Under the AI Act, a small business with a website chatbot or staff using ChatGPT has two main obligations: visitors must be told they are talking to an AI system, and people who work with AI must be helped to understand how it works and where it goes wrong. The strict “high-risk” rules cover other uses, such as hiring.
What the AI Act is
The AI Act is Regulation (EU) 2024/1689, the European law that sets common rules for artificial intelligence systems according to their risk. As a regulation, it applies directly in Romania, with no national transposing law. It entered into force on 1 August 2024 and mostly applies from 2 August 2026 (Article 113). Its timetable was amended by Regulation (EU) 2026/1744, known as the “Digital Omnibus on AI”, in force since 27 July 2026.
The law assigns obligations by role. A provider develops an AI system, or has one developed, and puts it into service under its own name (Article 3(3)). A deployer uses an AI system in its professional activity (Article 3(4)). A business whose staff work with ChatGPT is a deployer. A business that commissions a chatbot and publishes it on its website under its own name may be considered the provider of that chatbot.
What must a visitor talking to the chatbot be told?
Article 50(1) requires that people who interact directly with an AI system are informed that they are talking to an AI, unless this is obvious to a reasonably well-informed and observant person. The information must be given clearly and distinguishably, at the latest at the first interaction (Article 50(5)).
The obligation applies from 2 August 2026. Regulation (EU) 2026/1744 did not postpone it: it moved the deadlines for high-risk systems, not this information rule.
Formally, the obligation falls on the chatbot’s provider, in the sense above. In practice, it is easy for a business to check: the chatbot’s first message says that it is an AI assistant, not a person. With an off-the-shelf service, check that the supplier displays this message and that you can keep it. On the chatbots we build, the welcome message says so from the start.
Article 50(2) also requires text, images and audio that are generated to be marked automatically, in a format that software can read. This is an obligation on the providers of systems that generate content, such as the company that develops the AI model, not on the business that uses it on its website. For systems placed on the market before 2 August 2026, the deadline is 2 December 2026 (Article 111(4), added by Regulation (EU) 2026/1744).
What does the law ask when staff use ChatGPT?
Article 4 requires providers and deployers to take measures that support the “AI literacy” of their staff and of others who use AI on their behalf. The measures take account of people’s knowledge, their experience and the context in which they use AI. The obligation has applied since 2 February 2025. Since 27 July 2026, following the amendment by Regulation (EU) 2026/1744, the article states explicitly that a business does not have to guarantee a specific level of knowledge for each person.
The European Commission answers the ChatGPT question directly: a business whose staff use it, for example, for advertising copy or translations is covered by Article 4, and people should be told about the risks, such as made-up answers. The Commission also states that no certificate is needed, that a business can keep an internal record of training and that there is no mandatory course format.
Two transparency rules also concern the content you publish (Article 50(4)). Generated images, audio or video that look real (deepfakes) must be disclosed as artificially generated. Text generated with AI and published to inform the public on matters of public interest must be disclosed as such, unless a person has reviewed it and someone holds editorial responsibility for it.
Is a business chatbot a high-risk system?
As a rule, no. The European Commission groups systems by level of risk: chatbots fall under transparency risk, with the information obligation above, while the vast majority of AI systems used in the EU are minimal risk, for which the AI Act brings no new rules. High-risk systems are those in the areas listed in Annex III to the regulation, or safety components of certain regulated products (Annex I). A chatbot that answers questions about opening hours and prices, or an assistant that drafts emails, is not on that list.
A small business can still end up in high-risk territory through the way it uses AI. Examples from Annex III:
- sorting CVs, filtering applicants or assessing them when hiring (point 4(a));
- decisions on promotion, ending a contract, allocating tasks or monitoring staff performance (point 4(b));
- assessing the creditworthiness of individuals (point 5(b));
- risk assessment and pricing in life and health insurance (point 5(c)).
For these systems, the obligations apply from 2 December 2027, and for those in the regulated products of Annex I, from 2 August 2028 (Article 113(c), as amended). A business that uses such a system must, among other things, use it in line with the provider’s instructions and assign people with the necessary training and authority to oversee it (Article 26). Separately, using AI to infer a person’s emotions in the workplace has been prohibited since 2 February 2025, except for medical or safety reasons (Article 5(1)(f)).
How does the AI Act relate to the GDPR?
The AI Act does not replace the GDPR (Article 2(7)). When customer data goes into an AI assistant, whether through the chatbot on your website or pasted into ChatGPT by a member of staff, the GDPR rules still apply. Two matter straight away for a small business:
- use only the data needed for the task in question (GDPR Article 5(1)(c), “data minimisation”);
- if a supplier processes data on the business’s behalf, the relationship needs a contract that sets out what it does with the data (GDPR Article 28).
This gives a practical rule: do not enter customer data into an AI account for which the business has no such contract. What data to keep out of ChatGPT and how to use it safely is covered in the guide on ChatGPT in business. In our projects, data is stored on servers in the European Union, and if a service processes it outside the EU, we ask for the client’s approval first.
Who enforces the rules in Romania?
Each Member State designates the authorities that supervise the application of the AI Act (Article 70) and lays down the penalties, taking into account the interests of small and medium-sized businesses (Article 99(1)). The deadline for designating the authorities was 2 August 2025.
In Romania, a Government memorandum proposed ANCOM as the market surveillance authority and single point of contact, alongside other authorities for specific areas. According to ANCOM’s press release of 24 July 2026, the national implementing law is being drafted, and ANCOM and the other authorities will only be able to check compliance and impose penalties once it enters into force. On 2 November 2026, ANCOM was still referring to the national framework as being prepared. The obligations in the regulation, however, apply directly from the dates above, whatever the stage of the Romanian law.
What should a business with 5 to 20 people actually do?
- List the AI tools used in the business (the website chatbot, ChatGPT, translation, images) and the people who use them.
- Check the chatbot: its first message says that it is an AI assistant, and when it does not know an answer, it sends the customer to a person.
- Write short internal rules: which data must not be entered into AI assistants, who checks generated text before it is published and what to do with an answer that looks wrong.
- Run a short training session, suited to what each person does, on the limits of AI and made-up answers. Note the date and who attended.
- Check your contracts with AI suppliers: who processes the data, where and on what terms.
- If you use, or plan to use, AI for hiring, assessing staff or granting credit, look at that case separately, before 2 December 2027.
- Keep an eye on the Romanian implementing law and on the European Commission’s guidance, which may add detail.
How we can help
If you want to find out where AI helps in your business and which limits need to be set, an AI consulting session costs from 299 lei and ends with written recommendations. We build an AI chatbot for your website with the information message and the handover to a person agreed before launch. What the price includes is in the guide on what an AI chatbot costs, and the differences between an off-the-shelf service and one built for your business are in the guide on off-the-shelf or custom chatbots.
This guide explains the general rules and is not a substitute for legal advice on your business’s particular situation. Updated on 13 November 2026: the information has been checked against the official text of the regulation and the European Commission and ANCOM pages as of that date.