AI training has been mandatory for your SME since 2 February 2025, under article 4 of the European regulation on artificial intelligence. What shifts on 2 August 2026 is not the obligation itself, but its enforcement: national authorities will be able to inspect companies and impose fines. If your teams use ChatGPT, Copilot or any other AI tool without training, you are non-compliant, whatever your headcount.
Read the full guide
AI Training for Business: The Complete 2026 Guide (Funding, Certification, EU AI Act)
Answer a few questions and get a personalized assessment with recommendations tailored to your industry.
Assess my AI maturityRelated articles
Dive deeper with these complementary articles.
AI training for businesses in France costs between €350 and €3,300 ex. VAT per person in 2026, with up to 100% OPCO funding for SMEs under 50 employees. Since August 2, 2026, the EU AI Act requires every company using ChatGPT, Claude, or Copilot to guarantee minimum AI literacy across its teams, with sanctions up to €7.5M or 1.5% of global annual turnover. Here is how to design a truly operational, funded, and compliant program.
AI training has been mandatory for your SME since 2 February 2025, under article 4 of the European regulation on artificial intelligence. What shifts on 2 August 2026 is not the obligation itself, but its enforcement: national authorities will be able to inspect companies and impose fines. If your teams use ChatGPT, Copilot or any other AI tool without training, you are non-compliant, whatever your headcount. For the full picture, see the complete guide to AI training for business.
The 2 August 2026 date creates no new training obligation: it switches on enforcement. Many articles suggest the duty to train your teams starts in August 2026. That is wrong. Article 4 of the AI Act has applied since 2 February 2025. Since then, any employer deploying an AI system must ensure a sufficient level of AI literacy among the people involved.
The confusion comes from the regulation's staggered timeline. Prohibitions and the AI literacy duty took effect on 2 February 2025. Governance rules and obligations for general-purpose AI models followed on 2 August 2025. On 2 August 2026, the enforcement layer falls into place: national competent authorities, the penalty regime, real oversight. In other words, you have been under an obligation for eighteen months, but the risk of a fine becomes concrete in the summer of 2026.
Article 4 mandates an outcome, not a specific training format. The text requires providers and deployers of AI systems to take measures so that their staff, and people acting on their behalf, reach a sufficient level of AI literacy. That literacy covers the skills, knowledge and understanding needed to use these tools soundly and to weigh their risks.
In practice, the expected level depends on three factors: your teams' technical background, the context in which the tools are used, and the people affected by the AI. An accounting firm automating document sorting with generative AI does not have the same needs as an agency writing client content. The regulation sets neither a minimum duration nor a standard syllabus, which leaves room to adapt but puts the burden of proof on you.
No headcount threshold exempts you from article 4. An eight-person SME faces the same obligation as a group with thousands of employees. The test is not size but use: as soon as your company deploys an AI system in its activity, you fall within the text as a deployer.
Most SMEs are already deployers without realising it. A salesperson drafting follow-ups with ChatGPT, an assistant summarising minutes with Copilot, a communications officer generating visuals: all of these are covered uses. If those people have had no briefing on the tools' limits, on data-leak risks (GDPR) or on prompt quality (prompt engineering), you are not meeting the obligation, even if you never deployed AI in a technical sense.
From 2 August 2026, national authorities have the means to penalise. This is the date when the majority of AI Act rules come into force and enforcement begins, at both national and EU level. Member States must have designated their competent authorities and adopted their penalty regimes.
Here are the main application milestones:
| Date | What applies |
|---|---|
| 2 February 2025 | Prohibitions and AI literacy obligation (article 4) |
| 2 August 2025 | Rules for general-purpose AI models, governance |
| 2 August 2026 | Most rules apply, penalties, oversight, Annex III high-risk systems, transparency (article 50) |
| 2 August 2027 | Rules for high-risk AI embedded in regulated products |
For the legal detail of the training duty, the dedicated piece on what article 4 of the AI Act requires goes into the mechanism.
The delay being considered for some high-risk systems does not touch the training obligation. In 2026, the European Commission proposed, as part of the Digital Omnibus package, to link the application of rules on Annex III high-risk systems to the availability of support tools such as harmonised standards. This could push back the entry into application of certain high-risk obligations.
Beware the misreading: this proposal concerns high-risk systems, not AI literacy. Article 4 has applied since February 2025 and is not covered by this delay. An SME that believed itself shielded by the Digital Omnibus and skipped training its teams would be misreading the text. Until the regulation is formally amended, the prudent stance is to treat the training obligation as fully in force.
AI Act fines come in three tiers, with a protective mechanism for SMEs. Article 99 sets caps proportionate to the seriousness of the breach.
| Type of breach | Cap |
|---|---|
| Prohibited AI practices | 35 million euros or 7% of total worldwide annual turnover |
| Breach of other obligations | 15 million euros or 3% of turnover |
| Incorrect or misleading information | 7.5 million euros or 1% of turnover |
For a large company, the higher of the percentage and the fixed sum applies. For SMEs and start-ups, the rule reverses: the lower amount is used. Failing the training obligation does not fall in the heaviest tier, reserved for prohibited practices, but it remains a risk factor, especially after an incident tied to uncontrolled AI use. The fine is not the only stake either: a dispute with an employee or a client can turn the absence of training into an aggravating circumstance.
Without a written trail, a delivered training counts for nothing in an inspection. The article 4 obligation is an obligation of means, which means you must be able to show what you put in place. Documentation is the line between a compliant SME and an exposed one.
Keep at least: the syllabus followed and its objectives, attendance sheets or login records for remote sessions, individual training certificates, and a record of internal awareness actions (usage charters, framing notes). Training through a Qualiopi-certified provider produces these documents natively, which makes proof easier. If you handle upskilling internally, formalise each step anyway.
A realistic path to compliance takes a few weeks, not a day. Here is a pragmatic starting point:
This approach lines up with the broader reasons to train your teams on AI beyond the regulatory constraint alone: fewer errors, measurable time savings, controlled change management.
Is AI training really mandatory for a micro-business? Yes. Article 4 provides no headcount-based exemption. A micro-business using an AI tool in its activity is concerned just like a large company.
What actually happens on 2 August 2026? Most AI Act rules come into application and national authorities start to inspect and penalise. The training obligation itself has existed since 2 February 2025.
What kind of training satisfies article 4? The regulation imposes no set format. An action proportionate to the company's real uses and to the profiles concerned, backed by supporting documents, meets the obligation.
What penalties apply for a breach? AI Act fines reach up to 35 million euros or 7% of turnover for the most serious breaches. For SMEs, the lower of the percentage and the fixed sum applies.
Does the Digital Omnibus remove the training obligation? No. The proposal targets certain high-risk systems, not AI literacy. Article 4 still applies.
Can this training be funded? Yes. Several schemes exist, starting with your funding body. See how to fund this training and reduce your out-of-pocket cost.
Should everyone be trained the same way? No. The expected level depends on the role. Advanced use for teams producing with AI, lighter awareness for occasional use.
GrowthPerf is a Qualiopi-certified training provider specialised in AI and automation for SMEs and non-profits in the Paris region. We build your article 4 compliance from your real uses: tool mapping, literacy levels per profile, tailored training and the supporting documents required in an inspection. The goal is not to tick a box, but to make your teams autonomous and careful with AI, while securing your AI adoption in business.
To build a compliance plan before 2 August 2026, explore our AI training for business or go back to the complete guide to AI training for business.
