A high-risk AI system is, within the meaning of Article 6 of Regulation (EU) 2024/1689, known as the AI Act, an AI system that falls into one of two categories. The first, in paragraph 1, brings together systems linked to a product covered by the Union harmonisation legislation listed in Annex I. The second takes up one sentence of paragraph 2: “In addition to the high-risk AI systems referred to in paragraph 1, AI systems referred to in Annex III shall be considered to be high-risk.” Paragraph 3 adds a bounded derogation.
Article 3, devoted to definitions, contains no entry for this notion. We read here Article 6 and its annexes in the consolidated version of 27 July 2026, subsequent to Regulation (EU) 2026/1744. This reading classifies no system: the qualification is a matter for each organisation and its lawyers.
The first route runs through an Annex I product
Paragraph 1 sets out “both of the following conditions”, which are cumulative. Under point (a), the system is “intended to be used as a safety component of a product” covered by an Annex I act, or is itself such a product. Under point (b), that product is subject to a third-party conformity assessment pursuant to the same act.
Annex I has two sections: Section A, which includes for example the acts on toys or lifts, and Section B, in which Regulation (EU) 2026/1744 entered Regulation (EU) 2023/1230 on machinery. For systems linked to Section B products, Article 2(2) provides that “only Article 6(1), Article 60a, and Articles 102 to 112 shall apply”.
The 2026 regulation clarified the notion of safety component (Article 3, point 14). Paragraph 1a of Article 6 excludes from it systems “used solely for aspects not related to safety”, aspects it lists. Paragraph 1b brings back into it those “whose failure or malfunctioning would endanger health and safety”. Paragraph 1c bounds point (b): a product assessed by a third party “solely because of risks other than risks to health and safety” does not satisfy it.
The second route refers to the eight areas of Annex III
Under each area, Annex III describes use cases, whose recurring formula is “AI systems intended to be used”.
| Point | Area | Use cases covered, among others |
|---|---|---|
| 1 | Biometrics | Remote biometric identification |
| 2 | Critical infrastructure | Safety components in the management of road traffic |
| 3 | Education and vocational training | Admission, evaluation of learning outcomes |
| 4 | Employment, workers’ management and access to self-employment | Recruitment and selection, promotion, termination |
| 5 | Access to and enjoyment of essential private services and essential public services and benefits | Evaluation of the creditworthiness of natural persons |
| 6 | Law enforcement | Evaluation of the reliability of evidence |
| 7 | Migration, asylum and border control management | Examination of applications for asylum, visas and residence permits |
| 8 | Administration of justice and democratic processes | Assisting judicial authorities in interpreting facts or the law |
This table abridges: the qualification is made on the full wording of the annex, which carries its own exclusions.
The formula “intended to be used” echoes Article 3, point 12, which defines intended purpose as “the use for which an AI system is intended by the provider”, specified in particular by the instructions for use.
A derogation in paragraph 3, bounded by profiling
“By derogation from paragraph 2”, a system referred to in Annex III is not considered to be high-risk where it “does not pose a significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the outcome of decision making”. The second subparagraph applies it where one of its four conditions is met. The system is intended:
- to perform a narrow procedural task;
- to improve the result of a previously completed human activity;
- to detect decision-making patterns or deviations from prior decision-making patterns, without replacing or influencing the previously completed human assessment without proper human review;
- to perform a preparatory task to an assessment relevant for the purposes of the Annex III use cases.
The third subparagraph closes the derogation in one case: a system referred to in Annex III “shall always be considered to be high-risk where the AI system performs profiling of natural persons”. Article 3, point 52, refers for this term to Article 4, point 4, of the GDPR, Regulation (EU) 2016/679.
The derogation concerns paragraph 2 alone, and therefore Annex III alone. It has a counterpart, in paragraph 4: the provider relying on it documents its assessment before placing on the market or putting into service, and remains subject to registration under Article 49(2).
Provider and deployer facing a high-risk AI system
Among operators, Article 6 names only the provider, in its paragraph 4. Article 16 lists the obligations of the provider of a high-risk system, starting with compliance with the requirements of Section 2 of Chapter III (Articles 8 to 15); Article 26, those of the deployer, defined in Article 3, point 4. Our article on the AI Act obligations for businesses presents them with their timeline.
Article 25(1), point (c), considers as provider of a high-risk system the deployer who modifies the intended purpose of a system not classified as high-risk in such a way that it becomes high-risk. We draw one consequence from this, which is our own reading: an organisation using a purchased system has an interest in comparing its use with the intended purpose written in the instructions.
The terms Article 6 leaves to be qualified
Several terms carry the decision without appearing among the definitions in Article 3: “significant risk of harm”, “materially influencing”, “narrow” procedural task. Under paragraph 5, the Commission provides, “no later than 2 February 2026”, guidelines together with “a comprehensive list of practical examples of use cases of AI systems that are high-risk and not high-risk”.
Delegated acts may also amend the conditions of the derogation (Article 6(6) and (7)) and Annex III (Article 7).
Dates of application and upkeep of the classification
Under point (c) of the third subparagraph of Article 113, Sections 1, 2 and 3 of Chapter III, except Article 6(5), apply from 2 December 2027 for systems classified as high-risk under Article 6(2) and Annex III, then from 2 August 2028 for those classified under Article 6(1) and Annex I.
We propose a method, which the regulation does not prescribe: a classification sheet per system records the route examined, the annex point concerned, the derogation condition invoked, the answer on profiling and the version of the text read. It is reopened when the intended purpose, the regulation or the guidelines change.
Our page on AI governance describes the register and the classification grid that hold these sheets; our article on implementing AI governance, the stage at which they are drawn up. Each sheet thus carries two dates: its own, and that of the version of the regulation read.