EU AI Act: risk categories, obligations and deadlines for businesses

The EU Artificial Intelligence Act (Regulation (EU) 2024/1689) regulates AI according to the risk a system poses, so your first task is to establish which risk category each AI system falls into and whether you are its provider or its deployer. Most obligations fall on providers of high-risk systems, and the main exception to the original timetable is that the Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026) has postponed the high-risk rules: to 2 December 2027 for the high-risk use cases in Annex III and to 2 August 2028 for AI in regulated products (Annex I).

Prohibited practices have been banned since 2 February 2025, the rules for general-purpose AI models apply since 2 August 2025, and the transparency obligations since 2 August 2026. Because the Act is an EU regulation, it applies directly in the Netherlands; national authorities supervise it, and the European AI Office supervises general-purpose AI models.

What is an AI system under the Act?

An AI system is a machine-based system that operates with some degree of autonomy and infers from its input how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments. Traditional software that only follows rules fully defined by people generally falls outside the definition.

The European Commission has published guidelines on the definition and on the prohibited practices that help in borderline cases. The Act also reaches beyond the EU: it applies to providers outside the EU that place systems on the EU market, and to providers and deployers outside the EU whose system output is used in the EU.

What are the risk categories?

The Act distinguishes unacceptable risk (prohibited), high risk, limited risk with transparency obligations, and minimal risk without mandatory obligations. General-purpose AI models, such as large language models, have their own regime.

You classify a system by its intended purpose and context of use, not by its technology. A system that recommends films carries far less risk than one that helps decide who gets hired.

Which practices are prohibited?

Since 2 February 2025 the Act prohibits AI practices with an unacceptable risk. The main ones are:

• manipulative or deceptive techniques that distort behaviour and cause or are likely to cause significant harm;
• exploiting vulnerabilities related to age, disability or social or economic situation;
• social scoring that leads to unjustified or disproportionate treatment;
• predicting criminal offences solely on the basis of profiling or personality traits;
• building facial recognition databases through untargeted scraping of images;
• emotion recognition in the workplace and in education, except for medical or safety reasons;
• biometric categorisation to infer sensitive characteristics such as political opinions or sexual orientation;
• real-time remote biometric identification in public spaces for law enforcement, with narrow exceptions.

For employers, the ban on emotion recognition at work is the one that most often comes up in practice, for example in call-centre or productivity software.

Which AI systems are high-risk?

High-risk systems are AI used as a safety component of a regulated product (Annex I), and AI used in the areas listed in Annex III. Annex III includes:

• employment and workers management: screening CVs, evaluating candidates, decisions on promotion, dismissal or task allocation;
• education: admission and assessment of students;
• critical infrastructure: safety components in road traffic or the supply of water, gas, heating or electricity;
• access to essential services: creditworthiness assessment and pricing of life and health insurance;
• law enforcement, migration, border control and the administration of justice;
• biometrics, to the extent not prohibited.

A system in an Annex III area is not high-risk if it does not pose a significant risk of harm, for example because it only performs a narrow procedural task; a system that profiles natural persons always remains high-risk. If you rely on this exception, document your assessment.

When do the high-risk rules apply?

The original date for most high-risk obligations was 2 August 2026. Under the Digital Omnibus they now apply from 2 December 2027 for Annex III systems and from 2 August 2028 for AI in Annex I products.

The postponement gives more time, not an exemption. Data governance, technical documentation and conformity assessment take months to prepare. Systems already on the market before the application date are in principle only covered if their design changes significantly afterwards, with separate deadlines for systems used by public authorities.

What applies to limited and minimal risk?

Since 2 August 2026, people interacting with an AI system, such as a chatbot, must be told so unless it is obvious. Providers of systems that generate synthetic audio, images, video or text must mark the output in a machine-readable way, and deployers who publish deepfakes must disclose that the content is artificially generated or manipulated.

Minimal-risk systems, such as spam filters or AI in inventory management, have no mandatory obligations under the Act. The EU encourages voluntary codes of conduct, and the Act also addresses AI literacy of staff who work with AI systems.

What are the obligations of providers and deployers?

The provider develops a system, or has it developed, and places it on the market under its own name; the deployer uses a system under its own authority in a professional context. One organisation can be provider of one system and deployer of another.

Public bodies and certain private deployers, such as banks and insurers that assess creditworthiness or price life and health insurance, must also carry out a fundamental rights impact assessment before using a high-risk system.

When does a deployer become a provider?

A deployer that puts its own name or trademark on a high-risk system, substantially modifies it, or changes its intended purpose so that it becomes high-risk is treated as the provider. All provider obligations then apply, including conformity assessment.

Check this before you rebrand a supplier’s tool or retrain it for a new purpose, and record in the contract which party bears which obligations.

What are the fines?

For prohibited practices, fines can reach €35 million or 7% of worldwide annual turnover, whichever is higher. Most other breaches can be fined up to €15 million or 3%, and supplying incorrect or misleading information to authorities up to €7.5 million or 1%.

For SMEs and start-ups, the lower of the two amounts applies. Enforcement lies with the national supervisory authorities designated by the Netherlands and, for general-purpose AI models, with the European AI Office.

How does the AI Act relate to the GDPR?

The AI Act does not replace the GDPR: if your AI system processes personal data, both apply. You need a legal basis for processing personal data in training and use, you must inform the people concerned, and decisions based solely on automated processing that significantly affect people are restricted.

For processing likely to result in a high risk, the GDPR requires a data protection impact assessment (DPIA). Link it to the fundamental rights impact assessment under the AI Act to avoid doing the work twice. The Dutch Data Protection Authority (Autoriteit Persoonsgegevens) supervises the GDPR side.

How do you set up a compliance plan?

Start with an inventory, determine your role and the risk category for each system, and document every step. Use the time until 2 December 2027 to prepare, not to wait.

1. List every AI system you develop, buy, import or use, including AI functions in HR or CRM software.
2. Determine your role for each system: provider, deployer, importer or distributor.
3. Check each system against the prohibitions and against Annexes I and III, and record your reasoning.
4. For high-risk systems, analyse possible harms and plan measures such as bias testing and human oversight.
5. Put transparency measures for chatbots, generated content and deepfakes in place now.
6. Review contracts with AI suppliers so that you receive the instructions, logs and information you need as deployer.

What should you check as a provider?

• Is the system high-risk under Annex I or III, and have you documented any reliance on the narrow-task exception?
• Is your risk management and quality management system ready for 2 December 2027 or 2 August 2028?
• Does generated content carry a machine-readable marking?
• Is the system registered in the EU database before it is placed on the market?
• If you are established outside the EU: have you appointed an authorised representative in the EU for high-risk systems?

What should you check as a deployer?

• Do you use any prohibited practice, such as emotion recognition at work?
• Do people know when they are dealing with a chatbot or AI-generated content?
• For high-risk tools, such as recruitment software: have you assigned human oversight and informed your workers and works council?
• Do you keep the system logs for at least six months?
• Have you rebranded or modified a supplier’s system in a way that makes you the provider?

What can we do for you with the AI Act?

We advise businesses on the AI Act, the GDPR and the contracts with AI suppliers that go with them, as part of our work as IT lawyer.

• We classify your AI systems and record the analysis for supervisors.
• We determine your role per system and the obligations that follow from it.
• We review and negotiate contracts with AI suppliers and customers.
• We combine the DPIA and the fundamental rights impact assessment into one process.
• We advise on informing workers and the works council before AI is used at work.

Summary

• The AI Act regulates AI by risk: prohibited, high-risk, limited-risk with transparency obligations, and minimal-risk.
• Obligations depend on your role; a deployer that rebrands or substantially modifies a high-risk system becomes the provider.
• Prohibitions apply since 2 February 2025, general-purpose AI rules since 2 August 2025 and transparency obligations since 2 August 2026.
• The Digital Omnibus postponed the high-risk rules to 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
• Fines reach €35 million or 7% of worldwide turnover; the GDPR applies alongside the AI Act.

Frequently asked questions