Is an AI officer mandatory? What the law actually requires
No law requires an AI officer. What the EU AI Act asks for instead, where a named person is needed and when the role is worth having anyway.
In short
No, an AI officer is not required by law. The EU AI Act does not know the role, and German law does not require it either. What is mandatory: measures on AI literacy under Article 4, transparency under Article 50 and, for high-risk AI, human oversight by named, competent people. A responsible person makes these duties workable. The role itself is voluntary.
Auf Deutsch lesen: deutsche Fassung
“From how many employees do we need an AI officer?” The question sounds reasonable. Data protection has a threshold, after all. The answer is short: there is no threshold, because there is no obligation.
I take on the role for several companies as an external. I would have a reason to write something else here. That is why it comes first: nobody has to appoint an AI officer. What the law really asks for is further down. It is less than some offers suggest, and it affects more companies than most people think.
What the law says and what it does not
Negative claims are best checked by counting. In the consolidated German version of the EU AI Act of 27 July 2026, the word for AI officer appears zero times. The word for officer appears sixteen times, each time as part of data protection officer. The regulation knows providers, deployers, importers, distributors and authorised representatives. It does not know a person with the title AI officer.
There is no German statute that requires the role either. Data protection is different. Section 38 of the German Federal Data Protection Act requires a data protection officer once, as a rule, at least 20 people are constantly engaged in the automated processing of personal data. No such rule exists for AI. Anyone who names a headcount from which an AI officer becomes due has made it up.
If an offer reads as though the appointment were mandatory, one question helps: which article, which paragraph? There is no answer to that question.
What the EU AI Act requires instead
There are obligations all the same. Three of them also apply to companies that only use AI and develop nothing themselves.
- AI literacy (Article 4). Providers and deployers take measures to support the development of AI literacy of their staff. That is the wording of Regulation (EU) 2026/1744, in force since 27 July 2026. You do not have to guarantee a particular level for anyone. The article names no number of hours, no certificate and no specific training.
- Transparency (Article 50). The labelling and disclosure duties have applied since 2 August 2026, for example when people talk to a chatbot or content is artificially generated. The Digital Omnibus did not move this date.
- Prohibited practices (Article 5). They have applied since 2 February 2025. Most companies never come near them. It is still worth knowing before someone orders a tool that infers the emotions of the workforce from voice or facial expressions.
You are a deployer sooner than the word suggests. The regulation means anyone who uses an AI system under their own authority, except for purely personal, non-professional use (Article 3(4)). You do not need your own development for that. The assistant in the office suite is enough. So is the chatbot account the sales team uses to draft quotes.
A look at the consequences puts this in order. Article 99(4) lists the infringements for which fines of up to EUR 15 million or three per cent of annual turnover are provided. The deployer obligations under Article 26 are on that list, and so are the transparency obligations under Article 50. Article 4 is not. That does not make it optional. It does show where the legislator placed the weight.
None of these duties names a person. They are addressed to the company. How you distribute the work is up to you.
Where the law does require a named person
In one place the regulation becomes personal. Article 26(2) obliges deployers of high-risk AI systems to assign human oversight to natural persons. These persons need the necessary competence, training and authority, and the company must support them.
- This concerns high-risk AI. Under Annex III that includes systems that screen job applications or take decisions about workers. The text assistant in the office is not one of them.
- The date has moved. Under Article 113 as amended by Regulation (EU) 2026/1744, the high-risk obligations apply from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I systems.
- It means oversight of one specific system. That is a task at the workplace, not a staff function. Whoever operates the applicant tool and checks its suggestions is that person. An officer for the whole company is not what is meant.
Article 26(3) states expressly that deployers remain free to organise their own resources and activities. Here too, the regulation prescribes no role. It requires that someone looks who is able and allowed to.
Why responsibility is still not up in the air
No duty to appoint does not mean that nobody is responsible. Management is. In a German limited company, managing directors owe the care of a prudent businessperson (Section 43(1) GmbHG). Whether customer data ends up in a private chatbot account is a question of exactly that organisation.
In practice it looks like this: as long as nobody is named, AI questions land with whoever happens to know a little. Sometimes that is IT, sometimes the apprentice, sometimes nobody. Each of those answers is a decision in the name of the company. None of them is documented.
Anyone working to ISO/IEC 42001 cannot avoid an assignment. The standard is voluntary. Those who apply it must define roles and responsibilities (clause 5.3 and Annex A, control A.3.2). The standard does not ask for a title either. It asks that responsibility is settled and known. That can be one person, a small committee or a split across existing functions.
The third pressure comes from outside. Customers ask in questionnaires and supplier audits who is responsible for AI in your company. Three answers show the difference:
- Weak: “IT looks after that.” IT knows the tools it introduced. It does not know the others, and it does not decide about data in sales.
- Better: “Our quality manager.” A name. It stays open whether she knows about it, how much time she has and what she may decide.
- Solid: “Ms M., appointed in writing since March, four hours a week, report to management every quarter. Here is the mandate, here is the inventory.” A follow-up question rarely comes after that.
The smallest solution that holds
So you do not need an officer. You need settled responsibility. The smallest version fits on one page:
- One person with whom all AI questions land. With a name, in-house or external.
- A written mandate from management, stating tasks, time budget, authority and reporting line.
- An inventory of the AI tools running in the company, including the bought-in ones.
- A rule on which data may go into which tools.
- A record of who was briefed when and on what. That is your evidence for the measures under Article 4.
Whether this person is called AI officer is a matter of taste. The title has a practical advantage: everyone understands it at once. It has a drawback: it sounds like a statutory role with special rights. There are none. Protection against dismissal, freedom from instructions and direct access to management are in the law for the data protection officer. For the AI officer they exist only where you write them down.
One thing the mandate does not do: it does not take responsibility away from management. Tasks can be delegated. The duty to choose the person carefully, give them the means and check whether it works stays with the leadership.
What the role does in practice and how it fits into AI governance is covered here:
The AI officer: does your company need the role?When the role is worth having without an obligation
Voluntary does not mean superfluous. The role pays off as soon as one of these applies:
- More than a handful of AI tools are in use and nobody has the overview.
- Employees regularly ask what they are allowed to do.
- Customers want to know how you handle AI.
- The company builds its own AI applications or has them built.
- A tool touches applications, workers or decisions about customers.
If none of that applies, a rule on the noticeboard and a fixed contact person are enough. That too is settled responsibility. Nobody has to build a role for which the company has no work.
The role is not prescribed. What is prescribed is that the company knows what it does with AI. Without a responsible person it rarely does.
What you can do on Monday: ask three people in the company who they would turn to with an AI question. If you get three different answers, you have found your task. It is called settling responsibility. The title comes afterwards.
If you would rather fill the role externally first and decide later with real numbers:
External AI officerPrimary sources
- Regulation (EU) 2024/1689 (EU AI Act), consolidated version of 27 July 2026, Articles 3, 4, 5, 26, 50, 99 and 113
- Regulation (EU) 2026/1744 amending the EU AI Act, EUR-Lex
- Section 38 BDSG (German Federal Data Protection Act), data protection officers of private bodies
- Section 43 GmbHG, liability of managing directors
- ISO/IEC 42001:2023, AI management systems, clause 5.3 and Annex A, control A.3.2 (iso.org)
Frequently asked questions
Is an AI officer required by law?+
No. The EU AI Act does not mention the role anywhere, and no German statute requires it. What is mandatory are measures on AI literacy under Article 4, transparency under Article 50 and, for high-risk AI, human oversight by named, competent persons under Article 26(2).
From how many employees do you need an AI officer?+
There is no threshold. The limit of 20 people in Section 38 of the German Federal Data Protection Act applies to the data protection officer, not to AI. Whether the role pays off depends on the number of tools, the questions in the company and what customers require.
Does ISO/IEC 42001 require an AI officer?+
The standard requires that roles and responsibilities for AI are defined and assigned (clause 5.3, control A.3.2). It does not require a particular title. Responsibility can lie with one person, a committee or existing functions.
What happens without an AI officer?+
Nothing for the missing role itself, because it is not prescribed. Consequences follow from breaches of the actual duties, such as the transparency obligations under Article 50 or the deployer obligations under Article 26. Management carries the responsibility for those.
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.