AI training record: a template for Art. 4 of the AI Act

Under Art. 4 of the AI Act, companies that use AI systems must take measures to promote the AI literacy of their staff. Since July 2026, this has been an obligation to make an effort: a particular level, a certificate or an AI officer are not required. An internal record of which training and measures took place is sufficient. Below you will find a template, also available as a downloadable spreadsheet.

This article is not legal advice. It reflects the position as of 15 October 2026.

What changed in July 2026

Art. 4 has applied since 2 February 2025. In the original version, providers and deployers were to ensure, to their best extent, that their staff had a "sufficient level" of AI literacy. Many training providers turned this into a training obligation for all employees.

Art. 4 was amended by the Digital Omnibus, Regulation (EU) 2026/1744 of 8 July 2026. According to the European Commission, providers and deployers must now take measures that support the development of AI literacy. The measures depend on the knowledge, experience and training of the staff and on the purpose of use. There is no minimum level for individuals. Separately, the obligation to train people who oversee high-risk systems remains.

What this means in practice

In its questions and answers, the European Commission names three steps:

Clarify your role. Are you a provider, because you develop an AI system or offer it under your own name, or a deployer, because you use it? Most companies are deployers.
Assess the risk of the systems you use. Someone who uses a chat assistant for text needs different knowledge from someone who uses an AI solution to check invoices or pre-sort job applications.
Tailor the measures accordingly and record them. An internal record is sufficient.

In Germany, the Bundesnetzagentur (Federal Network Agency) has been the competent market surveillance authority since the KI-MIG of 22 July 2026. According to our information, the penalty provisions of the KI-MIG do not refer to Art. 4. Claims about high fines for missing training that have been circulating since August 2026 do not match this. If in doubt, have the position reviewed.

What counts as a useful measure

Training is useful if the person afterwards uses their tool more safely. The knowledge they need for this depends on their role:

Role	What the person should be able to do	Possible measure
Everyone who uses AI assistants	Know the capabilities and limits of a language model, recognise inadmissible data input, check results	2 to 3 hours of basics with exercises, read and confirm the AI policy
Business users of an AI solution	Recognise errors in the solution, know when the solution passes a case on and how approval works	Briefing on real cases, operating documentation
Managers	Assess use, evaluate risks, assign responsibility	Half-day workshop on benefits, limits and rules
IT and data protection	Vet providers, assess the data and permissions model and the logs	Technical training, checklist for new tools
Oversight of high-risk systems	Monitor results, intervene, switch off	Separate training under the obligations for high-risk AI

According to Bitkom, 70 per cent of companies with 20 or more employees already train their staff on AI (survey from September 2026). If you already provide training, often only the record is missing.

Template: AI training record

For each measure, record who took part and what was covered. These fields are enough:

Field	Example
Date	12/11/2026
Measure	Basic training on AI assistants
Format	In person, 3 hours, with exercises
Delivered by	Internal IT / external provider
AI system concerned	Microsoft 365 Copilot
Role of participants	Sales back office
Content	How it works, limits, data protection rules, checking results
Participants	Names or number, as agreed with the works council
Materials	Slides, AI policy version 1.2
Next refresher	11/2027 or when a new tool is introduced
Responsible	Name, function

The template is available as a downloadable spreadsheet. It opens in Excel, Numbers or LibreOffice.

What else belongs in the record

A list of the AI systems in use with purpose and user group. This is how you justify which role receives which training.
An AI policy that defines which tools are permitted and which data must not be entered into them. A template is in the article AI policy for companies.
A date for a refresher, at the latest when a new tool is added.

Agree with the works council whether attendance is recorded by name. For evidence towards an authority, it is often enough that the measure took place and who belonged to the target group.

Support

We train teams in working with AI tools and AI solutions. The training is based on the participants' roles and includes exercises on real tasks. We do not promise a certificate, because none is needed. There is more on the page about AI training.

Frequently asked questions

Is there an AI training obligation for companies?

Art. 4 of the AI Act requires companies that use AI systems to take measures to ensure the AI literacy of their staff. Since the amendment by Regulation (EU) 2026/1744 in July 2026, this is an obligation to make an effort. No particular level or form of training is prescribed.

Do employees need a certificate for AI literacy?

No. According to the European Commission's questions and answers, an internal record of training and other measures is sufficient. A certificate, an AI officer or a dedicated committee are not required.

Is there a fine if we do not provide training?

According to our information as of October 2026, the penalty provisions in the German implementing act KI-MIG (AI Market Surveillance Act) do not refer to Art. 4. A lack of training can still cause harm in other contexts, for example in liability for errors. If in doubt, have the current position reviewed by a lawyer.

Who supervises AI literacy in Germany?

Under the KI-MIG, the Bundesnetzagentur (Federal Network Agency) is the competent market surveillance authority. For individual sectors, the sector supervisors remain responsible, for example BaFin in financial services. The Bundesnetzagentur offers guidance on building AI literacy through its AI service desk.