The European Commission published draft guidelines on the classification of high-risk AI systems under the EU AI Act on 19 May 2026, opening a stakeholder consultation that was originally due to close on 23 June. Following requests from industry associations and other stakeholders, the Commission has extended the consultation deadline to 23 July 2026, with the final guidelines expected to be adopted by the end of the year. The guidance is intended to help providers, deployers, and market surveillance authorities determine whether a given AI system falls within the high-risk category under Article 6 of the Act.
The draft guidelines address the two routes by which a system becomes high-risk: systems that are safety components of products covered by the Union's harmonised product-safety legislation, and systems deployed in the specific use cases listed in Annex III that can significantly affect health, safety, or fundamental rights. The guidance places considerable weight on the intended purpose of a system - the use intended by the provider as reflected in instructions for use, promotional material, and technical documentation - as the anchor for classification. It also offers vocabulary for the harder questions: the distinction between an AI system that merely supports a human workflow and one that materially influences a decision, and the treatment of safety components and human oversight.
One of the more consequential clarifications concerns complex systems. The draft guidelines state that where several AI components interact and their combined outputs materially influence a decision in a high-risk use case, they should be assessed holistically and may be treated as a single AI system for classification purposes. This matters for organisations that build composite products from multiple models or modules, because it prevents classification from being avoided by fragmenting a system into individually low-risk components.
The extended timeline gives organisations a useful window to assess their own systems against the draft framework and to feed back before the guidance is finalised. Given that classification determines the weight of obligations that attach to a system under the AI Act, getting the analysis right is foundational rather than incidental.
Acompli perspective: Classification is the gateway to the AI Act's obligations - get it wrong, and either an organisation over-invests in low-risk systems or, more dangerously, under-prepares a system that is in fact high-risk. The draft guidelines' emphasis on intended purpose and on assessing complex systems holistically should prompt organisations to look closely at how their AI products are described and how their components combine. Reviewing your risk assessments against the draft classification framework now, and ensuring your records of processing capture where and how AI systems are deployed, will leave you well placed whether or not your systems ultimately fall within the high-risk tier.