Rules
What the AI Act is, its risk categories and roles, and what an organisation that only uses AI chatbots already needs to arrange today.
The EU AI Act is the European AI regulation, Regulation (EU) 2024/1689. It sets rules for those who make, sell or use AI systems, with stricter rules as the risk to people increases. The law applies directly in every Member State, including to an organisation that only uses an AI chatbot.
You may encounter three names for the law: AI Act, EU AI Act and AI regulation. They refer to the same text. This article explains its structure: what an AI system is, the roles the law recognises, its risk categories and what they mean for an ordinary organisation. We checked the articles mentioned against the text on the European Commission's AI Act Service Desk on 13 October 2026, incorporating the 2026 amendments.
A regulation does not first need to be transposed into Dutch legislation. The rules apply as set out in the European text. The Netherlands only arranges who supervises and how penalties are imposed. That implementing legislation is not yet in place.
The definition appears in Article 3. An AI system is a machine-based system that operates with some degree of autonomy and infers from its input how to generate output, such as predictions, content, recommendations or decisions. A chatbot falls within this definition. A spreadsheet with fixed formulae does not.
The law also distinguishes between an AI system and a general-purpose AI model. The model is the engine, for example the language model behind ChatGPT, Claude or Mistral Vibe (formerly Le Chat). The system is what you use: the chatbot with its window, settings and connections.
Under Article 2, the EU AI Act applies to providers placing AI on the EU market, wherever they are established, and to deployers established in the EU. Importers, distributors and manufacturers of products containing AI also fall within its scope.
Most organisations are deployers. You already are one if employees use a chatbot for their work.
It happens sooner than you might think. Article 25 lists three situations in which a user or other party is treated as the provider of a high-risk system: you put your own name or trade mark on an existing high-risk system, substantially modify such a system, or change a system's purpose so that it becomes high risk. That last situation can also involve a general chatbot. Anyone configuring one to rank job applicants should consider whether they are taking on a provider's obligations. Have this assessed before building it.
Article 2 lists several exceptions. The law does not apply to AI used exclusively for defence or national security, or to systems created solely for scientific research. A deployer's obligations also do not apply to a person using AI purely privately. Business use on a personal account is not private use: the organisation remains responsible.
The Commission describes four levels on its overview of the regulatory framework for AI. On its AI regulation page, the Autoriteit Persoonsgegevens (AP), the Dutch data protection authority, points out that you must establish which category a system you develop or use falls into.
| Risk category | Example | What it requires |
|---|---|---|
| Prohibited practice (Article 5) | Emotion recognition in the workplace; assigning people a social score that leads to disadvantage | Do not offer or use it |
| High risk (Article 6, Annexes I and III) | CV selection software; assessing entitlement to benefits; assessing creditworthiness | Provider: risk management, data quality, documentation, logging, human oversight. User: the obligations in Article 26 |
| Transparency (Article 50) | A chatbot on your website; a deepfake; AI text on matters of public interest | Tell people they are dealing with AI or that content is artificial |
| Minimal or no risk | Spam filter; AI in a computer game | No specific AI Act requirements; AI literacy still applies, and the GDPR where personal data is involved |
| General-purpose model (Chapter V) | The language model behind a chatbot | Obligations for the model maker, rather than the user |
Article 5 prohibits applications the legislator considers unacceptable. They include deceptive or manipulative techniques that harm people, exploiting vulnerability due to age, disability or social circumstances, social scoring, predicting criminal behaviour solely on the basis of profiling, untargeted collection of facial images for recognition databases, emotion recognition at work and in education, and categorising people by sensitive characteristics using biometrics. Two prohibitions were added in summer 2026, taking effect on 2 December 2026: AI creating intimate images of identifiable people without consent, and AI creating child sexual abuse material.
There are two routes to high risk. The first is AI as a safety component of a product already covered by European product rules, such as machinery or toys (Annex I). The second is a list of eight areas in Annex III: biometrics, critical infrastructure, education, employment and staff management, essential services and benefits, law enforcement, migration and border control, and the administration of justice and democratic processes.
Article 6 contains an exception. An Annex III system is not high risk if it poses no significant risk, for instance because it only performs a narrow procedural task or a preparatory task. That exception ceases to apply when the system profiles people: it is then always high risk. Our article on AI in HR explains what this means in practice for a personnel department.
Systems that interact with people or generate content are subject to the transparency rules in Article 50. Those rules and the transitional periods are detailed in our article on the AI Act in autumn 2026. For the largest group, which the Commission calls minimal or no risk, the AI Act imposes no specific requirements.
The law entered into force on 1 August 2024 and becomes applicable in stages. The dates appear in Article 113. An amending regulation, the digital omnibus on AI, moved some of those dates in summer 2026. According to the Commission, that amendment entered into force on 27 July 2026.
This is the position on 13 October 2026. The full timeline and what remains a proposal are covered in the autumn 2026 article linked above.
Less than many people think, but not nothing. A chatbot for summarising, writing and searching does not appear in Annex III. Even so, as a deployer you have obligations that apply today.
If you use a high-risk system, the obligations in Article 26 apply from 2 December 2027.
Under Article 86, people subject to a decision based on such a system can request an explanation of the system's role. Record these arrangements in your AI policy. Our article on writing an AI policy provides an outline.
The Commission's AI Office supervises the makers of general-purpose models. Member States designate supervisory authorities for the rest. This has not yet happened in the Netherlands. In a statement dated 9 October 2026, the AP says that it is the intended market surveillance authority, but has not yet been designated, and will only have powers once the Dutch implementing legislation takes effect. The rules themselves already apply.
The maximum amounts appear in Article 99. Implementing legislation determines how fines are imposed in the Netherlands.
For SMEs and start-ups, the lower of the two amounts applies instead of the higher. These are ceilings, rather than fixed rates: severity, duration and the organisation's size are taken into account.
The two laws address different things. The AI Act regulates the system: what it may do and how it is made and used. The GDPR regulates personal data: whether you may process it, how much, for how long and with which safeguards.
Article 2 of the AI Act states that the regulation is without prejudice to European personal data protection law. A chatbot outside every risk category can therefore still present a GDPR problem, for example if someone pastes a client file into it. Our article on GDPR legislation explains its requirements in plain language.
PrivaSearch makes documents searchable after personal data has been replaced with pseudonyms and a colleague has checked this. It searches and quotes. It does not assess people or make decisions. The page on the GDPR and the AI Act states that this supports classification outside the high-risk categories, depending on intended use and working practices. You classify your own use yourself.
Several elements align with what the law asks of a user. Human review is mandatory and cannot be skipped. Each answer refers to the document and page, so you can check the source. Actions are logged. And employees take brief AI literacy training, repeated periodically; without current training, the workspace does not open. The page How PrivaSearch works explains how the steps follow one another.
You can connect an AI assistant: Claude, ChatGPT Business and Mistral Vibe. What such an assistant can and cannot see appears on the page about AI chatbots and the GDPR. Pseudonymised data remains personal data for your organisation, so the GDPR continues to apply.
Your environment also creates an AI policy and a DPIA starter document based on your own configuration. These are starter documents, not legal advice. To see how this works, book a demo.
The AI Act is a European regulation setting rules for making, selling and using AI systems. The rules become stricter as the risk to people increases. A small number of applications are prohibited, a limited group is high risk and must meet strict requirements, and transparency rules apply to chatbots and generated content. The law applies directly in all EU Member States.
Yes. AI Act, EU AI Act and AI regulation are three names for the same law: Regulation (EU) 2024/1689. The Dutch government and the Autoriteit Persoonsgegevens usually call it the AI regulation; in everyday use, AI Act is more common. Parts of the text were amended in summer 2026 by the digital omnibus on AI. Always read the consolidated version.
Yes. Anyone using an AI system under their own authority for work is a deployer. You must avoid prohibited practices and take measures to support your people's AI literacy. The demanding requirements for high-risk systems apply only if you use the chatbot for a purpose in Annex III, such as selecting applicants.
When it is a safety component of a product covered by European product rules, or used in one of the eight Annex III areas, such as staff management, education, essential services or law enforcement. There is an exception for systems posing no significant risk, but it does not apply if the system profiles people. The purpose of use determines the classification.
Article 99 sets out three ceilings. For prohibited practices, €35 million or 7% of worldwide annual turnover; for most other obligations, €15 million or 3%; and for incorrect information to supervisory authorities, €7.5 million or 1%. For SMEs, the lower of the two amounts applies. How fines are imposed in the Netherlands will be governed by implementing legislation that is not yet in place.
No. The AI Act regulates the AI system; the GDPR regulates the personal data going in and coming out. Both apply alongside each other. An application facing few AI Act requirements may still require a data processing agreement, a record of processing activities entry and a DPIA under the GDPR. Consider both laws for every AI application.
The AI Act timeline after the digital omnibus: what applies already, what has moved to 2027 and 2028, and what AI chatbot users need to arrange now.
What a deepfake is, which rules apply now, which prohibition is coming and how organisations tackle fraud using a fabricated voice or video.
What the Dutch Algorithm Register and IAMA are, what is mandatory and what is not, and what a municipality arranges when staff use AI chatbots.
We show it with fictitious client files from your sector: upload, detection, review, search and the connection with your AI assistant. Afterwards you'll know exactly what your team can use safely.
Book a demo →