AI is not approaching the education and training sector, it is already in the middle of it. Teachers and course developers produce materials with generative tools, marketing teams write campaigns with ChatGPT, administrators summarise calls with AI note-takers, and here and there an algorithm looks over shoulders during assessment. The question for a training organisation is no longer whether AI plays a role, but whether anyone can say exactly which role, with which tools, and under which agreements.
That question has recently become a legal one as well. The EU AI Act applies in phases, and the phases that affect training providers most have already begun. This article sets out what applies now, what is coming, and the four steps that bring your baseline in demonstrable order.
If you would rather have this explained in an hour: on Thursday 17 September 2026 I present a free webinar (in Dutch) on exactly this topic for NRTO, the Dutch industry association for private training providers, open to the entire education sector. Registration is open on the NRTO website.
What applies today?
Two obligations deserve every training provider's attention, because they are not waiting.
The first is AI literacy. Article 4 of the AI Act has applied since 2 February 2025 and was reworded in July 2026 by Regulation (EU) 2026/1744, the Digital Omnibus. The current text asks organisations that provide or deploy AI to take measures that support the development of AI literacy among the people working with AI systems, taking into account their knowledge, experience and the context of use. No individual end level has to be guaranteed, but that does not make the duty lighter: the emphasis is on measures you can show. An organisation that cannot show what it has done has little to fall back on under the current text.
The second is transparency. The obligations of Article 50 have applied since 2 August 2026 and, unlike part of the high-risk obligations, were not postponed. For training providers this means concretely: a chatbot answering study or enrolment questions must make clear that a machine is on the other side. AI-generated images or video in campaigns or course materials trigger marking and disclosure duties. That touches the daily practice of marketing and communications teams directly.
What is coming?
The core obligations for standalone high-risk systems under Annex III were moved to 2 December 2027 by the Digital Omnibus. For the education sector that category matters, because Annex III lists systems for admission, assessment and proctoring. A provider that uses or plans to use AI in student assessment or admission decisions is well advised to know and classify those systems now. The date sounds far away, but a register, a role determination and a risk classification take months, and the heavier obligations build on them.
The AI Act also touches training providers in their role as employers. AI in recruitment and selection is a high-risk category, and that is precisely where organisations experiment most with tools that pre-sort CVs or score candidates.
The four steps
Getting the baseline in order does not have to be a big project. Four steps, in this order, take an average training organisation from unknown territory to a demonstrable starting position.
Determine your role per AI application
The AI Act distinguishes roles with different duties. Whoever uses a supplier's AI system is a deployer. Whoever builds an AI application or offers one under its own name, for example an own chatbot or an adaptive learning environment, may be a provider and then carries heavier obligations. Determine your organisation's role per application, because everything depends on it.
Map tools and users
Inventory which AI tools are used within the organisation, including the tools employees bought or use for free themselves. Record per tool who works with it, for what, and what information goes into it. That last point is sensitive for training providers in particular: participant data, assessments and exam materials do not belong in a public AI service without agreements.
Choose appropriate measures per role and risk
A uniform AI training for everyone is rarely the right interpretation. Article 4 asks for measures that fit knowledge, experience and context. A teacher using AI for grading needs different knowledge than a marketer generating images or an administrator summarising calls. Tie the measures to the inventory from step two, and record agreements about what is and is not allowed.
Record what you have done
The common thread in the AI Act is demonstrability. A register of AI applications, a role determination per system, written agreements and a record of who followed which training: together they form the dossier that lets you show a regulator, a client or an accreditation body that AI use in your organisation is policy, not accident.
The sector is picking this up together
The good news is that the training sector does not have to figure this out alone. NRTO is actively putting AI literacy on its members' agenda, and the webinar on 17 September is a concrete example: one hour, free, focused on what a training organisation can arrange on Monday. For the training side, with per-employee records that fit the dossier of step four, there is role-based AI literacy training for training providers, and the announcement with all practical details is also on embedai.nl.
Organisations that seriously walk through the four steps usually discover the work is more manageable than the legal text suggests. The ones that get into trouble are not those that start small, but those that wait until the question comes from outside.
Frequently asked questions about the AI Act for training providers
Sources
Newsletter
Every Tuesday, the AI Act week ahead in 5 minutes
A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.
Practical and short · No spam · One-click unsubscribe