Skip to content
All articles
ISO 42001 7 min read· by Lars Zimmermann

AI officer: the tasks in the first year, quarter by quarter

What an AI officer actually does in the first year: stocktake, rules, approvals, training, suppliers and a review. A plan in four quarters.

In short

In the first year an AI officer delivers four things: an overview of all AI tools, understandable rules with an approval route, briefed staff with a record and a reporting channel for incidents. At the end of the year there is a review with numbers. It shows how large the role needs to be and whether it is better placed in-house or externally.

Auf Deutsch lesen: deutsche Fassung

Job descriptions for AI officers often read like a wish list. Develop strategy, assess risks, foster innovation, ensure compliance. All correct. None of it says what the person does on the first Monday.

This article does. One year, four quarters, one result each that is on the table at the end. The plan works for an external just as it does for someone in-house. It assumes one thing: a written mandate from management with a time budget and authority. Without that, every task list is decoration.

First quarter: see what is there

The first task is counting. Which AI tools run in the company? IT rarely has the full answer because it only knows what it introduced itself. The assistant in the office suite came with the licence. The forecast in the ERP came with an update. The chatbot account in sales runs on a private email address.

The way therefore leads through the departments, with questions about the work: what do you write quotes with? What do you translate with? What do you use when it has to be quick? Asking about AI tools directly yields less, because many people do not know that their program contains one. Tone matters. Nobody gets into trouble for a tool they disclose.

What happens to a tool that runs without approval? The reflex is to ban it. It is usually wrong. A tool someone uses in secret solves a real problem. One of three classifications is more useful. It stays because it is uncritical. It stays with a condition, such as a company account only and no customer data. Or it is replaced by an approved tool that does the same job. A ban without a replacement moves the use to the private phone, where nobody sees it any more.

At the end of the quarter three things exist:

  • The AI inventory. For each tool: purpose, provider, data types, personal data, responsible person. Anyone working to ISO/IEC 42001 covers part of the resource documentation with it (Annex A, control A.4.2).
  • A first classification under the EU AI Act. For each tool: is this high-risk, does a transparency duty apply, or neither? For most office tools the answer is quick. The two or three cases where it is not quick are the important ones.
  • Three immediate rules. No customer data and no personnel data in public chatbots. No private accounts for company purposes. Register every new tool before first use. That is not yet a policy. It stops the biggest leaks while the policy is written.

Second quarter: rules and enablement

Now the AI policy is written. It builds on the inventory, that is, on what is really in use. A policy written before the stocktake regulates a company that does not exist.

A usable policy answers four questions: which tools are approved? Which data may go where? Who checks a result before it leaves the building? Who do I turn to when unsure? It fits on two pages and is written in the language of the shop floor. Management approves it. Without that signature it is a proposal (ISO/IEC 42001, A.2.2).

The question about data causes the most work. A split into three levels that anyone can apply without looking it up is practical:

  • Free: content that is public anyway and general questions without reference to customers or persons. It may go into any approved tool.
  • Company tool only: internal documents, quotes, analyses. They may go into tools for which a contract with the provider exists.
  • Only after individual approval or not at all: personnel data, health data, customer drawings under confidentiality, access credentials.

Then there is the approval route for new tools. It has to be fast. If an approval takes six weeks, the department uses the tool without approval from week two. A request needs five pieces of information: what the tool is, what it is needed for, which data goes in, where it is processed and who is responsible in the department. The answer comes within a few days and reads approved, approved with conditions or rejected with reasons. Rejections are kept too.

The third building block is briefing the workforce. Article 4 of the EU AI Act, as amended by Regulation (EU) 2026/1744, requires measures that support the development of AI literacy of staff. Nobody has to guarantee a particular level. In practice that means short briefings per group, on the tools that group really uses. Record who was briefed when and on what. That list is your evidence, and it costs ten minutes per session.

Third quarter: suppliers, incidents, customer questions

Suppliers first. Every tool in the inventory has a provider, and every provider deserves a few questions: where is the data processed? Are inputs used for training? Is there a data processing agreement, and has the data protection officer seen it? What happens to the data after the contract ends? The answers are usually in the contract documents nobody read when ordering (ISO/IEC 42001, A.10.3).

Then the reporting channel. Employees need a place to turn to when something goes wrong or looks odd: wrong information given to a customer, data in the wrong tool, an output nobody can explain. The channel only works if near misses are expressly welcome and nobody is punished for a report (A.3.3). In the first year every report is a good sign. Zero reports almost never mean zero incidents.

The channel includes a procedure for the first hour. Who is informed? Who may block a tool provisionally? Who talks to the customer if wrong information has gone out? If personal data is affected, the report goes to the data protection officer at once. A personal data breach must be notified to the supervisory authority without undue delay and, where feasible, within 72 hours (Article 33 GDPR).

The third point is customer enquiries. Customers ask in questionnaires and supplier audits how you handle AI. The AI officer builds a set of ready answers with evidence: inventory, policy, training records, approval records. By the third questionnaire the answer takes an hour instead of a week.

This quarter also includes a look at the transparency duties under Article 50 of the EU AI Act, which have applied since 2 August 2026. Does a chatbot on your website talk to customers? Do artificially generated images or texts go out? For systems that were already on the market before 2 August 2026, the transition period for marking under paragraph 2 ends on 2 December 2026.

Fourth quarter: review and decide

At the end of the year the AI officer reports to management. The report needs numbers, and they arise along the way if counting started at the beginning:

  • tools in the inventory, at the start and at the end of the year
  • approval requests and their outcome
  • people briefed per group
  • reports and what followed from them
  • customer enquiries answered
  • hours spent, split by these items

Three decisions follow from the numbers. How large does the role need to be in year two? Does it stay external, move in-house or become a mixed model? And is an AI management system under ISO/IEC 42001 worthwhile because customers ask for it or own AI projects are coming? The policy is reviewed at the same time. Rules nobody follows are badly written or badly explained (A.2.4).

What is not part of the role

  • The responsibility of management. The AI officer prepares, recommends and reports. Decisions on risk and accountability to the outside stay with the leadership.
  • The work of the data protection officer. The two work closely together. The GDPR assessment stays with the data protection officer.
  • Legal advice. Classifying a tool under the EU AI Act is a professional assessment. For binding advice in a dispute there are lawyers.
  • Auditing your own work. Whoever fills the role does not audit it. The internal audit needs objectivity and impartiality (ISO/IEC 42001, 9.2.2).

If one of these limits is missing from the mandate, the role becomes a catch-all. The person then does a bit of everything and may be held to account for decisions they were never allowed to take.

After one year the company should be able to answer four questions without thinking: what runs here? What is allowed? Who was briefed? Where do I report a problem?

What you can do on Monday: decide who does the stocktake and give that person four weeks. The first quarter needs no more than that to begin. Everything else builds on this list.

The inventory, the policy and the written mandate are available as free templates:

Free templates

The first year as an AI officer in four quarters

  1. Set the mandate

    Management issues a written mandate with tasks, time budget, authority, reporting line and limits.

  2. Take stock

    In the first quarter, record all AI tools, classify them under the EU AI Act and put three immediate rules in force.

  3. Create rules and an approval route

    In the second quarter, write the AI policy on the basis of the inventory, have it approved by management and set up a fast approval route for new tools.

  4. Brief the workforce

    Brief each group on the tools it actually uses and record who was briefed when and on what.

  5. Settle suppliers and reporting

    In the third quarter, review the tool providers, set up a reporting channel for incidents and near misses and prepare answers for customer enquiries.

  6. Review

    In the fourth quarter, report to management with numbers, review the policy and decide on the future scope of the role.

Share: LinkedIn E-Mail

Frequently asked questions

What are the tasks of an AI officer?+

They keep the inventory of AI tools, maintain the AI policy, approve new tools or recommend approval, organise briefings with records, run the reporting channel for incidents, answer customer enquiries about AI and report to management.

Where does an AI officer start?+

With the stocktake. Before rules are written, it must be clear which AI tools actually run in the company, including bought-in functions in standard software and privately used accounts. The inventory is the basis for everything else.

How much time does the role need?+

That cannot be said reliably at the start. The effort depends on the number of tools, own AI projects and customer requirements. Count the hours in the first year, split by task. After that you have a robust number.

May the AI officer ban tools?+

Only if management has given them that authority in writing. No law gives it to them. The mandate should state whether they may block a tool provisionally or only recommend a block, and who decides in a dispute.

Author & expert review: Lars Zimmermann · ISO/IEC 42001 Senior Lead Auditor & Senior Lead Implementer · ISO/IEC 27001 Lead Auditor & Lead Implementer (PECB)

Last updated: 10 November 2026. Researched and reviewed to the best of our knowledge; not a substitute for individual legal advice.

Questions about your own case?

In a free 15-minute intro call we assess where you stand on ISO 42001, ISO 27001 and the EU AI Act, honestly and without a sales pitch.

Continue reading