EU changes timetable for high-risk AI systems

The EU's AI omnibus entered into force on 27 July 2026. The regulation changes parts of the EU's AI Act, which governs how AI may be developed and used in the European market.

One concrete change is the timetable for some high-risk AI systems. These are systems that can affect, for example, recruitment, training, credit checks or access to important services. The requirements for these systems will come into effect on December 2, 2027.

For AI that is built into regulated physical products, such as machines, toys and elevators, the corresponding rules will come into effect on August 2, 2028. The European Commission links the longer deadlines to the need for standards and national supervisory functions to be completed.

The omnibus also changes other parts. Certain simplified obligations are extended from small and medium-sized companies to so-called small mid-cap companies. The requirement for companies to ensure AI knowledge is replaced by a non-binding call, while the Commission and member states are given greater responsibility for promoting the knowledge.

At the same time, a ban is introduced against AI systems that create sexually explicit or intimate material without consent, as well as material that shows sexual abuse of children. The regulation also gives the EU's AI office extended supervisory responsibility for certain systems.

This is not a blanket suspension of the entire AI Regulation. Bans on certain AI uses came into effect on February 2, 2025, and rules for general purpose AI models came into effect on August 2, 2025. Other parts have their own dates.

A Swedish business should therefore not assume that everything has been postponed. The right question is whether the business develops or uses a system covered by a certain category, what role the business has according to the law and what date applies to that particular obligation.

Därför spelar det roll

More companies gain time to prepare documentation, risk management and liability for high-risk AI systems. However, the longer schedule does not remove rules that already apply or other obligations with their own dates. A wrong interpretation can leave a company waiting for work that should already be underway.

Det här kan du göra

1. Map which AI systems you develop or use and who makes decisions using them.
2. Check whether a system can be classified as high risk and which application date applies to that particular category.
3. Document responsibility, data, human control and risk management even if a certain deadline has been moved.