AI system inventory template: the 18 fields

Identifying AI risks starts with a list of the AI systems in use. Most organisations do not have one, and the systems they forget are the ones nobody has assessed. This template gives the 18 fields to record for each system and why each is there.

#FieldWhat to recordWhy it is there
1SystemName, business function and components; a unique, persistent system IDLists each system once; the ID links to the register, the impact assessments and the logs
2Vendor or builtBuilt, bought, or bought and configured; supplier and contract; any model provider behind the productWho holds the documentation (AI Act Articles 13, 25(4))
3Model and versionModel name, version identifier, date of the last change notice, whether the version can be pinnedA vendor can change the model without telling you
4PurposeThe intended purpose as the provider states it, and the purpose in useUse outside the intended purpose changes the risk and can change your role
5UsersWho operates the system or acts on its output, by role and numberAI literacy and human oversight (Articles 4, 26(2))
6Persons affectedPeople whose money, access, rights or health the output affects, including those unaware of the system; children and vulnerable groups flaggedThe fundamental rights impact assessment (Article 27(1)(c))
7DataTraining sources if built; production inputs; personal and special-category data; where data is sent and keptData governance and the DPIA (Articles 10, 26(4))
8Role under the ActProvider, deployer or both; the ISO/IEC 42001 role; the Article 25 checkYour duties depend on your role (Articles 3, 25)
9ClassificationThe Article 6 outcome and a reference to the record, with the Article 5 screen and any Article 50 dutyProhibited, high-risk, transparency only or minimal (Articles 5, 6, 49, 50)
10OwnerOne named person accountable, with the model owner and data ownerEvery risk needs an owner
11StageOne of the seven lifecycle stagesPilots that quietly become production
12JurisdictionsWhere it is used and where the affected persons areWhich rules apply
13Period and frequency of useHow long and how often it is usedThe fundamental rights impact assessment (Article 27(1)(b))
14Impact assessmentsDPIA, fundamental rights impact assessment and ISO/IEC 42005-style assessment: reference, date, next dueArticles 26(9) and 27; ISO/IEC 42001 Annex A
15Last risk assessmentThe date, and the version assessedThe quarterly version check
16Discovery sourceDeclared, procurement, scan, audit or incidentMeasures how much shadow AI you are finding
17Pilot end dateThe date by which the pilot closes or goes through the deployment gateStops pilots running unassessed
18LogsWhere they are kept and for how longRecord-keeping and log retention (Articles 12, 26(6))

Fields 1 to 11 are the inventory of Chapter 5 of the book; fields 12 to 18 are what later chapters rely on. The workbook has the same fields as a sheet ready to fill in.

What the inventory feeds

Each line of the inventory decides what comes next. The role and the classification decide which duties of the AI Act apply (see EU AI Act obligations by role and date). The persons affected and the period of use feed the fundamental rights impact assessment. The system's purpose and data decide which lines of the AI risk register apply to it. For a bank or an insurer, a bought AI service is also an ICT service, and the same supplier details serve DORA's register of information.

Questions about AI inventories

What is an AI inventory?

A list of the AI systems an organisation builds, buys and uses, one line per system, with what each does, who uses it, whom it affects, its role and class under the AI Act, and an owner. It is the starting point for identifying AI risks, because a system nobody has listed is a system nobody has assessed.

Does the EU AI Act require an AI inventory?

Not in those words. But you cannot work out your role (Article 25), classify your systems (Article 6), meet the deployer duties of Article 26 or decide which systems need a fundamental rights impact assessment (Article 27) without knowing which systems you run. Financial entities must also keep a register of information on their ICT third-party arrangements under DORA (Article 28(3)), and AI services bought as ICT services belong in it.

How do you find AI that nobody declared?

Look beyond what teams declare: procurement and expense records, software contracts with AI features added, sign-ons to AI services, and incidents. Record where each system was found, so you can see how much was undeclared. Register line OPS-07 covers shadow AI.

The template in Excel

The free workbook holds the inventory sheet, the use-case template, the register with its scoring formulas and the FRIA mapping, all linked by the same fields.

Get the inventory template and the book

You get the links on this page straight away, and a copy by email. See how your email address is used.

Your files are ready. The links are personal to your email address and work for 7 days. A copy is on its way to your inbox.

    Start with Chapter 1, then Appendix A for the register. Questions about the method are welcome at rory@eonriskservices.com.