist in an organisation, what each is permitted to reach, who is accountable for it, and what lifecycle state it is in. It is the agent equivalent of a non-human identity inventory, with additional fields for delegation model and human-in-the-loop conditions. None of CSA MAESTRO, the IETF WIMSE draft on AI agent authentication or the NIST NCCoE concept paper prescribes a registry by name, but each assumes agents are uniquely identifiable.

What should an agent registry record?

Nine fields carry most of the value: a stable identifier surviving redeployment, a named individual owner, a one-sentence purpose, the systems and data classes it can reach with read versus write, credential type and expiry, delegation model, human-in-the-loop conditions, lifecycle state with a review date, and provenance indicating whether it was built in-house, supplied by a vendor, or embedded in a SaaS product.

How do you discover AI agents nobody registered?

Look where they authenticate rather than where they run: OAuth clients registered against your authorization server, third-party app grants in SaaS admin consoles, API gateway consumers that map to no known service, model provider API keys and their billing records, reachable MCP servers, and network egress to model endpoints from workloads with no declared AI purpose. Each of these is an existing control point, not a new product.

What are shadow agents?

Agents operating in your environment that are not in any inventory. Three categories to look for: vendor agents embedded in SaaS products you did not deploy and cannot inspect, developer-adopted tooling such as coding assistants and local MCP servers with repository access, and AI steps enabled inside existing workflow integrations so that no new grant appears. Policy alone does not prevent any of them, which is why discovery matters more than prohibition.

How often should an agent registry be reconciled?

At least quarterly, and monthly where agents have production write access. Reconcile in both directions: agents authenticating that are not registered, and registered agents that have not authenticated in ninety days. Report the delta rather than the total, because a count of registered agents tells a governance forum nothing about whether the registry is true.

Should an AI agent be registered as a non-human identity?

Yes, in this site's taxonomy an agent's identity is a non-human identity, so it belongs in the same inventory with extra fields: stable identifier, named owner, purpose, reach, credential type and expiry, delegation model and human-in-the-loop conditions. The IETF working-group draft draft-ietf-wimse-aims says agents MUST be uniquely identified; a registry is where that identification becomes checkable.

Is there a standard for registering or discovering AI agents?

Not a ratified one in our tracking on 16 October 2026. The A2A specification lets agents publish Agent Cards, which can be signed, describing skills and security schemes; that supports discovery between agents but is not an enterprise inventory. CSA's August 2025 paper on agentic IAM proposes an Agent Name Service. Neither tells you which agents your organisation runs, so the registry remains your own control.

Sources

From the publisher

HumanAudit, which publishes this site, is preparing the Agent Governance pack: an agent policy, inventory and owner register, delegated-authority standard, and kill-switch and revocation procedure. Planned for the first week of November 2026; not yet on sale, and no price has been published.

Monthly change log

Know when the NHI landscape changes.

One email a month from this reference: new deals, standards status changes and corrections. No sponsored content. Unsubscribe from any email.

We first email you a link to confirm. Privacy policy.