EU opens consultation on draft guidance for high-risk AI

The European Commission has published draft guidance on how to classify high‑risk artificial intelligence systems and has opened a public consultation that runs until June 23. The document aims to help users determine when an AI system should be treated as high‑risk under the EU AI Act.

The guidance is intended to help providers, deployers and other actors decide whether an AI system is a safety component of a product or can significantly affect health, safety or fundamental rights. It clarifies relevant provisions of the AI Act and includes practical examples showing how to assess classification in different sectors and use cases.

The draft is divided into two parts. The first part covers AI systems that are products themselves or safety components of products governed by sectoral rules, including the Machinery Regulation, the Toys Safety Regulation, the Radio Equipment Directive, medical device rules and the automotive and aviation regimes. The second part covers AI systems listed in Annex III of the AI Act that can significantly affect people’s health, safety or fundamental rights in specific use cases.

Article 6(3) is a central focus. The draft restates a four-condition carve‑out: the AI must perform a narrow procedural task; improve the result of a previously completed human activity; detect decision‑making patterns or deviations without replacing or influencing the human assessment; and perform a preparatory task for an Annex III assessment. The guidance also states that any Annex III system that performs profiling of natural persons is always high‑risk.

Patrick Sullivan, vice president of strategy and innovation at A‑LIGN, warned that many providers will struggle to apply Article 6(3). He added that the carve‑out and the Annex III exception have direct implications for human resources systems, credit decisioning and systems that build individual predictive profiles.

The guidance was published later than expected; it had been due in February. It follows political agreement on an AI Omnibus that set a new enforcement timeline: rules for systems used in areas such as biometrics, critical infrastructure, education, employment, migration, asylum and border control take effect on December 2, 2027, while rules for AI integrated into products, including robotics and industrial machinery, come into force on August 2, 2028.

The Commission invites contributions from AI providers and developers, organisations that use AI systems, public authorities, researchers, civil society organisations, supervisory bodies and members of the public. The consultation is open until June 23. The draft links classification decisions to existing sectoral rules and explains how Annex III should be interpreted in relation to profiling and other activities.