AI model and agent inventory is now the first governance test

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Collibra's blog post: AI governance starts with a live inventory of models, use cases and autonomous agents, because ownership, risk and data access cannot be governed reliably without it.

Why it matters: For IAM and governance teams, the practical issue is traceability: if you cannot inventory an AI asset at the point of deployment, you cannot assign accountability, review access or prove control coverage later.

Context

AI model and agent inventory is the record that tells an organisation what AI systems it runs, who owns them, what risk they carry and what data they can reach. In governance terms, that inventory is the control surface that makes AI oversight auditable rather than implied.

The article's core point is that spreadsheets and periodic reviews fail once AI estates start changing faster than manual tracking can absorb. For AI governance and identity teams, the problem is not just visibility in theory, but keeping the record current enough to support accountability, access scope and lifecycle decisions.

Key questions

Q: What breaks when AI systems are deployed without a complete inventory?

A: Governance breaks first, then auditability. If teams cannot identify every model, dependency, and owner, they cannot assign risk tiers, apply the right controls, or prove compliance when regulators ask. The result is shadow AI: systems that process data and influence decisions outside the governance perimeter, often for months.

Q: When should organisations prioritise inventory over broader AI governance controls?

A: Inventory should come first because every downstream control depends on knowing what exists. Risk classification, assessments, access review and lifecycle management all assume the asset is already identified. If the record is missing or stale, the organisation is governing a partial estate and cannot trust the rest of the programme.

Q: What are the signs that AI inventory governance is failing?

A: Common signs include duplicated tools, unknown model owners, incomplete lineage, inconsistent risk ratings, and AI deployments that appear in production before they are formally registered. Those symptoms show that discovery, approval, and governance workflows are no longer aligned.

Q: How should teams govern AI agents without treating the model as the only control point?

A: Governance should follow the full runtime, not just the model checkpoint. Teams need approval gates for tool use, version control for prompts and procedures, isolation for sub-agents, and traceability for every governed run. If the harness changes, policy changes, even when the model does not.

Technical breakdown

Why a live AI inventory matters more than a spreadsheet

A spreadsheet inventory is a snapshot, not a control. It goes stale as soon as new models, use cases or agents are deployed, which means the record no longer matches the runtime estate. In practice, that breaks the link between governance intent and operational reality. A live inventory captured at deployment becomes the source of truth for ownership, risk tier, assessment status and access scope. That is what makes traceability possible when auditors, risk owners or incident responders need a current answer.

Practical implication: move inventory creation into the deployment path so the record is created when the AI asset is born, not weeks later.

Why agents need different inventory treatment than models

Models predict, but agents act. That distinction matters because agents can query data, call tools and trigger workflows, which expands the governance surface beyond output quality into operational behaviour and access reach. An inventory that tracks only models records the safer half of the estate and misses the systems that can cause direct change. Once agents are treated as first-class inventory objects, ownership, lifecycle stage and data access become auditable attributes rather than assumptions.

Practical implication: classify agents separately from models and require explicit fields for action scope, connected tools and reachable data sources.

What capture-at-source changes in AI governance

Capture-at-source means the inventory is populated from code, deployment pipelines or orchestration metadata instead of manual entry. That reduces the gap between what was approved and what is actually running. It also makes it easier to attach assessments, risk tiers and named owners before production exposure. The governance win is not automation for its own sake, but a shorter distance between registration and accountability, which is where most inventories fail.

Practical implication: integrate inventory registration with CI/CD or deployment workflows so every new AI asset is registered before it can operate.

NHI Mgmt Group analysis

AI inventory is now the first governance test because visibility precedes every other control. If an organisation cannot state what AI systems exist, it cannot credibly assign ownership, classify risk or evidence oversight. That is true for models, but it becomes sharper for agents because runtime action expands the governance burden beyond a simple catalogue. The practitioner conclusion is straightforward: inventory completeness is the prerequisite for any defensible AI control model.

Agents expose the gap between approved capability and operational reality. A model may be approved in principle, while the agent built on top of it queries data, calls tools and triggers workflows in production. That means the governance object is not the model alone, but the behaviour plus the access path around it. Teams that inventory only model artifacts are documenting intent, not control.

Capture-at-source is the real governance shift, not just better record-keeping. Manual inventories decay because AI estates move faster than human reconciliation. When inventory creation happens at deployment, ownership and risk classification can travel with the asset instead of being reconstructed later from memory and spreadsheets. The implication is that AI governance moves from retrospective audit support to live control of the estate.

Inventory and lifecycle management are converging into one discipline for AI systems. The article shows that ideation, production and retirement are only meaningful if they are visible in the same record that holds owner, risk tier and data access. For governance teams, the hard part is no longer naming the process. It is maintaining a single control record that remains accurate as assets spawn, evolve and disappear.

From our research library:

• 76% of organisations cite shadow AI as a definite or probable problem, up from 61% in 2025, according to HiddenLayer's 2026 AI Threat Landscape Report.
• Read next: Shadow AI and AI Agent Discovery Guide

What this signals

AI inventory debt: once AI assets are registered manually, the record starts to diverge from production almost immediately, which means governance teams are auditing a version of the estate that no longer exists. That gap is most damaging where agents can act, because action without inventory coverage is action without accountability.

The practical shift is to treat inventory creation as a deployment control, not an administrative task. If registration, ownership assignment and risk classification happen at source, AI governance becomes measurable instead of reconstructive, and orphaned systems are easier to surface before they become policy exceptions.

For practitioners

• Implement deployment-time registration Register every model, use case and agent at the point it enters production, with a unique ID, owner, type and risk tier attached immediately.
• Separate agents from models in the inventory Require distinct records for systems that act, including the tools they can call and the data they can reach, rather than folding them into a model-only list.
• Automate inventory capture from code Pull asset metadata from CI/CD or orchestration pipelines so the inventory is updated from the deployed configuration instead of manual intake forms.
• Assign accountability before production Make named ownership and assessment status mandatory fields before any AI system can be treated as governed or production-ready.
• Review shadow and orphaned assets continuously Search for unregistered agents, duplicate deployments and stale records on a recurring basis so the inventory matches the runtime estate.

Key takeaways

• A live AI inventory is the control point that connects ownership, risk and data access to every model, use case and agent in production.
• Manual or periodic inventory methods fail quickly because AI estates change faster than spreadsheet-based governance can track.
• Capture-at-source is the operational answer because it keeps registration, accountability and lifecycle status aligned with the deployed system.

Key terms

• AI Model Inventory: The authoritative record of AI assets in production, including ownership, risk tier, lifecycle stage, data access, and assessment status. It exists to support accountability, auditability, and governance. Unlike a catalog, it is designed to answer who is responsible and whether the asset remains within approved bounds.
• Agentic AI: Autonomous AI systems capable of planning, deciding, and taking actions, including calling APIs, writing code, and orchestrating other agents, with minimal human oversight. Agentic AI introduces new NHI risks as agents must authenticate to external services.
• Capture-at-Source: A registration approach where asset metadata is created from code, deployment or orchestration events rather than manual entry. It reduces drift by making the inventory reflect the same system that actually ships and runs in production.
• 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.

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.