Compliance Guide

NIST AI RMF GOVERN 1.6 Requirements for an AI System Inventory

GOVERN 1.6 is the NIST AI RMF subcategory that addresses maintaining an AI system inventory. It is a brief outcome statement and does not prescribe inventory fields, scope, ownership, or update processes, so each organization must define them.

What GOVERN 1.6 states and what it leaves open

Because the subcategory leaves scope, fields, ownership, and update processes undefined, the organization has to set them. A defensible inventory covers internally built, third-party, and agentic systems, records risk-relevant attributes, assigns named owners and update triggers, and produces evidence that it is actually maintained.

From outcome statement to working inventory
AreaWhat to define
ScopeDecide which internal, third-party, and agentic systems count.
Data elementsRecord the attributes that support risk-based oversight.
Ownership and triggersName accountable owners and define what forces an update.
EvidenceRetain records showing the inventory is complete and current.

Define inventory scope first

An inventory needs a clear boundary before it needs fields. Decide which internal, third-party, and agentic systems count, so that internally built systems, vendor-provided systems, and agents are all covered by the same inventory.

Data elements to record

Record the attributes that support risk-based oversight. For each system, capture the following.

  • Purpose and intended use, including affected users and the decisions the system informs.
  • Accountable business owner and technical owner.
  • Models used, their provenance, and the data sources they draw on.
  • Third-party dependencies, such as model providers, hosted services, and external tools.
  • Risk tier and deployment status, from proposed through retired.
  • For agents: identity, granted permissions, connected tools, and data access.

Assign ownership and update triggers

Name accountable owners for each inventory entry and define what forces an update. Without named owners and explicit triggers, an inventory is a one-time snapshot rather than a maintained record.

Evidence an assessor would reasonably expect

Retain records showing the inventory is complete and current. The goal is to be able to demonstrate that the inventory is actually maintained, not only that it exists.

Where automated and runtime-derived data helps

Automated and runtime-derived data can support the agent entries in the inventory, which call for identity, granted permissions, connected tools, and data access.

Keep the agent portion of your AI inventory current

See how runtime governance can supply agent identity, permissions, tool-call activity, and audit logs to support your inventory process.

Explore Runtime Governance