Skip to content

Sovereignty and law · 5 min read · As of November 2026

The EU AI Act for mid-sized companies: what applies since 2026 and what to do

The EU AI Act takes effect in stages, and some stages shifted in 2026. We sort out the deadlines and say what you should tackle now.

The EU AI Act regulates AI by risk class. For a mid-sized company, much depends on its role as provider or deployer and on the risk class of the application. As soon as AI is in use, the AI Act usually has to be checked. What matters is which duties apply when and which risk class your own application falls into.

At a glance

  • Bans on certain practices and the duty of AI literacy have applied since 2 February 2025, and the duties for providers of general-purpose AI models (GPAI) since 2 August 2025.
  • A simplification package adopted in June 2026, the “Digital Omnibus”, moved the deadline for stand-alone high-risk systems from 2 August 2026 to 2 December 2027. The transparency duties and the start of enforcement were left untouched and have applied since August 2026.
  • ISO/IEC 42001 works as a roadmap to meet the documentation duties in an orderly way, whatever the exact state of the deadlines.

The deadlines after the Digital Omnibus

Date What applies
2 February 2025 Prohibited practices (Art. 5) and AI literacy requirements take effect
2 August 2025 Duties for providers of general-purpose AI models (GPAI) take effect
2 August 2026 Transparency duties under Art. 50 (among others the labelling of AI interactions and chatbots) and the start of enforcement and penalties apply as originally planned
2 December 2026 The transition period for technical marking (watermarking) of synthetic content and for technical safeguards against the prohibited practices added in 2026 (Art. 5) ends
2 December 2027 Duties for stand-alone high-risk AI systems under Annex III take effect (moved from 2 August 2026)
2 August 2028 Duties for high-risk AI systems embedded in products that are already regulated (Annex I) take effect

The Council of the European Union gave its final approval to the Digital Omnibus on 29 June 2026. It has been in force since 27 July 2026. The transparency duties have applied since August 2026. The deadline for high-risk systems was postponed. The information comes from legal analyses and reflects the position in November 2026. Before you decide anything, check the deadlines against the Official Journal of the EU or have them reviewed legally.

Determine the risk class

Not every AI application is “high risk”. The Act distinguishes prohibited, high-risk, limited-risk (with transparency duties) and minimal-risk applications. In mid-sized companies, many applications are low or limited risk, such as internal assistants or document search over your own files. What counts is the concrete use. The same language model can be minimal risk in one application and high risk in another, for example in automated hiring decisions.

Use ISO/IEC 42001 as a roadmap

ISO/IEC 42001:2023 is the first international standard for AI management systems. It provides a structure for responsibilities, risk assessment, data handling and documentation, and it overlaps in large parts with the documentation duties of the AI Act. It gives you a basis for classification and documentation, even if your own application only falls under a high-risk duty in 2027 or 2028.

Build the evidence into the architecture

Documentation, logging, data residency and access control belong in the technical solution. An architecture that records every model call (time, model used, result, failure) produces this evidence during normal operation and saves you from reconstructing it later. The same logging supports the traceability that the GDPR also asks for. More in the article AI and the GDPR in business.

What a mid-sized company should do now

  1. List all AI applications used in the company, including those that individual employees use without official approval.
  2. Assign each application to a risk class. The classification depends on the use and does not follow a model name.
  3. For applications that interact with people or produce content, check whether the transparency duties in force since August 2026 are met, for example labelling as an AI interaction.
  4. For applications that are likely to fall under the high-risk deadlines of 2027 or 2028, use the lead time and do not wait for the due date.
  5. Take a structure such as ISO/IEC 42001 as the frame and bundle individual measures within it.

How data control and traceability can be put into practice in daily operation is shown on the page about sovereign infrastructure.

Frequently asked questions

Does the EU AI Act affect small companies too?

Yes, depending on the application. Many use cases in mid-sized companies are low or limited risk. The duties follow the risk class and the use, whatever the company size.

Did the Digital Omnibus remove the high-risk duties?

No, it postponed them. For stand-alone high-risk systems under Annex III the new date is 2 December 2027, and for systems embedded in regulated products under Annex I it is 2 August 2028. The transparency duties under Art. 50 and the start of enforcement and penalties stayed at the original date in August 2026.

What is the quickest sensible first step?

A complete inventory of all AI applications actually used in the company, including unofficial use by employees, and a first classification by risk class. That lays the groundwork for every further measure, and a later postponement of a deadline does not change it.

What does ISO/IEC 42001 have to do with the EU AI Act?

The standard does not replace any legal duties. It is a practical roadmap that maps many of the proofs the AI Act asks for in an orderly way, such as risk assessment, data handling and documentation.

Questions on your EU AI Act classification?

We discuss with you how to classify your AI applications and which steps come first.