AI in manufacturing: overlap with the Machinery Regulation

Short answer: For AI in manufacturing the central question is whether the AI system is a safety component of a product already covered by European product safety legislation. If so, Annex I of the AI Act classifies the system as high-risk, and the new Machinery Regulation applies at the same time. This is not a choice between two rule sets but two conformity tracks meeting on one product.

Annex I: the link to product safety

The AI Act has two routes to high-risk. Annex III lists standalone applications (such as recruitment or critical infrastructure). Annex I does something different: it makes an AI system high-risk when it is a safety component of a product covered by listed sectoral legislation — and the Machinery Regulation is on that list. An AI that decides when a robot arm stops or a press locks is therefore a high-risk system.

The Machinery Regulation

The Machinery Regulation (Regulation (EU) 2023/1230) replaces the old Machinery Directive and applies from early 2027. New is that it addresses AI-driven and self-learning safety functions explicitly. Machines with evolving behaviour, with fully or partly self-learning systems, and with software updates that change safety behaviour face specific requirements. The regulation demands that the manufacturer controls the risks of that learning behaviour throughout the whole lifecycle.

One product, two tracks

The hard part is alignment. The AI Act requires a risk management system, technical documentation, data quality, logging and human oversight for the AI part. The Machinery Regulation requires its own risk assessment and essential safety requirements for the machine as a whole. The CE marking must cover both. In practice the conformity assessment is therefore integrated: a single technical file that satisfies both regimes, so you do not prove the same thing twice with different outcomes.

Predictive maintenance and quality control

Not every factory AI is high-risk. Predictive maintenance, vision-based quality inspection or process optimisation are often not a safety component and then fall outside Annex I. The test stays functional: if the system co-determines the safety of persons, it tips into high-risk. The same mechanism appears in other sectors, such as AI in energy.

What to do

• Classify per function: is the AI system a safety component or not?
• Align the two files: integrate the AI Act documentation with the Machinery Regulation risk assessment.
• Control learning behaviour: record how you manage the risks of self-learning or updatable systems.
• Plan around timelines: high-risk obligations and the Machinery Regulation do not run exactly in step; check application dates.
• Involve your notified body early if external conformity assessment is needed.

For industrial AI the core mistake is treating the AI Act and machine safety as separate tracks. It is one product with two converging requirement sets.