Independent analysis by NHI Mgmt GroupSource article: Oak, “How to Inventory the AI Agents Running in Your Enterprise” (November 11, 2026)

TL;DR: Oak argues that AI agent inventories work only when each agent is treated as an identity with an accountable owner, effective permissions, tool usage and deployment scope, because spreadsheet-based registration misses shadow agents and duplicated deployments. Identity review must shift from static asset tracking to continuous governance or agent risk stays invisible.


At a glance

What this is: NHI Mgmt Group's analysis, drawing on Oak's blog post: AI agent inventories become governable when discovery is continuous and each agent is recorded as an identity with owner, permissions, tool use and deployment scope.

Why it matters: IAM and security teams need agent inventories that survive duplication, shadow deployment and ownership drift, because without identity-level records they cannot certify access or assign accountability.

👉 Read Oak's guide to inventorying AI agents as identities


Context

AI agent inventory breaks when teams treat agents as assets or model instances instead of identities. The missing control is not just discovery, but a governed record that ties each agent to an accountable owner, the human or team it acts for, and the effective permissions it can actually exercise.

Discovery has to begin at the identity surfaces where agents already authenticate, including OAuth grants, service principals, MCP servers and repository keys. A spreadsheet cannot keep pace with agents created faster than governance steps, so the inventory has to follow access rather than creation paperwork.


Key questions

Q: How do teams discover unmanaged AI agents across the enterprise?

A: Teams discover unmanaged AI agents by combining cloud and application inventory, API gateway logs, IAM and PAM records, endpoint telemetry, and software asset discovery to find runtime processes, service accounts, tokens, and orchestration tools that act autonomously. A complete inventory should also map each agent to the human owner, the data and tools it can reach, and any secrets or delegated permissions it uses.

Q: Why does one AI agent deployed in several places still need one record?

A: Because the agent is one identity with one accountable owner, one policy set, and one revocation point, even if it runs in multiple hosts or clouds. A single record ties its runtime permissions, tool access, secrets, and audit trail together so access reviews, rotation, and shutdown work consistently. Without that record, scope drifts and orphaned credentials remain active.

Q: Why is defining permissions important for AI agents?

A: Defining permissions is crucial for AI agents to minimize the risk of unauthorized data access. Without clear permissions, agents can inadvertently interact with sensitive data, leading to potential security breaches and compliance issues.


Technical breakdown

Why AI agent inventories fail when deployments are counted separately

A single AI agent can exist across several tenants, cloud accounts and tool endpoints while still being one identity. Counting each deployment as a separate row understates the real blast radius and fragments ownership. The operational mistake is to inventory placements rather than the identity that spans them. When those rows are not correlated, no reviewer sees the full set of permissions or the combined reach of the agent.

Practical implication: Correlate duplicate deployments into one identity record before certification or access review, so ownership and effective reach are assessed as a single governance object.

What belongs in an AI agent identity record

A usable agent record needs more than a name and a system tag. It needs the accountable human owner, the subject the agent acts on behalf of, effective permissions after inheritance and delegation, the tools actually used, and every deployment location. That structure separates responsibility from implementation detail and makes the record answerable during review. Without those fields, the inventory becomes a catalog of labels rather than a governance control. See the Agentic AI Identity Guide.

Practical implication: Define a minimum agent record schema that captures ownership, delegated scope, actual access and runtime footprint before allowing the agent into production.

Why continuous discovery matters more than a one-time register

An agent inventory decays as soon as new tools are granted, old owners leave or shadow agents appear outside formal onboarding. Continuous discovery treats the inventory as a living view of the identity layer, not a project deliverable. That matters because the risk changes with access drift, not with the calendar. The inventory should therefore be refreshed whenever agent identity surfaces change, not only at audit time. See the Shadow AI and AI Agent Discovery Guide.

Practical implication: Move from periodic spreadsheets to event-driven discovery and review cycles tied to grants, deployments and ownership changes.


Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.


NHI Mgmt Group analysis

AI agent inventory is an identity governance problem, not an asset management problem. Oak's core point is that agents become governable only when they are recorded as identities with owners, delegated purpose and effective permissions. That framing matters because the inventory has to answer who is responsible and what the agent can reach, not merely what was deployed. Practitioners should treat the inventory as part of access governance, not a separate AI catalog.

Duplicate deployments create identity blind spots. One agent deployed in several places is still one identity, but separate rows make its total reach look smaller than it is. That is a governance failure because reviewers certify fragments instead of the full access footprint. The implication is that correlation across tenants, accounts and servers is a prerequisite to meaningful review.

Shadow agents are a discovery problem first and an enforcement problem second. Agents running on a builder's credentials inherit whatever that builder can access, which makes unregistered agents especially hard to contain. The useful concept here is identity correlation debt: the longer an agent exists without a linked owner and scoped record, the harder it becomes to determine who answers for it and what it can do. Governance teams should prioritise discovery at the point of authentication, not at the point of form submission.

Continuous review is the only cadence that matches agent change rate. Static registers age out as soon as grants, deployments or owners change. The practical lesson for IAM and IGA teams is that AI agent governance has to move to event-driven review triggers and usage-based reassessment, or the inventory will reflect yesterday's access rather than today's reality.

From our research library:

What this signals

Identity correlation debt: governance breaks when the same agent is allowed to exist as several partial records. As deployments multiply, the inventory starts to understate total reach and overstate how much accountability the programme actually has. Security teams should expect correlation, not enumeration, to become the real control point for agent oversight.

AI agent inventories will increasingly sit between discovery tooling and certification workflows. That means IAM and IGA programmes will need a control that can absorb frequent change without turning every update into manual overhead, especially as NIST SP 800-53 Rev 5 Security and Privacy Controls become a stronger reference point for access governance.


For practitioners

  • Establish a single accountable owner per agent Require every AI agent record to name one accountable human owner, even when the agent serves a wider team or workflow. Shared mailboxes and generic team aliases should not carry ownership.
  • Correlate duplicate deployments into one identity Merge rows for the same agent across SaaS tenants, cloud accounts and MCP servers so reviewers see total reach rather than partial views. Use deployment correlation as a certification prerequisite. See the AI Agent Observability, Audit and Incident Response Guide.
  • Inventory from identity surfaces, not forms Start discovery at OAuth grants, service principals, repository keys and MCP servers so shadow agents appear even when no one registered them. Registration should confirm what discovery already found.
  • Record effective permissions and actual tool use Capture the permissions an agent can exercise after inheritance and the tools it actually invokes, then compare those two sets for excess access. Tool usage gaps often reveal overreach that a static grant review misses.
  • Trigger review on ownership and access changes Re-review an agent when a new tool grant appears, a deployment is added or the owner leaves. Continuous discovery should feed event-driven micro-certifications instead of waiting for a fixed annual cycle.

Key takeaways

  • AI agent inventories only work when each agent is governed as an identity with an owner, delegated purpose and effective access.
  • Counting deployments separately hides the real reach of an agent and weakens accountability across tenants and tools.
  • Continuous discovery and event-driven review are the controls that keep agent inventories current enough to govern.

Related NHI Mgmt Group resources: Agentic AI Identity Guide · Shadow AI and AI Agent Discovery Guide · AI Agent Authorisation Guide · AI Agent Observability, Audit and Incident Response Guide


Standards & Framework Alignment

This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.

OWASP Agentic AI Top 10 and OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST AI RMF, NIST CSF 2.0 and CIS Controls v8 set the governance and control requirements practitioners need to meet.

FrameworkControl / ReferenceRelevance
OWASP Agentic AI Top 10ASI03 — Identity & Privilege AbuseAI agents inherit and exercise privileges across deployments, making privilege abuse central to this inventory problem.
Recommendation — Map agent records to privilege boundaries and verify delegated access against intended scope.
NIST AI RMFGOVERN — AI Governance and AccountabilityThe article is about accountable AI inventory and review processes, which sit in AI governance.
Recommendation — Establish governance for AI agent inventory ownership, review cadence and accountability.
OWASP Non-Human Identity Top 10NHI-05 — Overprivileged NHIAI agents are non-human identities whose effective permissions can exceed their intended task scope.
Recommendation — Review AI agent permissions against task scope and remove excess access that the record exposes.
NIST CSF 2.0PR.AA-05 — Access Permissions, Entitlements and AuthorizationsAgent inventories are about knowing and governing actual entitlements across systems and deployments.
Recommendation — Document and review each agent's entitlements as part of access governance.
CIS Controls v8CIS-5 — Account ManagementThe post focuses on owning, tracking and retiring agent identities as accounts in use across environments.
Recommendation — Manage AI agents as accounts with named ownership, scope and retirement tracking.

Key terms

  • AI Agent Identity: The digital identity used by an autonomous AI agent to authenticate to external systems, APIs, and services. Managing AI agent identities is an emerging and rapidly evolving area of NHI security.
  • Shadow AI: AI agents, copilots, or connected tools operating without full visibility or governance from security teams. Shadow AI becomes an identity problem when those systems authenticate with unmanaged tokens, service accounts, or OAuth apps that can reach production resources.
  • Effective Permissions: Effective permissions are the access an identity can actually use after role inheritance, scope, and policy are applied. In Azure AI environments, they often matter more than the assigned role name because inherited rights can widen access to data, logs, and secret stores.
  • Event-Driven Certification: Event-driven certification is a review model that triggers access validation when an identity changes state, such as a role move, onboarding, or termination. It reduces reliance on fixed review cycles and gives IAM teams a better chance to act while access is still relevant.

What's in the full article

Oak's full blog post covers the operational detail this post intentionally leaves for the source:

  • The step-by-step discovery workflow for finding AI agents across SaaS, cloud, repositories and MCP servers
  • The inventory row schema used to capture owners, effective permissions, tool usage and deployment locations
  • The handling of duplicate deployments when one agent appears across multiple environments
  • The review and certification pattern for keeping agent records current as grants and owners change

👉 Oak's full post covers discovery surfaces, record fields and continuous review for AI agents.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.
NHIMG Editorial Note
Published by the NHIMG editorial team on November 16, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org