How does the AI Act apply to machinery after the Digital Omnibus?

Machinery left the AI Act's direct high-risk regime on 27 July 2026

The AI Act machinery regulation question changed shape on 27 July 2026, when Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force. Its Article 1(41) made two edits to Annex I of the AI Act: in Section A, point 1, which listed the old Machinery Directive 2006/42/EC, "is deleted", and in Section B a new point 21 was added for Regulation (EU) 2023/1230 on machinery.

That move matters because of what Section B means. The first sentence of Article 2(2) of the AI Act, as amended, reads: "For AI systems classified as high-risk AI systems in accordance with Article 6(1) related to products covered by the Union harmonisation legislation listed in Section B of Annex I, only Article 6(1), Article 60a and Articles 102 to 112 shall apply." Its second sentence adds that Articles 57, 58 and 59 on sandboxes "shall apply only in so far as the requirements for high-risk AI systems under this Regulation have been integrated in that Union harmonisation legislation".

Recital 42 of the amending regulation gives the reason. The two instruments "might lead to overlaps", and, "Given the specific nature of machinery and the machinery sector", it found it "appropriate to move to a sectoral approach". The Digital Omnibus guide covers this as one of four headline changes; this page works through what it puts in place.

Where the AI requirements went: Article 3 of Regulation (EU) 2026/1744

The Omnibus is titled as amending three regulations, and the third is the Machinery Regulation itself. Article 3 of Regulation (EU) 2026/1744 makes three amendments to Regulation (EU) 2023/1230.

A duty to write AI requirements into Annex III. New paragraphs added to Article 8 of the Machinery Regulation require the Commission to "adopt delegated acts in accordance with Article 47 of this Regulation to amend Annex III to this Regulation by adding health and safety requirements in respect of Artificial Intelligence (AI) systems that are classified as high-risk pursuant to Article 6(1)" of the AI Act, where they are a safety component of, or are themselves, a product covered by the Machinery Regulation. The content is fixed by reference: "Those requirements shall ensure that the relevant requirements set out in Chapter III, Section 2, and Articles 17, 19, 72 and 73 of Regulation (EU) 2024/1689 are reflected." The Commission must also "ensure a level of protection consistent with that Regulation", and "Those delegated acts shall apply by 2 August 2028."

A presumption of conformity borrowed from the AI Act. New Article 20(10) provides that until harmonised standards or common specifications on high-risk AI are referenced or adopted under the Machinery Regulation, high-risk AI systems that comply with the relevant harmonised standards or common specifications under Articles 40 and 41 of the AI Act "shall be presumed to be in conformity with the essential health and safety requirements set out in Annex III to this Regulation as regards high-risk AI systems".

A new delegated power. Article 47(2) of the Machinery Regulation, as replaced, confers the Article 8 power on the Commission "for a period of five years from 27 July 2026", subject to the usual two-month objection period for Parliament and Council in Article 47(6).

Read together, the list in Article 8 tells you what survives the move: the Section 2 requirements (risk management, data and data governance, technical documentation, record-keeping, transparency, human oversight, accuracy, robustness and cybersecurity), the Article 17 quality management system, Article 19 log retention, Article 72 post-market monitoring and Article 73 serious incident reporting. What the Omnibus fixed is the date by which they must apply, not the date by which the Commission must adopt the act, so check the Commission's delegated acts register for the Article 8 act before planning against its wording.

The Machinery Regulation already regulates machine learning

The delegated act is not the first AI rule in the Machinery Regulation. Regulation (EU) 2023/1230 applies from 20 January 2027 under Article 54, a date fixed by the corrigendum published in OJ L 169 of 4 July 2023, which replaced the originally printed 14 January 2027. From that date it already contains provisions written for learning systems.

• Third-party conformity assessment for learning safety functions. Annex I Part A 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 with embedded systems of that kind "that have not been placed independently on the market, in respect only of those systems". Article 25(2) allows only module B plus C, module H or module G for Part A categories, all of which involve a notified body.
• Risk assessment over the learning lifecycle. Annex III Part B, point 1, requires the risk assessment to include hazards "foreseeable at the time of placing the machinery or related product on the market as an intended evolution of its fully or partially self-evolving behaviour or logic".
• Human-machine interaction. Annex III section 1.1.6 requires the interface to be adapted to operators of machinery "with intended fully or partially self-evolving behaviour or logic", and, where relevant, the machinery to communicate "its planned actions (such as what it is going to do and why) to operators in a comprehensible manner".
• Control systems. Annex III section 1.2.1 requires control systems of such machinery not to "cause the machinery or related product to perform actions beyond its defined task and movement space", and requires recording of data on the safety-related decision-making process, retained for one year.

Recital 55 of the Machinery Regulation confines the third-party provisions on safety software to "systems with a fully or partially self-evolving behaviour using machine learning approaches ensuring safety functions" and excludes "software incapable of learning or evolving". Under Article 25 of the Machinery Regulation, a product reaches a notified body in two ways: its category is listed in Annex I Part A, or it is listed in Part B and was not built to harmonised standards or common specifications covering all the relevant essential requirements. Either way it is then required to undergo a third-party conformity assessment, which is the condition in Article 6(1)(b) of the AI Act.

Is your machinery AI high-risk? The narrowed safety component test

Article 2(2) keeps Article 6(1), so classification still matters. An AI system is high-risk on the Annex I route where it is a safety component of a product, or is itself a product, covered by Annex I, and that product "is required to undergo a third-party conformity assessment". The Omnibus narrowed both limbs with an amended definition and three new paragraphs.

1. Is it a safety component? Article 3(14) as amended defines it as a component "which fulfils a safety function for that product or AI system, or the failure or malfunctioning of which endangers the health and safety of persons or property", and a component fulfils a safety function "where its intended purpose is to prevent or mitigate risks to health and safety of persons or property".
2. Is it only a non-safety function? Article 6(1a): AI systems "solely used for non-safety related aspects of user assistance, performance optimisation, service efficiency, automation or convenience or quality control shall not qualify as safety components".
3. Would its failure endanger health and safety anyway? Article 6(1b) overrides the previous step: such systems "shall qualify as safety components".
4. Is third-party assessment required for safety reasons? Article 6(1c): a product that needs third-party assessment "solely due to risks other than risks to health and safety, in particular risks relating to the distribution of radio spectrum or electromagnetic interference that do not affect health and safety" is not considered as fulfilling Article 6(1)(b).

Recital 7 of the amending regulation adds that "The mere fact that an AI system is integrated into or operates within a product that is subject to Union harmonisation legislation does not, in itself, mean that it fulfils a safety function."

A hypothetical, to show how the steps run. A collaborative robot uses a machine-learning vision model to detect people and trigger a protective stop. That model's intended purpose is to mitigate a risk to health and safety, so it is a safety component under Article 3(14). Whether it falls in Annex I Part A of the Machinery Regulation depends on whether it has fully or partially self-evolving behaviour: if it does, point 6 covers it as an embedded system, while recital 55 keeps software "incapable of learning or evolving" outside the third-party provisions. A second model on the same robot tunes cycle times for throughput. That is performance optimisation under Article 6(1a), unless its failure could endanger health and safety under Article 6(1b), which is the question to document. Article 2(2) covers only systems classified under Article 6(1), so whether a machinery-related system also falls in an Annex III use case is a separate test. The classifier runs both routes.

Section A against Section B: what applies

The carve-out is limited to Section B. The eleven Section A acts are listed in the FAQ below. Section B now holds civil aviation security, two- and three-wheel vehicles, agricultural and forestry vehicles, marine equipment, rail interoperability, motor vehicle approval and type-approval, unmanned aircraft under Regulation (EU) 2018/1139, and machinery.

Two Section A points need flagging. First, Article 2(13) is new: the application of specific requirements in Articles 9 to 15 and 17 to 25 "may be limited" where Section A legislation provides "an equivalent or higher level of protection" and the limitation "does not reduce the overall level of protection". It is not self-executing. "By 2 August 2027, the Commission shall adopt delegated acts" specifying the systems, the requirements, the conditions and the scope. Until such an act exists, nothing is limited. Second, Article 43(3) as amended keeps the sectoral procedure as the vehicle and lets existing Section A notified bodies assess Section 2 compliance, but requires them to "apply for designation in accordance with Section 4 of this Chapter by 28 January 2028". The conformity assessment article works through those routes.

Article 60a: real-world testing for Section B products

Apart from Article 6(1), the final provisions and the conditional reference to the sandbox articles, Article 60a is the only AI Act article that Article 2(2) applies to Section B. Its title is "Testing of high-risk AI systems covered by Union harmonisation legislation listed in Section B of Annex I in real-world conditions outside AI regulatory sandboxes". Under paragraph 1, Member States "may allow" testing by providers or prospective providers of AI-enabled Section B products "with a view to assessing and verifying the conformity of those systems with the requirements laid down in Articles 8 to 15".

The conditions are in paragraphs 2 to 6. A Member State that allows it must adopt a framework, alone or jointly, and notify the Commission "before implementing it". The framework must include "a mandatory real-world testing plan" agreed between the provider or prospective provider and the national competent authority or relevant authority under the Section B legislation, under paragraph 5(a), must ensure compliance with Article 60(2), (3), (4)(d) to (j) and (5) to (9), and must "ensure a high level of protection of health safety and fundamental rights". Paragraph 6 makes the testing subject to the sectoral legislation, which does not override Article 60a to the extent needed to enable the testing.

Before planning a test of AI-enabled machinery in public space, check whether the Member State concerned has notified a framework under Article 60a(3).

Dates and what to check now

The full timeline keeps these alongside the Annex III date of 2 December 2027. For a machinery manufacturer or a supplier of AI safety components, five checks follow from the text.

1. Classify each AI function separately against Article 3(14) and Article 6(1a) to (1c), and record why a function is or is not a safety component. The answer decides whether Article 6(1) applies at all.
2. Map machine-learning safety functions to Annex I Part A points 5 and 6 of the Machinery Regulation. If they are listed, a notified body is involved from 20 January 2027 regardless of the AI Act.
3. Build to the Article 8 reference list now. The delegated act must reflect Chapter III Section 2 and Articles 17, 19, 72 and 73, so documentation, logging, post-market monitoring and serious incident reporting processes designed to the AI Act text will map across.
4. Track the AI Act harmonised standards. Under Article 20(10) of the Machinery Regulation they carry a presumption of conformity for machinery until machinery-specific AI standards exist.
5. Do not extend the carve-out to other products. A product under the Radio Equipment Directive, the Lifts Directive or the Medical Devices Regulation stays in Section A, and the high-risk systems guide and the explainer on what counts as high-risk apply to it in full.

The carve-out also does not reach AI that is not high-risk under Article 6(1). Article 2(2) applies only to systems "classified as high-risk AI systems in accordance with Article 6(1)", so machinery AI outside that classification remains subject to the rest of the AI Act on its own terms, including Article 5 and, where its conditions are met, Article 50.