Share
EU AI Act at work: what Romanian employers must do about ChatGPT
Since 2 February 2025, every company in Romania whose employees use ChatGPT, Copilot or Gemini has had a legal obligation that few managers can explain: it must ensure AI literacy among the people who use these tools. The rule comes from Article 4 of the EU Artificial Intelligence Regulation (the AI Act), and it does not stop at companies that build the technology. It also covers those that simply use it, which means almost any office where someone asks an AI assistant to draft a sales offer or summarise a report.
In summer 2026 the wording was revised, and the Romanian law that is supposed to set out who enforces the rules and what penalties apply has still not been adopted. That is no reason for companies to wait. For employees, training and written rules on AI are becoming part of everyday office life. For HR, what is at stake is a set of documents that need to exist before anyone asks for them.
What Article 4 requires after the 2026 amendments
Regulation (EU) 2026/1744, part of the "Digital Omnibus" package published in the Official Journal of the EU on 24 July 2026, changed the wording of Article 4. Originally, companies had to ensure, to the best extent possible, a sufficient level of AI knowledge among their staff. Now they must take measures to support that knowledge. Lawyers describe the change as a shift from an obligation of result to an obligation of means: the company no longer guarantees a specific level, but it is still required to train its people in the lawful and responsible use of AI.
The regulation does not prescribe a particular format, a minimum number of hours or a certification. Nor does it require every employee to reach the same level: an accountant who summarises invoices with Copilot and a salesperson who writes offers with ChatGPT need different things. What matters is something else: the company must be able to show what it has actually done. Simply having staff read written instructions is not considered enough.
Who enforces the rules in Romania, and from when
As a general rule, the AI Act has applied across the European Union since 2 August 2026. Romania, however, still has no national law, even now in November, naming the supervisory authorities, the procedure and the penalty regime. In a memorandum adopted on 12 March 2026, the Government proposed ANCOM (the National Authority for Management and Regulation in Communications, Romania's telecoms regulator) as the market surveillance authority and single point of contact, with other institutions to take over specific areas, such as high-risk systems in the financial sector.
In a statement issued on 24 July 2026, ANCOM said it would only be able to inspect and sanction breaches of the regulation once the national legislation, still being drafted, comes into force. The absence of that law does not make the regulation optional, because it applies directly in all member states. And one law is already fully in force: when an employee pastes customer data into a public chatbot, the incident also falls under the GDPR.
For employees: what changes at your desk
Do you use AI at work? In the coming months you may well receive an internal policy setting out what you can and cannot do with these tools, and be invited to a training session. It is not a test of your competence, nor a sign of distrust. It is an obligation on your employer, and having clear rules protects you too.
A few things you can do right now:
-
Ask for the written rules: which tools are approved, what information you are allowed to enter, and who checks AI output before it goes to a client.
-
Don't put customer data into a public chatbot: especially from a personal account. The GDPR already applies here, whatever the status of the Romanian law.
-
Check before you send: a made-up answer from an AI assistant that reaches a client or your manager becomes your problem, not the software's.
-
Keep proof of your training: and if you haven't had any, ask. Knowing how to use AI responsibly is a real advantage when you're looking for other jobs.
Want to know how other companies handle this before you apply? It's worth reading employer reviews written by people who already work there. They often reveal more than the job ad about how well organised a company is with its digital tools.
For HR and employers: the steps that matter now
For most companies, compliance doesn't require a huge project, just a few well-documented steps. Tudor Dumitrescu, a senior software engineer and founder of Academia Programatorilor, sums up the problem like this: "Most problems with AI at the office aren't technical, they're everyday ones: a document with customer data pasted into a public chatbot, a made-up answer forwarded without checking, a fake message taken as genuine. For most companies, Article 4 doesn't mean bureaucracy, it means a few clear rules and people who understand why they exist."
-
An inventory of tools: find out which applications people use, in which departments and for which tasks, including personal accounts used for work purposes.
-
A short acceptable-use policy: approved tools, the types of information that must never be entered into them (customer data, personal data, confidential documents) and the obligation to check the output.
-
Role-specific training: sales, accounting and recruitment use AI differently, so they need different examples. You don't need a certification or a fixed number of hours.
-
Some form of assessment: a short quiz or practical exercise shows that people have understood the rules, which a document sent out for reading cannot prove.
-
Records: who was trained, when, on what content and with what assessment result. This is the file you will present once the national law comes into force.
Link your AI policy to the GDPR procedures you already have. The most likely incidents, such as customer data pasted into a public chatbot, can be penalised right now under data protection rules, without waiting for the law implementing the AI Act.
AI in recruitment: the 2 December 2027 deadline
The AI Act classifies systems used in recruitment and workforce management as high-risk. For these, additional obligations apply from 2 December 2027. They cover, for example, tools that screen applications or help assess candidates.
For HR, the time left until 2 December 2027 is time for an audit: which AI features the recruitment platform already has, which decisions they influence and who can review them. The selection team is, in any case, among the first that needs the training required by Article 4.
For candidates, asking whether an employer uses AI in hiring is a legitimate question at a job interview. The way it's answered says something about how transparent the process is and about the company culture.
The AI Act doesn't ask companies to turn their employees into programmers. It asks that the people who use these tools know what they are doing and what they must not do, and that the company can prove it. For employees, it's a good reason to ask for clear rules. For HR, it's a cheap checklist today that will matter far more on the day ANCOM, or whichever authority is designated, gains the power to impose penalties.
Read all the articles about
HR Metrics HR Coach Guide Workforce planning Legislative Balance Editorial Economic EthicsArticle written by:
Comments
0 comments
Access your account and add your comment
HR Metrics
Subscribe to the Newsletter
Read articles of interest from Undelucram.ro contributors