What the AI Act says about testing

A dated summary with links to the official texts: what testing it requires, from when, and who assesses conformity. Checked on 2026-11-14.


What the AI Act is

The EU AI Act is Regulation (EU) 2024/1689, amended by Regulation (EU) 2026/1744, in force since 27 July 2026.

What testing it requires

For high-risk systems, Article 9 requires testing against probabilistic metrics and thresholds defined in advance (Article 9(6) and (8)) and suited to their purpose, and the technical documentation must include dated and signed test reports (Annex IV, point 2(g)). These requirements will apply from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I products.

Who assesses conformity

For Annex III systems, points 2 to 8, the provider assesses conformity through internal control, without a notified body. A test like this one provides evidence the provider can use in that assessment.

Assistants that talk to people

If the assistant converses with people, the obligation to tell them they are talking to an AI (Article 50) applies from 2 August 2026.

Who decides whether a system is high-risk

Whether a system is high-risk is a legal question (Article 6 and Annex III of the Regulation) for your legal adviser to assess. I do not classify it.

If the system is not high-risk

If the system is not high-risk, Article 9 does not bind it, and this test serves as good practice.

Harmonised standards and guidance

The reference of the first harmonised standard, EN 18286:2026 (quality management system), was published in the Official Journal on 14 October 2026 (Implementing Decision (EU) 2026/2274); it gives a presumption of conformity with the Article 17 requirements it covers. The Spanish AI agency's (AESIA) guides of December 2025 are not binding and predate the 2026 amendment.