olivlawolivlaw

From books

An artificial intelligence management system is not built beside the security one but inside it: the same clauses, a different risk register.

Parlamentul European și Consiliul Uniunii Europene · Regulamentul (UE) 2024/1689 de stabilire a unor norme armonizate privind inteligența artificială · 2024 · Regulamentul (UE) 2024/1689 (regulamentul privind inteligența artificială), art. 9 alin. (1) — sistemul de management al riscurilor2 minutes readpublic domain
“A risk management system shall be established, implemented, documented and maintained in relation to high-risk AI systems.”Parlamentul European și Consiliul Uniunii Europene · Regulamentul (UE) 2024/1689 de stabilire a unor norme armonizate privind inteligența artificială · 2024 · Regulamentul (UE) 2024/1689 (regulamentul privind inteligența artificială), art. 9 alin. (1) — sistemul de management al riscurilor

The harmonized structure allows one management system with two risk registers, not two parallel systems.

Regulation (EU) 2024/1689 applies in tiers. The prohibited practices and the general provisions apply from 2 February 2025, the duties for general-purpose models from 2 August 2025, and the regulation's general application from 2 August 2026. For the high-risk systems of Annex III, the "digital omnibus" package deferred the obligations to 2 December 2027 — the state of play in September 2026. Article 9 requires a risk management system established, implemented, documented and maintained across the whole life cycle of the high-risk system: an iterative process, not a folder. This is where the harmonized structure of management system standards pays off. ISO/IEC 42001:2023, the standard for artificial intelligence management systems, uses the same clauses 4-10 as ISO/IEC 27001, so context, roles, internal audit and management review are run once. What is added is specific: the inventory of systems, the impact assessment on people, training data, human oversight.

Earlier passages showed that the two systems share the same management clauses. But sharing clauses does not mean one document. Article 9 requires its own risk register for artificial intelligence, separate from the security one. A security risk says: someone steals your customer data. An artificial intelligence risk says: your lending system systematically rejects applications from one neighbourhood. Both live in the same annual review, at the same table, but with different vocabulary. In practice, you meet the team once and discuss two registers. Whoever merges them loses each one's specificity; whoever fully separates them doubles the meetings. The next step: who should own each register in the organisation.

Why it matters Whoever treats artificial intelligence compliance as a separate project pays twice for exactly the same management clauses.

2 February2025 —2 August2025 —2 August2026 —2 December2027 —
The calendar as it stands in September 2026.

← Back to the feed