You cannot audit what you cannot list. That plain fact is why the AI system inventory sits at the foundation of nearly every governance and audit program. Ask a mid-sized organization how many AI systems it runs and the honest answer is usually "fewer than we really have," because AI hides inside vendor software, spreadsheets with macros, marketing platforms and the browser tabs of curious employees. This guide shows how to build and maintain an AI system inventory: what to include, which fields to record, how to discover the AI you did not know about and how to keep the list from rotting. If you want formal training in audit practice, see the Certified AI Audit Professional certification.

What is an AI system inventory?

An AI system inventory, also called an AI register or catalog, is a central, maintained record of every AI model, tool, feature and use case an organization builds, buys or uses. Each entry captures who owns it, what it does, what data it touches, how risky it is and where it stands in its lifecycle. It is the map every other control depends on.

Why the inventory matters for audits

Auditors usually start here. The inventory defines the audit universe, drives risk-based sampling and exposes governance gaps at a glance. A missing or stale inventory is itself a finding, because it means risk decisions are being made without knowing what exists.

It shows what could be audited and in what order
It links systems to owners who can answer questions
It supports risk tiering and regulatory mapping
It reveals shadow AI and unmanaged vendor tools
It gives incident responders a head start

What counts as an "AI system" for inventory purposes

Define it broadly enough to catch real risk, and clearly enough that people can apply it. Include the following.

Models and systems you build

Machine learning models, scoring engines, recommendation systems, forecasting tools and generative applications developed in-house, along with their pipelines.

Tools and platforms you buy

Third-party software with AI features, such as resume screeners, fraud detectors, support chatbots, analytics platforms and copilots embedded in office suites. If a vendor's tool makes or shapes decisions using AI, it belongs.

Foundation models and APIs

External models accessed through APIs, including which applications call them and for what.

Informal and shadow use

Public chatbots, browser extensions and personal accounts used for work. Record them as unmanaged uses so that they can be addressed, not ignored.

Use cases, not just products

One tool can serve several uses with different risk. Record the use case separately: a language model used for drafting internal notes is a different risk from the same model screening candidates.

Fields to record

Keep the template focused enough that people will fill it in. The table shows a practical core set.

Field	What to capture
Name and ID	Unique identifier and common name
Description and purpose	What it does and the business problem it solves
Owner and sponsor	Named business owner and technical owner
Type and source	Built in-house, vendor product, foundation model API or informal tool
Data used	Categories, sensitivity, personal data, sources
People affected	Customers, employees, applicants, public, vulnerable groups
Risk tier	Rating and the reason for it
Lifecycle stage	Idea, development, pilot, production, retired
Controls and evidence links	Assessments, test reports, approvals, monitoring
Vendor details	Supplier, contract, sub-processors, model versions
Review dates	Last review, next review, last audit

Link, do not duplicate

Keep the inventory lean by linking to detailed documents such as model cards and assessments, instead of copying them in.

How to discover what you did not know about

Self-reporting alone misses things. Combine methods.

Ask people

Survey department heads and run short interviews with product, HR, marketing, finance, operations and customer support. Ask what tools they use to save time, and make it safe to answer honestly.

Check the paper and money trails

Review procurement records, software subscriptions, expense claims, vendor lists and contracts for AI features and services.

Check the technical trails

Scan cloud accounts, API usage, code repositories, network and identity logs for AI services and model calls. Developers and security engineers can automate much of this, and many strengthen those skills through tech certifications.

Make registration the easy path

Add an inventory step to procurement, project intake and architecture review, so new AI enters the list on day one.

Keeping the inventory alive

A stale inventory is worse than none, because it gives false comfort.

Assign ownership and rhythm

Name an inventory owner, require system owners to confirm entries on a schedule and review high-risk entries more often. Trigger updates when systems change, vendors update models or new uses appear.

Integrate with other processes

Connect the inventory to risk registers, incident management and audit plans. When an incident occurs or a vendor announces a model change, the inventory should show who and what is affected.

Protect integrity

Control who can edit entries and keep change history. Tamper-evident approaches are discussed in deep tech certifications. Make sure the people maintaining and approving the inventory are qualified, with verification through HireValid where credential checks apply.

Share carefully

Some inventory content is sensitive. Publish summaries externally only when accurate and appropriate. A digital marketing course can help communications teams describe AI use accurately and transparently.

Conclusion

An AI system inventory catalogs models, tools and use cases with owners, data, risk tiers, lifecycle stages and evidence links. Define "AI system" broadly, record uses as well as products, discover hidden AI through people, paper and technical trails, and keep the list current with owners, triggers and integrations. It is the quiet foundation under every good audit.

Frequently asked questions

These are the questions people ask most about AI inventories.

1. What is an AI system inventory?

A central record of all AI models, tools and use cases an organization builds, buys or uses.

2. Why do auditors start with the inventory?

It defines what can be audited and drives risk-based sampling.

3. What should be included?

In-house models, vendor AI tools, foundation model APIs and informal uses.

4. What is shadow AI?

AI tools used without formal approval or visibility.

5. What fields should I record?

Owner, purpose, data, affected people, risk tier, lifecycle stage, vendor and review dates.

6. Should I list use cases separately?

Yes, because risk depends on how a tool is used.

7. How do I find hidden AI?

Combine surveys, procurement review and technical scanning.

8. Who owns the inventory?

A named inventory owner, with system owners responsible for their entries.

9. How often should it be updated?

On a schedule and whenever systems, vendors or uses change.

10. Can a spreadsheet work?

Yes to start, as long as it has controls and version history.

11. When do I need a platform?

When volume, integrations or audit demands outgrow a spreadsheet.

12. Should vendor features be included?

Yes, if AI makes or shapes decisions.

13. How does it link to risk tiers?

Each entry carries a risk tier that sets review depth.

14. Should the inventory hold full documents?

No. Link to model cards, assessments and test reports.

15. How do I get staff to register AI tools?

Make it easy, tie it to procurement and avoid punishing honest disclosure.

16. Can inventories be audited?

Yes, for completeness, accuracy and currency.

17. How do I test completeness?

Sample systems found in other records and check whether they are listed.

18. Do I need to list retired systems?

Keep a record with retirement date and data disposal details.

19. Is the inventory sensitive?

Parts may be, so restrict access and share summaries carefully.

20. What is the biggest inventory mistake?

Building it once and letting it go stale.