Which AI skills to train first by role and sector?

Why start from tasks rather than tools?

A single session on "generative AI" for all employees is a poor answer to very different needs. A lawyer who has contracts summarised, a technician who consults a document base and a recruiter who screens applications do not expose the company to the same risks.

The AI Act points in the same direction. Its Article 4 requires measures that take into account people's knowledge, the context of use and the persons on whom the systems are used. The European Commission's Q&A states that there is no one-size-fits-all training.

The gap is real. According to the Commission's Joint Research Centre (JRC, 2025), 30% of EU workers use AI systems in their work, but only 15% have received AI training.

Which frameworks to rely on?

The main public framework is DigComp, the European digital competence framework. Its version 3.0, published by the JRC in November 2025, has 5 competence areas, 21 competences, 4 proficiency levels (basic to highly advanced) and more than 500 learning outcomes. AI is integrated across the framework: each of the 21 competences includes AI-related elements, labelled as such.

Regulation (EU) 2026/1744, which amended Article 4 of the AI Act, cites DigComp among the European frameworks the Commission and Member States may take into account to promote AI literacy.

Several DigComp competences correspond directly to professional uses of AI:

• Evaluating information (1.2): checking a generated answer, spotting an error or an invented source;
• Protecting personal data and privacy (4.2): knowing what must not be entered into an external tool;
• Integrating and re-elaborating digital content (3.2) and Copyright and licences (3.3): reusing generated content correctly;
• Identifying and addressing digital competence needs (5.4): knowing what still needs to be learned.

Which skills for which profile?

Profile	Priority skills	Regulatory point of attention
All users	How a language model works, errors and hallucinations, data not to enter, checking outputs	Article 4 of the AI Act
Regular users (marketing, HR, finance, legal, support)	Writing requests, assessing quality and reliability, using AI features in business software, bias	GDPR for the data processed
Technical and data teams	Model integration, evaluation and testing, data quality, security	Provider obligations if the company places a system on the market
Managers	Work organisation, performance assessment, supporting teams	Information and consultation of the CSE (employee representative body) on new technologies (companies with at least 50 employees in France)
Decision-makers and people in charge of human oversight	Real capabilities and limits, legal framework, governance, prioritising use cases	Article 26 of the AI Act for high-risk systems

The Regulation sets no training duration per profile. Duration depends on people's starting point and the tools used: an initial assessment avoids re-training what is already mastered.

What changes by sector?

The sector does not change the basic skills. It determines the use cases on which to anchor training and the risks to address first.

• Banking and insurance. Creditworthiness assessment or credit scoring of natural persons is among the high-risk systems in Annex III of the AI Act. Teams need to recognise these uses and the rules that apply.
• Health and social care. Health data is sensitive data under the GDPR. Checking generated information and professional responsibility come first.
• Human resources. Recruitment, selection and decisions on the employment relationship also fall under Annex III. HR teams need to be able to identify these cases.
• Training and education. Admission, assessment of learning outcomes and exam proctoring in education and vocational training institutions are also listed in Annex III.
• Industry and B2B services. Use cases concern technical documentation and support for technicians; confidentiality of technical data is the main risk.

Obligations for Annex III high-risk systems apply from 2 December 2027.

How to build the AI training plan?

1. List the AI tools in use and the populations using them.
2. Identify, by profile, the tasks where AI is used or planned, and the associated risks.
3. Formulate the target skills, using DigComp or an internal framework.
4. Assess people's starting level.
5. Build modules anchored in the company's real use cases.
6. Keep a record of the actions taken and check what people can do afterwards.

The overall approach is described in AI Act Article 4: mandatory AI training, what companies must do and in AI maturity in companies: how to assess it and what to do next.