Guidelines for high-risk AI systems

Co authors: Ramona Ahmadi, Julius Haas

On May 19, 2026, the European Commission published the Draft Commission Guidelines on the classification of high- risk AI systems.[1]

The guidelines are intended to help companies, public authorities and other responsible parties assess more easily whether an AI system qualifies as a high-risk system under Article 6 of the AI Act.[2] This shall ensure that a consistent application and enforcement of the AI Act across the EU. The guidelines themselves do not create any new obligations, but they clarify how the existing obligations under the AI Act should be applied.

When is AI in machines/robots considered “high-risk”? When it comes to AI in machines and robots, two main categories are typically considered:

1. Safety component of a product (Annex I AI Act)

AI is considered high-risk if:

• it is intended to be used as a safety component of a machine, or
• constitutes a “product” itself covered by the relevant harmonisation legislation (e.g., the Machinery Directive) and
  the product itself is required to undergo a third-party conformity assessment.[3]

2. Use cases from Annex III of the AI Act

AI systems also carry a high risk if they fall into one of the application areas listed there, such as:

• Critical infrastructure,
• Safety-critical control of systems,
• Access to essential private and public services,
• Education and employment.[4] 

Outlook

The provisional agreement must now be approved by the Council and the European Parliament and formally adopted by the legislative bodies. EU lawmakers aim to finally adopt the legislation before August 2, 2026.[5]

This is intended to ensure that the new rules are in place in time before the current regulations for high-risk AI systems take effect (in this regard, please see our newsletter on the Digital Omnibus). This will provide businesses and government agencies with a more predictable regulatory framework, but at the same time, they will need to adapt to the new requirements early on.

Practical Note

Manufacturers and suppliers of AI for machines are advised to begin

• systematically assessing and documenting whether AI functionalities should be classified as safety components of a product or as products themselves
• mapping use cases against Annex III of the AI Act to identify high-risk classifications at an early stage
• integrating existing conformity assessment processes (EU Machinery Regulation) with AI requirements (see also our newsletter on the Digital Omnibus / political agreement of May 2026),
• with risk assessment and technical documentation in a manner that complies with both product safety and AI regulations.

Especially in borderline cases (e.g., assistance functions vs. safety-critical control systems), early legal classification is crucial for market access and for reducing liability risks.