Inventorying AI systems for ISO 42001

The first step towards ISO 42001 is knowing what you have. Without a complete overview of your AI systems, you cannot assess risks, set a policy or build a management system. The EU AI Act also states that organisations must be able to classify their systems by risk level, and that starts with the inventory too. You can read more about that classification on the page high-risk AI classification.

Most organisations discover during this step that they use more AI than they thought. Purchased software packages with AI components, built-in recommendation algorithms, automatic scheduling systems: it all counts. And anything that is not mapped falls outside your AIMS, with all the risks that brings.

What counts as an AI system?

The definition in the AI Act is deliberately broad: any machine-based system that generates outputs such as predictions, recommendations, decisions or content based on input. That covers more than the obvious applications.

Type of AI system	Examples
Generative AI	ChatGPT, Copilot, internal LLM tools
Classification and scoring	Credit scores, fraud detection, spam filters
Recommendation systems	Product recommendations, news feeds, matching platforms
Image recognition	Quality control, facial recognition, medical image analysis
Predictive analytics	Churn prediction, demand forecasts, maintenance prediction
Automated decisions	Routing, prioritisation, application processing
Speech recognition	Transcription, voice assistants, customer service bots
Scheduling systems	Staff planning, logistics, resource allocation

Not sure whether something counts as an AI system? Use this rule of thumb: if the system independently generates outputs based on data (even without explicit programming for each case), it counts.

Shadow AI: the hidden stock

Shadow AI is AI use that falls outside the view of IT and compliance. Employees use free AI tools, colleagues share ChatGPT outputs in client documents, and departments buy SaaS tools with built-in AI features.

This is not only a governance problem for ISO 42001. It is also a risk for the EU AI Act: if you do not know about high-risk AI, you cannot meet the requirements for it either.

How to detect shadow AI:

Ask HR and procurement. Which tools were bought outside the standard IT process? Which tools does HR reimburse for employees?
Analyse IT traffic. Which AI services reach your network? Tools such as Netskope or Zscaler can make this visible.
Interview department heads. Ask directly: “Which AI tools do you use, including free or personal accounts?”
Check app stores. Which browser extensions and mobile apps do employees allow to be installed?

Shadow AI causes the biggest unpleasant surprises during audits. An employee who runs client texts through ChatGPT without permission is already a finding, even if you did not know about it.

The AI register: structure and example table

An AI register is the central document of your inventory. It is also a mandatory element of your AIMS. For each system, you record at least:

Name and description
Owner (person or department)
Provider (built in-house, SaaS, open source)
Purpose of use
Users and target group
Data processed (including whether it concerns personal data)
Risk level (minimal, limited, high: see high-risk AI classification)
Status (in use, in development, retired)
Date of last assessment

Below is an example of a completed register for a medium-sized HR technology company:

System	Owner	Provider	Purpose	Data	Risk level	Status
CV ranker	Head of Recruitment	Built in-house (Python/ML)	Score and rank job applicants	Personal data, CV text	High-risk	In use
Scheduling assistant	HR Operations	SaaS supplier X	Create work rosters	Employment data	Minimal	In use
Service desk chatbot	IT Manager	Azure OpenAI	Answer internal helpdesk questions	Internal knowledge base	Limited	In use
Turnover predictor	Analytics	Developed in-house	Predict attrition risk per employee	Personal data from the HR system	High-risk	In development
Email filter	IT Manager	Microsoft 365	Filter spam and phishing	Email content	Minimal	In use

The CV ranker and turnover predictor in this example fall under high-risk AI (Annex III of the EU AI Act: recruitment and employment management). That means stricter requirements once the high-risk obligations apply.

Step-by-step plan for the inventory

Define the scope

Determine which part of the organisation you are inventorying. Is it the whole organisation, or only the parts you include in the AIMS scope? Document this.

Consult sources

Gather input from several sources: IT asset management, procurement systems, conversations with department heads, contracts with software suppliers. Also look at cloud and SaaS subscriptions.

Detect shadow AI

Use the methods above. Schedule short conversations with key people in each department. Ask explicitly about free tools and personal use.

Draw up the register

Fill in the register based on the information you have gathered. Be thorough but pragmatic: a complete register with basic information is better than a half-finished register with perfect details.

First risk classification

Classify each system by risk level. This does not have to be a final assessment yet, but it sets priorities for the deeper analyses. See high-risk AI classification for the criteria.

Record ownership

Make sure each system has a named owner. No owner means no accountability, and every auditor will record that as a nonconformity.

Keep the register alive

Schedule half-yearly reviews. Agree that new AI systems are reported before they go into use, not after.

Tooling for AI inventory

You do not have to keep the register in a spreadsheet (although that works well for smaller organisations). Specialised tools are available.

Approach	Examples	Suitable for
Spreadsheet	Excel, Google Sheets	Small to medium-sized, up to about 20 systems
GRC platform	ServiceNow GRC, OneTrust	Large organisations with existing GRC tooling
AI governance tools	Credo AI, IBM OpenPages, Holistic AI	Organisations with many AI systems
ISMS tools with an AI module	Vanta, Drata, Secureframe	Organisations that also manage ISO 27001

Which tool you choose matters less than whether you keep the register up to date. An up-to-date spreadsheet is better than a neglected enterprise tool.

What you deliver after this step

After the inventory, you have:

An up-to-date AI register with all known systems
A first risk classification per system
Insight into shadow AI use
Named owners per system

This is the basis for the gap analysis and step-by-step plan, the risk and impact analyses per system and the build-up of your AI management system.