Mandatory AI training in companies: what article 4 of the European regulation really says

Yes, some form of AI training is now expected of most European employers. Since 2 February 2025, article 4 of Regulation (EU) 2024/1689, the AI Act, has imposed a so-called AI literacy obligation: any organisation that deploys or uses AI systems must take measures so that the people who use them have a sufficient level of understanding. It is not a bureaucratic nightmare, it is not a diploma, and it does not target only technical roles. A hotel whose front desk uses ChatGPT to draft replies to guests is concerned, just as much as a large group.

I am Tiffany Weltman, a trainer in generative artificial intelligence for the hotel and restaurant industry, based in Paris. Since January 2024, I have trained more than 2,500 professionals in the sector, with engagements at Accor, Paris Society, Airelles and the Plaza Athenee, and on the order of 250 days delivered. My organisation is Qualiopi-certified, which makes the training fundable, often through the sector's OPCO (AKTO for the French hospitality sector). In this article, I explain what article 4 really requires, what it does not require, and where to start without complicating your life.

Who the obligation applies to

The common belief is that the AI Act only concerns software vendors or high-risk AI systems. That is wrong for article 4, which targets both the providers and the deployers of AI systems. In practical terms, this covers a very large majority of employers.

In a hotel or a restaurant, you are very likely concerned if:

• your teams use ChatGPT, Claude, Gemini or Copilot to write emails, replies to reviews or social posts;
• your front desk or your sales team relies on a tool that includes AI (translation, summarising, suggested replies);
• your marketing generates visuals or text with an AI tool;
• a business software you use embeds AI features, even without you having explicitly chosen it.

In other words, the question is not whether you are concerned, but to what extent. The more AI touches sensitive matters (guest data, HR decisions, public content), the higher the expected level of literacy.

What article 4 really requires

The text asks the employer to take measures to ensure, as far as possible, a sufficient level of AI literacy among its staff and among the people who use these systems on its behalf. The key word is sufficient: the expected level is proportionate to each person's role, their knowledge and the context of use.

Concretely, a sufficient level of literacy means that your teams know:

• what a generative AI tool is, what it can do and its limits, starting with the plausible-sounding errors it produces without warning;
• which data can be entered into it, and above all which data is never entered (this is directly linked to the topic of AI and GDPR in hospitality and restaurants);
• how to reread, check and correct an output before sending it to a guest;
• when a human must take back control.

This foundation can be passed on through structured training, but also through clear internal guidelines and awareness sessions. What matters is that it is real, adapted to the jobs and documented. Here is an example of a simple instruction, the kind I have people write during training:

Rewrite this reply to a guest review in a warm, professional tone, in English, without inventing any detail that is not in my text and without mentioning any commercial offer. Here is the review and my rough reply: (paste them here)

This kind of prompt, where you frame the tool instead of letting it improvise, is one of the reflexes that AI literacy aims to build. A second useful reflex: systematically ask the tool to flag what it is unsure about, rather than filling the gaps with something invented. These two habits, learned in a few minutes, prevent most day-to-day slip-ups.

What article 4 does not require

This needs to be said clearly, because many businesses worry for no reason. Article 4 does not ask you:

• to create an AI manager role or a dedicated department;
• to put every employee through an exam or an individual certification;
• to produce dozens of pages of procedures that no one will read;
• to give up on AI as a precaution.

The obligation is a proportionate best-efforts obligation, not a crushing administrative burden. For a front desk team, a few hours of practical training, grounded in real daily cases, already cover the essentials. It is a common-sense reflex that becomes a reasonable obligation, not a legal threat to brandish. And this rise in skills takes nothing away from the teams: it gives them back time, absorbs repetitive tasks and returns energy to guest service.

How a Qualiopi training helps you comply

A Qualiopi-certified training ticks several boxes at once. First, it structures and documents the process: programme, objectives, assessment, certificate. This is exactly the trace that shows the employer has taken measures, as article 4 requires. Second, it is fundable, often through the sector's OPCO (AKTO for the French hospitality sector), which sharply reduces the cost for the business.

On substance, a training tailored to your teams is not limited to a theoretical presentation. It starts from your jobs: front desk, reservations, F&B, spa, management. It builds lasting reflexes around data, proofreading and verification. And it provides use cases you can reuse right away, so the time invested pays off from the very next week. From what I see in the field, trained teams often recover on the order of 4 to 8 hours a week on repetitive tasks such as replying to reviews, translation or drafting emails, but this depends heavily on the establishment and the tools already in place.

It is also the moment to connect AI use to your business tools. At CheckChak, the hotel scheduling software I co-founded, every rule applied (working time, rest, split shifts, premium pay set out in the French national HCR collective agreement, IDCC 1979) is justified, and the cost is displayed before the schedule is published. It is the same logic as AI literacy: a powerful tool, but always explainable and kept under human control.

Where to start

There is no need to overhaul everything. Here is a simple, realistic path:

1. Take stock of the real AI uses in your establishment, department by department. You will often be surprised to see how many tools already include it.
2. Set two or three clear internal rules about guest data: what you enter, what you never enter.
3. Train first the teams that use AI daily, the front desk and marketing above all, with concrete content and cases from your own property.
4. Keep a record: programme, participants, certificate. This is what makes your compliance with article 4 tangible.
5. Anchor the reflexes over time, reviewing them once or twice a year.

If you are looking for an operational starting point, I lay out a step-by-step method in my guide to getting started with AI in your establishment in 30 days. Article 4 is not one more constraint: it is a chance to equip your teams properly, secure your uses and save precious time, in the service of your guests.

This article provides practical guidance and does not constitute legal advice.