EU AI Act for businesses: what applies to you and when

Who the EU AI Act applies to

The AI Act (Regulation (EU) 2024/1689 (opens in a new tab)) entered into force on August 1, 2024. Article 2 sets its reach, and it goes beyond companies based in the EU. The Act covers:

• Providers that place AI systems or general-purpose AI models on the EU market, wherever the provider is based.
• Deployers of AI systems that are established or located in the EU.
• Providers and deployers outside the EU when the output of their AI system is used in the EU.
• Importers and distributors of AI systems and product manufacturers that sell AI inside their own products.

The third point matters for companies in the United States, the UK or Asia. A U.S. company that runs an AI chat agent for European customers can fall under the Act without an office in the EU. Providers of high-risk systems based outside the EU must also appoint an authorized representative in the EU before they sell there (Article 22).

Some uses are outside the Act. It does not cover AI used only for military, defense or national security purposes, and a person using AI for purely personal, non-professional reasons is not a deployer.

Provider, deployer, importer or distributor

Most duties attach to a role. The definitions are in Article 3, and the same company can be a provider for one system and a deployer for another.

Roles can change. Under Article 25 (opens in a new tab), a deployer, distributor or importer becomes the provider of a high-risk AI system if it puts its own name on it, makes a substantial modification to it or changes the intended purpose of a system so that it becomes high-risk.

This rule matters for companies that build AI agents on a general-purpose AI system. If you configure such a system to filter job applications, you can become the provider of a high-risk system, with the full set of provider duties.

The risk tiers of the AI Act

The Act sorts AI by the risk of the use, not by the technology. The European Commission describes four levels: unacceptable risk, high risk, transparency risk and minimal risk. General-purpose AI models have their own rules.

Prohibited practices (Article 5)

Since February 2, 2025, Article 5 (opens in a new tab) has banned these practices in the EU:

• Manipulative or deceptive techniques that cause significant harm
• Exploiting vulnerabilities linked to age, disability or a social or economic situation
• Social scoring that leads to unjustified or unrelated harmful treatment
• Predicting crime based only on profiling or personality traits
• Untargeted scraping of facial images to build recognition databases
• Emotion recognition at work and in education, except for medical or safety reasons
• Biometric categorization to infer sensitive traits such as race, religion or sexual orientation
• Real-time remote biometric identification in public spaces for law enforcement, with narrow exceptions

The Digital Omnibus adds two bans from December 2, 2026: AI systems that generate intimate images of an identifiable person without consent, and AI systems that generate child sexual abuse material. Our guide to the Digital Omnibus on AI explains how they work.

Minimal risk

AI that falls into none of the other tiers is minimal risk. The AI Act adds no specific rules for it beyond the AI literacy duty. Other laws still apply, such as the GDPR for personal data.

High-risk AI systems

High-risk systems carry the heaviest duties. There are two groups. The first is AI that is a safety component of products that already need safety checks under EU law, such as medical devices, toys or lifts (Annex I). The second is AI used in the sensitive areas listed in Annex III:

• Biometrics, such as remote biometric identification and emotion recognition
• Critical infrastructure
• Education and vocational training, such as admissions and exam scoring
• Employment, such as recruiting, filtering applications and evaluating staff
• Access to essential services, such as credit scoring, life and health insurance pricing and emergency call triage
• Law enforcement
• Migration, asylum and border control
• Justice and democratic processes

An Annex III system is not high-risk if it only performs a narrow procedural or preparatory task and does not materially influence decisions, under the conditions in Article 6(3). A system that profiles people stays high-risk. The provider must document this assessment and register the system in the EU database.

Providers of high-risk systems need risk management, data governance, technical documentation, logging, human oversight, accuracy and cybersecurity measures, a quality management system and a conformity assessment. Deployers must follow the instructions for use, assign trained people to oversee the system, keep its logs for at least 6 months and inform workers before using it at work (Article 26).

These rules now apply from December 2, 2027 for Annex III systems and August 2, 2028 for Annex I products. The Digital Omnibus on AI (opens in a new tab) moved them from the original dates of August 2, 2026 and August 2, 2027.

Transparency duties under Article 50

Article 50 (opens in a new tab) has applied since August 2, 2026. For most companies that use AI agents, it is the tier that matters today:

• Providers must design AI systems that interact with people so that those people are told they are dealing with an AI, unless this is obvious from the context.
• Providers of generative AI must mark synthetic audio, images, video and text in a machine-readable way. Systems already on the market before August 2, 2026 have until December 2, 2026.
• Deployers of emotion recognition or biometric categorization systems must inform the people exposed to them.
• Deployers must disclose deepfakes, and AI-generated text published to inform the public on matters of public interest, unless a person has reviewed the text and holds editorial responsibility.

The information must be clear and given at the latest at the first interaction. Our guide to Article 50 covers each duty in detail, and the AI disclosure checklist turns them into steps for voice and chat agents.

General-purpose AI models

General-purpose AI (GPAI) models, such as the large language models behind most AI agents, can perform a wide range of tasks. Since August 2, 2025, their providers must keep technical documentation, give information to companies that build on the model, have a copyright policy and publish a summary of the training content (Article 53). Models with systemic risk have extra duties (Article 55).

The Commission can enforce these duties, including with fines, from August 2, 2026. Models placed on the market before August 2, 2025 must comply by August 2, 2027, according to the Commission's guidance for GPAI providers (opens in a new tab). If you build agents on a model from an AI provider, the model duties sit with that provider. You still need the documentation it must give you, because your own duties depend on it.

Key dates of the EU AI Act

The table shows the dates as they stand after the Digital Omnibus on AI (Regulation (EU) 2026/1744 (opens in a new tab)) entered into force on July 27, 2026.

For a short summary of what the 2026 delay moved and what it left in place, read what still applies after the Digital Omnibus.

The AI literacy duty (Article 4)

Article 4 has applied since February 2, 2025, and the Digital Omnibus rewrote it. Providers and deployers must now take measures to support the AI literacy of their staff and of other people who operate or use AI systems on their behalf. The new text states that they do not have to guarantee a specific level of literacy for any individual.

The duty covers providers and deployers of all AI systems, not only high-risk ones. The measures should fit the staff's knowledge and training, the context of use and the people the AI affects. The Commission's AI literacy Q&A (opens in a new tab) says no certificate is needed, suggests keeping an internal record of training and confirms that national authorities supervise the rule from August 2, 2026.

In practice this means short training for each role, written rules for AI tool use and a record of who completed what.

Penalties under Article 99

Article 99 (opens in a new tab) sets the maximum fines. Each Member State sets its own penalty rules within these limits:

For SMEs and start-ups, each fine is capped at the lower of the two amounts. The Digital Omnibus extends this lower cap to small mid-cap enterprises for the two lower tiers. The Commission fines GPAI providers directly under Article 101, up to EUR 15 million or 3%.

A 6-step EU AI Act compliance plan for AI agents

This plan fits a company that uses or builds voice, chat, sales or back-office agents. Start now even if you have no high-risk system, because Articles 4, 5 and 50 already apply.

1. Build an AI inventory

List every AI system you build, buy or let staff use: what it does, who it talks to, which data it reads and which actions it can take. Include AI features inside your CRM or help desk, and tools staff use without approval ("shadow AI"). Note where users and outputs are, because output used in the EU can bring the Act into play.

2. Classify each system

Check each entry against Article 5, then Annex I and Annex III, then Article 50. In most cases, AI agents for support, booking or sales fall under Article 50, not the high-risk rules. Agents that filter job applicants, score credit or price life or health insurance can be high-risk. Record your reasoning. Our EU AI Act risk checker gives a first indication.

3. Assign roles and owners

For each system, write down your role. Watch the Article 25 triggers that turn you into a provider. Name one person who owns the compliance file for each system.

4. Put transparency in place

Make each agent tell people it is an AI at the first interaction, and label deepfakes and AI-generated public-interest text you publish. If you provide generative AI, check that outputs carry machine-readable marking. The Commission published a voluntary code of practice on marking and labeling AI-generated content (opens in a new tab) on June 10, 2026.

5. Document what you have

Keep a short file per system: purpose, role, risk class and reasoning, data sources, human oversight, logs, test results and the disclosure text. For high-risk systems, this grows into the technical documentation of Article 11 and the risk management system of Article 9.

6. Review vendor terms

Ask each AI vendor about its role under the Act, the documentation it provides, how it meets Article 50 marking and where it processes data. For high-risk systems, Article 25(4) requires a written agreement with suppliers of AI components on the information and technical access you need.

Our AI security consulting starts with the inventory and the classification, because every later step depends on them. Before launch, we attack-test each agent's disclosure, data handling and permissions and document the results for your compliance file.