Regulation (EU) 2024/1689: Article 4

AI Act, Article 4: AI literacy training at work is a legal obligation

Article 4 of the AI Act has applied since 2 February 2025 and covers every company whose employees use artificial intelligence systems, including everyday use of ChatGPT, Claude, Gemini or Copilot. Since 2 August 2026 the national supervisory authorities are operational: from that date, inspections can start.

What Article 4 says

“Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf… This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.”

Wording of Regulation (EU) 2026/1744, the Digital Omnibus, in force since 27 July 2026. The earlier text asked to “ensure a sufficient level”: an obligation of result, replaced by an obligation of means.

In plain terms: if people in your organisation use AI, you are required to take real steps so they can use it competently, in a way proportionate to role, context and risk. The obligation applies to companies of every size: the regulation sets no thresholds.

The dates that matter

• 1 August 2024: the AI Act enters into force across the Union.
• 2 February 2025: Article 4 becomes applicable: the AI literacy obligation is live.
• August 2025: the civil liability profile: damage caused by untrained staff exposes the company to claims.
• 2 August 2026: national supervisory authorities become operational: inspections and sanctions can start.

Who it applies to

Providers (those who develop or market AI systems) and deployers (those who use them professionally). The second category is enormous: the professional firm using ChatGPT for email, the SME that switched on Copilot, the company running a support chatbot. Every size, every sector.

What "sufficient AI literacy" means

Real, proportionate skills: understanding what AI can do and where it fails, writing effective requests, recognising hallucinations and biased output, protecting personal and confidential data, knowing when the decision must stay human. A written policy alone falls well short: it takes real, measurable, documented training.

Compliance in 4 steps

1. Map who uses AI: every person touching AI tools is within the obligation.
2. Train with practice, by role: real scenarios from each person's job: HR, sales, finance, legal, operations, industry.
3. Measure the skills: objective evaluation, scores, per-person progress: competence becomes a number.
4. Document with verifiable certifications: a documented record for audits and disputes.

How Grace solves Article 4

Grace Certified turns the obligation into training your team actually enjoys: 520 real professional scenarios and a gym of everyday cases, AI evaluation across 7 dimensions with a 0-100 score, certifications with a public verification URL (valid 18 months) and team reports as a documented record. A team starts the same day and the first documented record exists after the first week.

Document your team's training →

Frequently asked questions about Article 4

Does Article 4 of the AI Act apply to small companies?

Yes. The obligation covers providers and deployers of AI systems of every size and sector: the regulation sets no size thresholds. If your staff uses AI at work, the obligation applies.

We use ChatGPT or Copilot: are we covered by the training obligation?

Yes. Anyone using third-party AI systems in a professional context is a "deployer": daily use of ChatGPT, Claude, Gemini or Copilot by employees falls squarely within Article 4.

Is a written internal policy enough to comply with Article 4?

A policy alone falls well short of the requirement: the regulation asks for measures that support the development of AI literacy, meaning real knowledge and skills, proportionate to each role and context of use. That takes actual training, measurable and documented.

What does a company risk by postponing AI training?

National supervisory authorities have been operational on the AI Act since 2 August 2026. On top of that sits civil liability: if an untrained employee causes damage using AI, the company answers for it, and the lack of documented training weighs in any dispute.

What kind of training satisfies Article 4 of the AI Act?

Training proportionate to role, context and risks: understanding what AI can do and where it fails, using it safely on real cases from your own job, recognising hallucinations and sensitive data. Practice on real scenarios with measurable evaluation is the most defensible form.

What must a company be able to document in case of an audit?

Who was trained, on what, when and with what outcome. Grace certifications carry a public verification URL and team reports show progress and scores per person: a documented record.

How long does it take to comply with Article 4?

With Grace a team starts the same day: people train on scenarios from their own sector, evaluation is instant and certifications come as levels are reached. The first documented record exists after the first week.

Informational content updated July 2026: for a legal assessment of your specific situation, consult a professional. See also AI Hygiene & GDPR.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.