AI Omnibus in force on 27 July: machinery AI moves into the Machinery Regulation

Regulation (EU) 2026/1744 of 8 July 2026 — the Digital Omnibus on AI — has been in force since 27 July 2026, published in the Official Journal on 24 July: article 4 sets it on “the third day following that of its publication”. The title states the reach: it amends three regulations, (EU) 2024/1689 (the AI Act), (EU) 2018/1139 (civil aviation) and (EU) 2023/1230 (the Machinery Regulation).

One effect is immediate: article 1(40)(c) adds a point (d) to article 113 of the AI Act — “Articles 102 to 110 shall apply from 27 July 2026”, the provisions that graft AI onto sectoral product legislation. But the consequence almost nobody will join up concerns machinery builders.

The two lines that move machinery

Article 1(41) amends Annex I to the AI Act: “in Section A, point 1 is deleted” — that was Directive 2006/42/EC — and “in Section B, the following point is added: 21. Regulation (EU) 2023/1230”. Article 1(2)(a) then rewrites article 2(2) of the AI Act: for high-risk AI systems related to products covered by Section B legislation, “only Article 6(1), Article 60a and Articles 102 to 112 shall apply”. Chapter III — the substantive requirements — no longer applies directly to that machinery.

This is not an exemption: article 6(1) still classifies as high risk an AI system that is a safety component of an Annex I product — Section A or B — subject to third-party conformity assessment. What changes is the text that writes the requirements. Recital 42 gives the official rationale: applying the two regulations together “could lead to overlaps”, while a level of protection “consistent” with the AI Act must be ensured; hence, “given the specific nature of machinery”, the choice to “move to a sectoral approach”.

Where the requirements end up

In Annex III to the Machinery Regulation, the document every engineering office already works from. Article 3(1) adds two subparagraphs to article 8 of Regulation (EU) 2023/1230: the Commission “shall adopt delegated acts […] to amend Annex III […] by adding safety and health protection requirements” for AI systems “classified as high-risk pursuant to Article 6(1)” of the AI Act because they are safety components of a product covered by that regulation, or are themselves such a product. The content is constrained: those requirements must ensure “compliance with the relevant requirements set out in Chapter III, Section 2, and in Articles 17, 19, 72 and 73” of the AI Act — high-risk requirements, quality management, automatically generated logs, post-market monitoring, serious-incident reporting. And they apply “by 2 August 2028”.

One file, one route, one body

Annex I, Part A of the Machinery Regulation lists, at point 5, “safety components with fully or partially self-evolving behaviour using machine learning approaches ensuring safety functions”, and at point 6 machinery embedding such systems where they have not been independently placed on the market. Part A means article 25(2): EU type-examination followed by conformity to type (modules B and C), full quality assurance (H) or unit verification (G). Internal production control is not an option: the notified body was already there.

The practical point. Once the AI requirements sit in Annex III they become essential requirements like any other: verified inside that procedure, documented in the technical file of Annex IV, Part A referred to in article 10(2). One file, one route, one counterpart, instead of two parallel stacks. And the body will have to read them: article 30(7)(c) requires its personnel to have “adequate knowledge and understanding of the essential health and safety requirements set out in Annex III”. It will not run through the new Annex XIV the Omnibus adds to the AI Act: the “AIP” notified-body codes refer back to Section A (recital 43) and the list starts at AIP 0102. For machinery there is no longer a code.

What is not yet written

The requirements. Today Annex III contains no line specific to high-risk AI: it will arrive by delegated act, and of that future text we know the perimeter, not the wording. Article 3(3) states that the delegation runs for five years “from 27 July 2026” and that Parliament and Council have two months, extendable by a further two, to object.

There is a bridge. Article 3(2) adds a paragraph 10 to article 20: until harmonised standards or common specifications for AI exist under the Machinery Regulation, systems complying with those adopted under articles 40 and 41 of the AI Act “shall be presumed to be in conformity” with the Annex III requirements on AI. And the new article 60a opens the way to testing in real-world conditions authorised by Member States.

One date recurs in the text near the machinery provisions, 2 September 2027: it does not concern them. Article 1(30) replaces article 72(3) of the AI Act and requires the Commission to adopt “guidelines, including a template, on the post-market monitoring plan by 2 September 2027” — a template that recital 41 makes voluntary. Since Annex III will have to cover article 72 as well, that document precedes the delegated acts: a reference point, not a deadline for manufacturers.

The same pattern, in aviation

Article 2 does not touch only article 27 of Regulation (EU) 2018/1139: it adds the same paragraph 3 to articles 27, 31, 32, 36, 39, 50 and 53 — when adopting implementing or delegated acts on AI systems that are safety components, “the requirements set out in Chapter III, Section 2, shall be taken into account”. Recital 44 describes this as “a technical correction” relative to article 108 of the AI Act.

The rest of the Omnibus sits elsewhere: the high-risk calendar — 2 December 2027 and 2 August 2028 — and the new article 4 on AI literacy.

What to do now

Three moves, possible before the delegated acts exist. First: take stock of which machines have machine learning inside a safety function: points 5 and 6 of Annex I. Second: read the technical file against Annex IV, Part A and mark where nothing today corresponds to risk management, data governance, logs and post-market monitoring: those empty boxes are the 2027 workload. Third: ask your notified body how it will build AI competence.

The work runs on drawings, bills of materials, risk assessments and manuals — the same documents already in play for instructions for use — that is, on the company’s trade secrets. That is why we run assisted document compliance analysis with dedicated, closed AI, detached from the open web, in two modes: on-premise in the client’s own environment, or on our dedicated cloud — reserved for the single client, with a dedicated VPN, the data centre in Italy and premises we staff ourselves. In both cases the design never leaves the perimeter: that is the principle behind our platform.

Do you know how many of your machines fall under points 5 and 6 of Annex I to the Machinery Regulation? Half an hour with one of our experts is enough for a first map.