Glossary · EU AI Act: high-risk AI

Conformity assessment (AI Act)

Also known as: conformity assessment, CE marking for AI, harmonized standards

Under the EU AI Act, conformity assessment is the process of demonstrating that a high-risk AI system meets the requirements of the Act before it is placed on the market or put into service. Depending on the system, it is carried out by the provider through internal control or with a notified body, and it ends with an EU declaration of conformity and CE marking.

AI regulation
EU

In one sentence

AI Act conformity assessment demonstrates that a high-risk AI system meets the requirements, by internal control or with a notified body, before market entry.

Example

The provider of an AI recruitment tool performs internal control under Annex VI, signs the EU declaration of conformity, affixes the CE marking and registers the system in the EU database.

How it applies

Procedures: For most Annex III systems, the provider assesses conformity itself (internal control, Annex VI). For biometric systems, a notified body is involved unless harmonized standards have been fully applied (Annex VII). AI in Annex I Section A products is assessed in the procedure of the product legislation — for example the medical device conformity assessment — which then covers the AI requirements too. Machinery is in Section B: its AI functions are assessed under the Machinery Regulation, where Article 20(10) lets AI Act harmonized standards support conformity until machinery-specific ones exist.
Harmonized standards: Systems that follow harmonized standards published in the Official Journal are presumed to conform. The European standards bodies CEN and CENELEC are developing these standards in their joint technical committee JTC 21; their delay was a main reason for postponing the high-risk rules.
Result: An EU declaration of conformity, the CE marking and registration in the EU database. A substantial modification requires a new assessment.
Technical documentation: The technical documentation is the evidence of the assessment. Document why each requirement is met and where the proof can be found.

Compared with the USA, Canada and China

There is no general pre-market assessment for AI in the USA or Canada outside regulated sectors such as medical devices, where the FDA and Health Canada review AI-enabled devices. China requires a security assessment and filing before a generative AI service with public-opinion attributes goes online, but this is a review by the regulator rather than a conformity assessment against product requirements.