A machinery maker selling into the EU should book a notified body for every safety function that uses self-evolving machine learning before 20 January 2027, because the Machinery Regulation, which applies from that date, lists those functions among the categories assessed by a notified body, and the AI requirements due for such machines by 2 August 2028 will be added to that same regulation.
Regulation (EU) 2023/1230 shall apply from 20 January 2027, as a corrigendum of July 2023 set it, and Member States shall not impede the making available of machines placed on the market in conformity with the Machinery Directive before then. Its Annex I, Part A names safety components with fully or partially self-evolving behaviour using machine learning approaches ensuring safety functions and, in respect only of those systems, machinery embedding such systems that have not been placed independently on the market. Article 25(2) allows three procedures for both, EU type-examination followed by conformity to type, full quality assurance or unit verification, and a notified body takes part in each. The regulation does not define self-evolving; a recital says the third-party provisions should not apply to software incapable of learning or evolving, and programmed only to execute certain automated functions.
The Digital Omnibus on AI, Regulation (EU) 2026/1744, in force since 27 July 2026, moved the Machinery Regulation to Section B of the AI Act's Annex I, where for high-risk AI systems only Article 6(1), Article 60a and Articles 102 to 112 shall apply. Under an amended Article 8 of the Machinery Regulation, the Commission shall adopt delegated acts adding requirements for those AI systems to its Annex III, reflecting the AI Act's high-risk requirements and its articles on quality management, logs, post-market monitoring and serious incidents. Those delegated acts shall apply by 2 August 2028.
From January the risk assessment shall include hazards foreseeable at placing on the market as an intended evolution of self-evolving behaviour in machinery designed to operate with varying levels of autonomy. Control systems of such machinery shall keep the machine within its defined task and movement space, record safety related decision data and retain it a year for a national authority's reasoned request, and allow the machine to be corrected at all times. Where sensor data controls safety related operations on sensor-fed, remotely-driven or autonomous machinery, the technical documentation describes, where appropriate, the system's capabilities and limitations and its data, development, testing and validation processes. A camera that stops a palletising cell when a person steps in, on a model that keeps learning on site, belongs on the list: protective devices designed to detect the presence of persons are in the regulation's indicative list of safety components.
In the first month the head of engineering lists every safety function on machines to be placed on the EU market from 20 January 2027 and marks those using self-evolving machine learning. In the second, the product compliance manager writes the risk assessment and documentation entries for each. In the third, the chief operating officer signs with a notified body; an operating partner can ask each portfolio machinery company for that list.
The list can stop at a component its supplier placed on the market independently, and at any function that serves no protective measure, the purpose by which the regulation defines a safety function. Until standards or common specifications for high-risk AI are referenced or adopted under the Machinery Regulation, a system meeting the AI Act's relevant ones shall be presumed to conform to its Annex III requirements for high-risk AI systems, under the Omnibus's new Article 20(10).
