EU AI Act + Robotics: Annex II Machinery Regulation Interface

Main article: High-risk AI Annex III — it sets out the second classification route (Art. 6(2) AI Act) with all eight Annex III areas.

1. Interface between Machinery Regulation 2023/1230 and the AI Act 2024/1689

Short answer: for a robot with AI, two regulatory strands apply side by side — the sector law of Machinery Regulation (EU) 2023/1230, one of the Union harmonisation acts listed in Annex I to the AI Act, and the AI Act (EU) 2024/1689 itself, which via Art. 6(1) links to precisely that sector law; since Regulation (EU) 2026/1744 (in force since 27 July 2026), the AI Act obligations of this route apply from 2 August 2028 (Art. 113(3)(c)(ii) AI Act).

Machinery Regulation (EU) 2023/1230 applies from 20 January 2027 (Art. 54(2); 14 June 2023 is merely the date of adoption) — on that date Machinery Directive 2006/42/EC is also repealed. AI Act Annex I is legally binding from 2 August 2028 (postponed by Regulation (EU) 2026/1744, in force since 27 July 2026; Art. 113(3)(c)(ii) AI Act). Robotics with AI falls under both.

Annex I to the AI Act is not itself a catalogue of obligations but a list of Union harmonisation legislation. Only Art. 6(1) turns it into a classification: an AI system is considered high-risk where it is intended to be used as a safety component of a product covered by Annex I, or is itself such a product, and is subject to third-party conformity assessment. The reference point is therefore always the specific product, not the company. Practice recommendation: carry out the allocation per product and per AI function — for the same product line, a manufacturer can be both the provider of a high-risk AI system and the deployer of a system that is not high-risk.

2. High-risk classification

Short answer: a robot with an AI safety component is high-risk under Annex I to the AI Act only where both conditions of Art. 6(1) are met — (a): the AI system is intended to be used as a safety component of a product covered by the Union harmonisation legislation listed in Annex I (which includes the Machinery Regulation), or is itself such a product, and (b): that product, or the AI system as a product, must undergo a third-party conformity assessment. Without an obligation of third-party assessment, Art. 6(1) does not apply. Independently of this, a robot may be high-risk via Annex III (Art. 6(2)), for example as a safety component in the area of critical infrastructure (Annex III no. 2).

The two routes differ in their consequences. The exception in Art. 6(3) — no high-risk classification absent a significant risk to health, safety or fundamental rights, for example in the case of a narrowly defined preparatory task — is, by its wording, tied to the systems referred to in Annex III, and even there it does not apply where natural persons are profiled. Where classification runs via Annex III no. 2 (critical infrastructure), that very number is the sole exception from provider registration in the EU database under Art. 49(1) AI Act.

Practice recommendation for the allocation (four steps): describe the AI function and output per product line; check the Annex I allocation of the product; clarify the third-party assessment obligation under (b); record the role as provider (Art. 3 no. 3) or deployer (Art. 3 no. 4) and document it with reasons. Step 3 is answered not by the AI Act but by the sector law to which Art. 6(1)(b) links — for machinery, Machinery Regulation (EU) 2023/1230, linked in the source list at the end; which conformity assessment procedure it prescribes for your product category is not assessed in this article. In case of doubt, the higher classification is the more robust route — it costs effort, whereas too low a classification costs up to EUR 15 million or 3 % of worldwide annual turnover under Art. 99(4).

3. CE marking: one marking, two strands of requirements

Short answer: there is no double CE marking. Where a high-risk AI system is also covered by other Union legislation which likewise provides for CE marking, the (single) CE marking indicates that the system also meets the requirements of that other legislation (Art. 48(5) AI Act). Likewise, only a single EU declaration of conformity is drawn up, covering all applicable Union legal acts (Art. 47(3) AI Act). What is doubled, therefore, are the requirement and conformity assessment strands (Machinery Regulation + AI Act), not the marking. A Notified Body can be qualified for both strands at once (machinery + AI). Practice recommendation: make this dual qualification a selection criterion as soon as a Notified Body has to be involved for your product type — the procurement route is described in section 8. Whether, and under which procedure, such involvement arises in a given case depends on the respective strand and is not assessed in this article.

In practice this means: you maintain two strands of evidence but mark once. The AI strand requires technical documentation under Annex IV (Art. 11 AI Act) and a conformity assessment (Art. 43 AI Act); these obligations of the Annex I route apply from the date stated in section 6. Practice recommendation: set up the Annex IV content as a separate, clearly delimited part of the product file rather than scattering it through the existing machinery documentation — that keeps the AI part capable of being presented and updated in isolation vis-à-vis a Notified Body.

4. Safety functions

Short answer: where the AI function is classified as high-risk under Art. 6(1), the full set of requirements for high-risk AI applies: risk management system (Art. 9), data governance (Art. 10), technical documentation under Annex IV (Art. 11), record-keeping (Art. 12), transparency towards deployers (Art. 13), human oversight (Art. 14) as well as accuracy, robustness and cybersecurity (Art. 15).

For the safety part, the relevant product standards must also be considered depending on the robot type — ISO 10218 for industrial robots, ISO 13482 for personal care robots. Practice recommendation: record in the Annex IV file, per product line, which of the requirements under Art. 15 AI Act (accuracy, robustness, cybersecurity) you cover through the application of standards and which remain open as AI-specific issues. Whether any of these standards is harmonised for the purposes of the AI Act and therefore carries a presumption of conformity is not assessed in this article.

The catalogue does not end once the product is placed on the market: post-market monitoring (Art. 72) and the response to serious incidents (Art. 73) are added. Practice recommendation: set out in writing, per safety function, who may intervene, how the intervention is technically possible and how it is logged — human oversight is owed by Art. 14, while the specific arrangement is left to you and belongs in the Annex IV documentation. Also design the record-keeping under Art. 12 so that it technically permits the deployer's retention obligation of at least six months (Art. 26(6)) — provider and deployer sides must match at this point.

5. SME robot manufacturers

Short answer: for small and medium-sized manufacturers, the role is decisive first: the provider (Art. 3 no. 3) bears the system obligations from point 4 as well as quality management (Art. 17) and conformity assessment (Art. 43); the deployer (Art. 3 no. 4) bears the use obligations under Art. 26. Anyone who substantially modifies the purpose or functionality of a third-party AI system, or places it on the market under their own name, themselves becomes a provider under Art. 25 AI Act. For SMEs and start-ups, Art. 99(6) AI Act provides that the lower of the two values (amount or percentage of turnover) applies within the fine framework.

Those affected are above all drone manufacturers as well as manufacturers of robotic arms, cobots and service robots — for all four, what is decisive is not the product category but the allocation of role and Annex I status under section 2. The effort scales with the role:

Practice recommendation on budgeting: plan using the line for your role. The staff figures are experience values, not statutory requirements; external costs, in particular for a Notified Body, depend on the product and the procedure and can only be quantified reliably by obtaining quotes.

Practice recommendation: clarify the role contractually before you integrate technically. If you buy in an AI component and market the robot under your own name, Art. 25 applies — you then owe the provider obligations even though you did not train the model. Have the supplier deliver the information you need for the Annex IV documentation and the conformity assessment.

6. Transition periods

Short answer: the transition periods of the AI Act follow from Art. 113 AI Act, those of the Machinery Regulation from Art. 54(2) Regulation (EU) 2023/1230; the two dates decisive for robotics are already named in section 1, and until then Machinery Directive 2006/42/EC continues to apply. All rows of the table below state statutory dates of application — not practice recommendations; the preparatory steps for the AI strand are set out in the FAQ at the end of the article.

Independently of the Annex I date, the AI literacy obligation under Art. 4 has applied since 2 February 2025 — including for teams that develop or operate robotics AI. Under Art. 4 AI Act in the version applicable since 27 July 2026, providers and deployers must take measures to support the development of the AI literacy of their staff; no particular level of competence is owed. According to the Commission FAQ (non-binding), a certificate is not required; internal proof of training is sufficient. Breaches of Art. 4 do not fall within the fine framework of Art. 99(4); penalties are governed by national law (Art. 99(1) AI Act). On the Digital Omnibus: it has been in force as Regulation (EU) 2026/1744 since 27 July 2026; the postponement for Annex I to 2 August 2028 and that for Annex III to 2 December 2027 is therefore applicable law (Art. 113(3)(c) AI Act).

7. Obligations matrix: providers and deployers

Short answer: the provider owes the system requirements from point 4 — including risk management (Art. 9) and the technical documentation under Annex IV (Art. 11) — plus a quality management system (Art. 17) and the conformity assessment (Art. 43). The deployer checks the relevance of the input data (Art. 26(4)), keeps logs for at least six months (Art. 26(6)) and informs the workers concerned (Art. 26(7)).

All rows are statutory obligations of the AI Act; the trigger column states the norm-specific link of the respective row. For the Annex I route — that is, AI in machinery — they apply from the date stated in section 6. The rows that expressly link to Annex III apply, since Regulation (EU) 2026/1744, from 2 December 2027 (Art. 113(3)(c)(i) AI Act). The fundamental rights impact assessment (FRIA) under Art. 27 AI Act links to the Annex III route — the triggers are public bodies as well as Annex III no. 5(b) and (c); it does not follow from the Annex I route.

8. Germany and Austria: what differs nationally

Short answer: as regards the substance of the AI Act obligations, nothing — the AI Act is a directly applicable EU regulation, and Art. 6(1), Annex I, the provider and deployer obligations as well as all deadlines apply identically in Germany and Austria. Only the competent authorities (Art. 70 AI Act) and the penalty regime within the framework of Art. 99(1) AI Act are governed nationally — in Germany since 29 July 2026 by the German AI Market Surveillance and Innovation Promotion Act (KI-MIG) (§ 2(1): Federal Network Agency (Bundesnetzagentur), in so far as sectoral authorities are not competent; § 15(3): fines of up to EUR 50,000); as at the date of this article (17 August 2026), Austria has not designated a market surveillance authority under Art. 70 AI Act. This section does not assess the machinery strand.

Each Member State establishes or designates at least one notifying authority and at least one market surveillance authority; these exercise their powers independently, impartially and without bias. One market surveillance authority is designated as a single point of contact; the contact details are to be made publicly available under Art. 70(2) AI Act (deadline: 2 August 2025; Art. 70(1) and (2) AI Act).

In Germany, the market surveillance authority has, since 29 July 2026, been the Federal Network Agency (Bundesnetzagentur) (§ 2(1) KI-MIG), in so far as sectoral authorities — such as BaFin — are not competent; it is at the same time the single point of contact under Art. 70(2) AI Act (§ 6(1) KI-MIG). As at the date of this article (17 August 2026), Austria has not designated a market surveillance authority under Art. 70 AI Act; the AI service point of RTR is an information and advisory body without supervisory powers.

Practice recommendation: before 2 August 2028, clarify for each sales market, on the basis of these contact details, which market surveillance authority is the single point of contact and which Notified Body can cover both the machinery and the AI strand for your product type. Both require lead time that cannot be arranged at short notice.