Minimal risk
Residual category with no definition in the Regulation: AI systems that are not prohibited by Article 5, are not classified as high-risk under Article 6, do not trigger the transparency obligations of Article 50 and are not general-purpose models under Chapter V. The expression does not appear in the articles: it is industry vocabulary.
It is where the vast majority of the tools a company uses sit, and where the obligations are few, but not zero.
Which obligations it carries
It has no classification of its own because it is not a category of the Regulation: there is no article, no date and no obligation attached to the category as such. What does apply to a system sitting here, exactly as to any other: Article 4 — taking measures to support the development of AI literacy — is a legal obligation of providers and deployers applicable from 2 February 2025, and it is a measures-based obligation, not one of result; and the prohibitions of Article 5 are a legal obligation applicable from 2 February 2025. Neither contains a risk qualification. And where the system interacts with people, generates synthetic content or constitutes a deep fake, the transparency obligations of Article 50 apply by virtue of that nature, whatever the label.
What it is not
It is not “unregulated”. Articles 4 and 5 do not distinguish by category, and Article 50 can be triggered by the nature of the system — interaction with people, synthetic content, deep fakes — even if the system is not high-risk. It cannot be attested or declared to anyone: there is no register, seal or minimal-risk notification. And it is not the same as the Article 6(3) exception, which operates within Annex III, requires the provider to document its assessment before placing the system on the market and subjects it to registration under Article 49(2). Confusing them turns a documentary obligation into an absence of obligation.
The nuance almost nobody captures
That the category does not exist in the Regulation has a practical consequence that is overlooked: nobody grants it to you, so your own documentation sustains it. If someone asks why a tool was not treated as high-risk, the useful answer is not “it was minimal risk” — which means nothing legally — but the dated reasoning of why it fits no Annex III area and not Article 6(1).