What Article 4 says today
The article binds providers and deployers of AI systems. “Deployer” is what the Act calls whoever uses an AI system under its authority in a professional activity: a company where people work with an AI assistant is one. The European Commission puts it with this example: a company whose employees use ChatGPT to write advertising text or translate has to comply with Article 4, and its people should be informed about the specific risks, such as hallucination.
What they have to do, in the words of the article, is “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”, taking into account what they know, their training and what they will use the systems for. And the text adds: “This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.”
What changed in July 2026
Regulation (EU) 2026/1744, the “Digital Omnibus on AI”, was published in the Official Journal of the European Union on 24 July 2026 and has been in force since the 27th. The Commission had proposed removing this obligation from companies; the final text keeps it, but softens it.
| From 2 February 2025 to 26 July 2026 | Since 27 July 2026 | |
|---|---|---|
| What to do | Measures to ensure, to the best extent possible, a sufficient level of AI literacy. | Measures to support the development of AI literacy. |
| Each person’s level | A “sufficient level”. | No specific level has to be guaranteed for any individual. |
| Help for small companies | The article said nothing. | The Commission and the Member States must support them, and the Commission must publish practical examples of how to comply. |
| Who is bound | Providers and deployers. | The same. |
Source: Article 4 of Regulation (EU) 2024/1689, in its original version and in the version consolidated on 27 July 2026.
What is expected of a company, according to the Commission
The European Commission has a questions and answers page on Article 4, updated on 20 October 2026. It is guidance, not law, but it’s the most specific material there is:
- That the company has a general understanding of what AI is, how it works, which AI it uses and what its risks and opportunities are.
- That it considers its own role (whether it builds the system or only uses it) and the risk of each system.
- That the measures follow from that: according to what each person knows and what they will use it for.
- Having staff read the instructions for use is not sufficient.
- No certificate is needed, nor any measuring of what each employee knows. The company can keep an internal record of training and other initiatives.
- There is no need to appoint an AI officer or set up a committee.
The Commission also says it imposes no format and no specific training: each company has to adapt it to what it does.
Who supervises it and what can happen
Supervision lies with national market surveillance authorities, not with the Commission, and it started on 2 August 2026. Article 99 of the Act, the one on fines, sets no amount for Article 4: it leaves penalties to each Member State’s law. The Commission adds that a penalty is more likely if there is an incident caused by a lack of training or guidance.
In Spain, the Commission lists the Spanish Artificial Intelligence Supervision Agency (AESIA) as the point of contact, with its designation still pending final adoption. The draft organic law that was to set the penalties under the Act lapsed on 6 October 2026, when parliament was dissolved.
We’re not lawyers, and this guide doesn’t tell you what you’re exposed to: that’s for your adviser to look at. What it does tell you is what the law asks and how to respond to it.
How a small company can comply
- List the AI tools used in the company and who uses them.
- Run an initial training session for everyone who touches them: what it is, what it can’t do, the risks (made-up answers, data that must not go in) and the house rules.
- Adapt it to each job: accounts doesn’t need the same as the engineering office.
- Write a short usage policy and explain it in person.
- Write it down: the date, who was there and what was covered. That record is what you’ll be able to show.
- Repeat it from time to time, and when new people join or the tools change.
Other dates in the Act, if you use AI
| Date | What applies |
|---|---|
| 2 February 2025 | Article 4 (AI literacy) and the prohibited practices. |
| 27 July 2026 | The new wording of Article 4. |
| 2 August 2026 | The transparency obligations of Article 50: for example, disclosing that an image, audio or video that looks real has been generated or manipulated with AI. And supervision by national authorities begins. |
| 2 December 2027 | The high-risk systems of Annex III, which the Omnibus postponed. |
| 2 August 2028 | High-risk systems embedded in products, also postponed. |
Source: Article 113 of the Act, in the version consolidated on 27 July 2026.