How to Read the AI Act After the Omnibus

Regulation (EU) 2026/1744 answers the question raised at the beginning of this series in a structural way: the EU AI Act no longer has one effective date because a date now belongs to a particular provision, system, actor, and lifecycle state. This final synthesis brings those variables together. It also separates what the Digital Omnibus has settled from what still depends on standards, authorities, delegated acts, national frameworks, and facts about a real system.

Read the law on two levels: calendar and classification

The first level is the legal calendar. The amending regulation itself entered into force on 27 July 2026, three days after its publication on 24 July. That date tells us when Regulation (EU) 2026/1744 became part of the EU legal order. It does not make every amendment applicable at once.

The second level is the system or activity to which a provision attaches. The amended Article 113 keeps Chapters I and II applicable from 2 February 2025, subject to the stated exception for the new Article 5 provisions. Articles 102 to 110 apply from 27 July 2026. The new Article 5 provisions on specified non-consensual intimate material and child sexual abuse material apply from 2 December 2026. The Chapter III, Sections 1 to 3 high-risk requirements, subject to the text’s exception for Article 6(5), are split between 2 December 2027 for systems classified under Article 6(2) and Annex III, and 2 August 2028 for systems classified under Article 6(1) and Annex I.

Those dates are not competing opinions about when “the AI Act” begins. They are cumulative application states. The general framework can already apply while a particular high-risk obligation waits for 2027 or 2028. A new prohibition can apply in 2026 even though a product-related high-risk assessment waits until 2028. The amending regulation can be in force while a delegated act, standard, or national framework needed for practical implementation is still being prepared.

The date therefore answers only one part of the question. To know what it means, identify the classification route. The Annex III route is use-based: the system falls within a listed high-risk area such as a biometric, employment, education, critical-infrastructure, migration, asylum, or border-control use, subject to the final text and the actual facts. The Annex I route is product-based: the AI system is a safety component of a product governed by Union harmonisation legislation, and its function and failure consequences connect it to health, safety, or property. The Commission’s reference to lifts and toys helps illustrate the second route, but it is not an exhaustive classification decision.

That distinction makes the 2027/2028 split intelligible. The dates do not separate “old AI” from “new AI,” or software from hardware. They separate two legal relationships. A system can be high-risk because of a listed use. Another can be high-risk because it performs a safety function inside a regulated product. A provider must identify which relationship exists before selecting the date.

The product route also demonstrates why classification and conformity assessment must not be collapsed. The amended safety-component concept focuses on an intended function that prevents or mitigates health-and-safety risks. A system used only for convenience, user assistance, performance optimisation, service efficiency, automation, or non-safety quality control is not a safety component merely because it is installed in a regulated product. But if failure or malfunctioning would endanger health and safety, a provider’s convenience label cannot settle the matter.

Once a system is on the product route, sectoral coordination may reduce duplication under conditions. For Section A legislation, Article 2(13) permits specific AI Act requirements to be limited only when the sectoral law gives an equivalent or higher level of protection and the limitation does not reduce the AI Act’s overall protection. The Commission must identify the systems, requirements, conditions, and scope through delegated acts. Until that work identifies a particular case, the mere existence of a related product regulation does not prove that an AI Act duty has disappeared.

The machinery change makes the same point in a different form. Moving machinery from Section A to Section B of Annex I shifts more of the detailed AI safety interface into the machinery regime and limits the direct AI Act provisions in the way Article 2(2) specifies. It does not remove the safety objective.