The AI Act and buildings: what Regulation (EU) 2024/1689 changes for an operator

Regulation (EU) 2024/1689, the AI Act, is the first horizontal framework governing artificial-intelligence systems in the Union. It does not regulate a technology, it regulates uses, ranked by risk: unacceptable (banned), high risk, limited risk (transparency duties) and minimal risk.

It binds both providers of AI systems and their deployers — in our field, the operators using a tool that embeds AI.

A building operator is almost never a provider under the regulation: they did not develop the system. They are a deployer.

That changes the nature of the duties: a deployer does not have to produce the model's technical documentation, but must know which AI uses run inside their tools, ensure use consistent with the instructions, guarantee effective human oversight, and inform people where required.

In practice the first audit question is not 'is your AI compliant?' but 'do you know where AI sits in your estate?' — and very few operators can answer it.

The original timeline has been amended, so be wary of the 2024 and 2025 tables still circulating.

• 2 February 2025 — banned practices and the AI literacy duty (Article 4): applicable.
• 2 August 2025 — general-purpose AI (GPAI) model duties and the governance strand: applicable.
• 2 August 2026 — general application of the regulation, including the Article 50 transparency duties.
• Annex III high-risk systems — the political agreement on the Digital Omnibus, reached between the Council and the Parliament on 7 May 2026, defers their applicability to 2 December 2027 (from 2 August 2026), and to 2 August 2028 for AI embedded in already-regulated products (Annex I).

Status as of 22 July 2026: the text was adopted by Parliament on 16 June, approved by the Council on 29 June and signed on 8 July 2026; it enters into force on publication in the Official Journal, which is imminent. The deferral touches neither the Article 50 transparency duties nor the AI literacy duty.

The topic feels remote until you run the inventory. In an equipped tertiary estate you routinely find:

• predictive maintenance and remaining-useful-life estimation;
• anomaly detection on sensor time series;
• setpoint optimisation for heating, ventilation and cooling;
• natural-language assistants over operational data;
• sometimes people counting or occupancy analytics — the most sensitive case, where personal data enters.

Most of these uses are limited or minimal risk. But having no inventory at all is a governance failure in every case.

Without waiting for the high-risk deadlines, a serious operator should be able to produce on request:

• the inventory of AI uses present in their tools, with each purpose;
• the risk classification retained for each and its justification;
• the human oversight measures in place;
• the transparency provided to occupants and staff where the use requires it;
• the traceability of AI-assisted decisions that had an operational effect.

This is the file public buyers and large accounts already ask for in tenders — well before any penalty applies.

Sensia Hypervision ships a module that inventories every AI use in the product — anomaly detection, maintenance recommendations, natural-language assistant, correlations — and maps each to the matching obligations of the regulation.

The output is a dashboard whose contents you reuse as-is in a governance review or a tender response. File export is on the roadmap.

Two design choices are worth flagging: inference runs on OVHcloud in France on open models, which makes the chain of custody easy to demonstrate; and the system abstains when it lacks sufficient evidence rather than producing an unverifiable recommendation.

• Regulation (EU) 2024/1689 of 13 June 2024 laying down harmonised rules on artificial intelligence — EUR-Lex
• The European Commission's AI Office
• Digital Omnibus on AI — political agreement of 7 May 2026, adopted by Parliament on 16 June, approved by the Council on 29 June, signed on 8 July 2026

This page is an explainer, not legal advice. The high-risk timeline takes effect on publication of the amending text in the Official Journal.