EU AI Act Explained: Obligations, Timeline and Fines for Businesses

The EU AI Act explained for businesses: how the risk categories work, what providers and deployers must do, how the timeline has shifted after the Digital Omnibus agreement, and the fines for getting it wrong.

Abstract illustration of connected nodes representing artificial intelligence systems

Key points

  • The AI Act regulates AI systems according to risk: prohibited, high-risk, limited-risk with transparency duties, and minimal risk.
  • Bans on certain practices and an AI literacy duty have applied since February 2025; rules for general-purpose AI models since August 2025.
  • EU legislators reached a provisional agreement in May 2026 to postpone the main high-risk obligations to December 2027 and August 2028.
  • The Act applies to organisations outside the EU whose systems, or their outputs, are used in the EU.

Regulation (EU) 2024/1689, the Artificial Intelligence Act, entered into force on 1 August 2024. It is the first comprehensive AI law from a major jurisdiction and it applies in stages. Most companies are affected not as developers of AI but as users of it, which the Act calls deployers. Knowing which role you hold, and which risk category each system falls into, is the starting point.

The risk categories

CategoryExamplesConsequence
Unacceptable riskSocial scoring; manipulative techniques causing harm; untargeted scraping of facial images; emotion recognition in workplaces and schools (with exceptions); certain biometric categorisation and predictive policingProhibited
High riskAI as a safety component of regulated products; systems listed in Annex III, such as recruitment and worker management, credit scoring of individuals, life and health insurance pricing, access to education and essential services, and some law-enforcement and migration usesExtensive obligations before and after placing on the market
Limited riskChatbots; systems generating synthetic text, images, audio or video; deepfakesTransparency: people must be told they are interacting with AI, and synthetic content must be marked
Minimal riskSpam filters, inventory forecasting, most business analyticsNo specific obligations; voluntary codes

Separate rules cover general-purpose AI (GPAI) models, such as large language models. Their providers must publish technical documentation and summaries of training content and respect copyright, with additional duties for models posing systemic risk.

Who has obligations

  • Providers develop an AI system or have one developed and place it on the market under their own name. They carry the heaviest duties for high-risk systems: risk management, data governance, technical documentation, logging, human oversight, accuracy and robustness, conformity assessment and registration.
  • Deployers use an AI system in a professional capacity. For high-risk systems they must follow the instructions for use, assign competent human oversight, monitor operation, keep logs, inform workers and affected individuals and, in some cases, carry out a fundamental rights impact assessment.
  • Importers and distributors must verify that systems carry the required conformity evidence.

A deployer becomes a provider if it puts its own name on a high-risk system, substantially modifies one, or changes the purpose of a system so that it becomes high-risk. Fine-tuning a model for recruitment screening is an example worth legal attention.

The timeline

DateWhat applies
1 August 2024Entry into force
2 February 2025Prohibited practices; AI literacy duty (Article 4)
2 August 2025Obligations for providers of general-purpose AI models; governance and penalties provisions
2 August 2026General date of application, including most transparency duties. This was also the original date for Annex III high-risk systems.
2 December 2027Annex III high-risk obligations, under the Digital Omnibus agreement
2 August 2028High-risk AI embedded in regulated products (Annex I), under the Digital Omnibus agreement
The December 2027 and August 2028 dates reflect the provisional political agreement on the Digital Omnibus on AI announced in May 2026. Confirm the final text in the Official Journal before relying on them.

The postponement responds to delays in the harmonised standards that providers need to demonstrate conformity. It does not change the substance of the high-risk regime, and procurement teams at large customers are already asking suppliers about readiness.

Penalties

  • Prohibited practices: up to €35 million or 7% of worldwide annual turnover, whichever is higher.
  • Most other obligations, including those for high-risk systems: up to €15 million or 3%.
  • Supplying incorrect information to authorities: up to €7.5 million or 1%.

For small and medium-sized enterprises the lower of the two amounts applies.

What every business should do now

  1. Build an inventory of AI systems in use, including features embedded in HR, finance, CRM and other software, and tools staff have adopted on their own.
  2. Classify each use by role and risk category. Pay particular attention to recruitment, performance management, credit and insurance decisions.
  3. Check for prohibited uses, which have been unlawful since February 2025.
  4. Meet the AI literacy duty. Staff who operate or use AI systems must have sufficient understanding for their role. Record the training provided.
  5. Ask suppliers how they classify their products and what documentation they will supply.
  6. Adopt a usage policy. See writing a generative AI policy.
  7. Align with data protection. The AI Act sits alongside the GDPR, which already restricts solely automated decisions with legal or similarly significant effects.

The United Kingdom

The UK has no equivalent statute. Its approach relies on existing regulators applying cross-sector principles, with data protection, equality, consumer and financial services law doing most of the work. UK companies fall under the EU Act when they place AI systems on the EU market or when the output of their systems is used in the EU.

Frequently asked questions

Does the AI Act apply to a company that only uses tools such as chatbots?

Yes, as a deployer. For most everyday uses the duties are light: AI literacy, transparency where relevant, and avoiding prohibited practices.

Is using AI to screen job applicants high-risk?

Yes. Recruitment and selection systems are listed in Annex III. Deployers must also inform candidates and worker representatives.

Are small businesses exempt?

No, though the Act includes simplified documentation, regulatory sandboxes and lower fine ceilings for SMEs.

The bottom line

The delay to the high-risk rules gives more time, not a reason to wait. An inventory, a classification and a literacy programme are low-cost steps that are already expected, and they are the foundation for everything that follows.

Important: This article is general information, not legal advice. Application dates were subject to legislative amendment at the time of writing; check the consolidated text on EUR-Lex.

Sources and further reading

Scroll to Top