Imagine this: you have been a warehouse forklift driver for years, relying on instinct and experience to sense danger and steer clear of it. Recently, however, you were trained to operate a vehicle that uses artificial intelligence (AI) to detect whether a collision is imminent and, if so, to engage the emergency braking system. One day, a colleague steps out unexpectedly from behind a stack of boxes. The system fails to recognise the person and the emergency brake never engages. There is no warning explaining why the system failed, and no record that would allow the driver or a safety representative to establish whether the fault lay with the sensors, the training data, or the AI system itself. You slam on the brakes just metres before impact. AI did not save your colleague — you did.

This scenario illustrates the growing importance of artificial intelligence in safety-critical machinery — and the consequences when such systems fail. It is precisely why AI-specific safeguards matter. Workers and safety representatives need to know that safety-critical systems have been properly tested, that they perform reliably in real working conditions, that failures can be traced and investigated, and that effective human oversight remains possible. AI is becoming an integral part of machinery, increasingly performing functions once carried out by mechanical safeguards or human judgement. It is no longer an optional add-on; it now performs a growing number of safety-critical functions. If that AI fails, workers can suffer serious injury or death. Where AI is responsible for protecting workers, ensuring that it meets robust safety requirements is not an administrative burden; it is a necessity. That is why the decision to remove the Machinery Regulation from Annex I, Section A of the AI Act is so concerning.

Agreed in May 2026 and now in force, lawmakers in the European Union (EU) adopted the “Digital Omnibus on AI”. It was designed, among other things, to provide “simpler, innovation-friendly rules for artificial intelligence”, and as a consequence it removed machinery products, and AI-powered safety components in machinery, from the direct scope of the AI Act and its strict high-risk requirements. The AI Act establishes legally binding obligations designed specifically for AI systems, covering issues such as data governance, bias, cybersecurity, and quality management; the Machinery Regulation, as it currently stands, contains no equivalent AI-specific obligations. With the reference to the Machinery Regulation deleted from the AI Act, those obligations no longer apply to AI safety components in machinery.

The ostensible reason for this purge was to avoid duplication between the Machinery Regulation (EU) 2023/1230 and the AI Regulation (EU) 2024/1689, but in effect it temporarily leaves machinery manufacturers without an equivalent AI-specific legal regime to govern safety functions. To address the regulatory gap, the European Commission has launched a proposal to amend the Machinery Regulation to include requirements on high-risk AI systems in machines. Through a so-called “delegated act”, the Commission will have two years to incorporate into the Machinery Regulation the essential health and safety requirements applicable to AI safety components, without having to redraft the entire Regulation.

AI safety left stranded at the roadside

How did we get here, and what pre-empted this about-turn on AI safety? The proposal followed sustained lobbying by sections of the machinery industry, which called on lawmakers to streamline compliance requirements as part of a broader drive to boost European competitiveness. But in clearing the way for the Digital Omnibus, lawmakers have left AI-specific safeguards for machinery stranded at the roadside. Rather than continuing under the AI Act’s directly applicable high-risk regime, those safeguards have been diverted into the Machinery Regulation — a regulatory lay-by where they will remain until replacement rules are finally ready.

These protections cannot sit in that lay-by until August 2028, with workers and manufacturers left watching the clock.

At the European Trade Union Confederation (ETUC), we are concerned about this unnecessary fragmentation of the EU’s AI safety architecture. The push for streamlined compliance to boost Europe’s productivity must not become an excuse to weaken worker protections. The AI Act already contained mechanisms to prevent administrative duplication, so this decoupling creates an avoidable safety risk for workers. Removing machinery from the Act’s directly applicable high-risk regime creates uncertainty over how — and how completely — its safeguards will be transferred to the Machinery Regulation.

Until specific harmonised standards are developed under the Machinery Regulation for high-risk AI systems, manufacturers will be able to rely on the relevant standards developed under the AI Act. This should provide some continuity and greater certainty during the transition. But standards cannot replace the law. They can explain how manufacturers should meet a legal requirement, yet they cannot create a protection that the legislation itself does not guarantee. They can provide a bridge between the two regimes; they cannot fill gaps in the legal foundations beneath them.

This uncertainty is a serious concern for workers and manufacturers alike. Nobody yet knows precisely how every relevant high-risk requirement of the AI Act will be reflected in the Machinery Regulation, or how the new framework will operate in practice. Workers need a firm guarantee that the transfer will not produce a watered-down version of the existing safeguards. Manufacturers, for their part, need to know which requirements will apply, how compliance will be assessed, and whether further changes may follow. The previous arrangement may not have been perfect, but replacing clear and directly applicable requirements with a regime whose details are still to be developed inevitably creates uncertainty. The Commission must therefore act well before the 2028 deadline, providing a clear timetable, concrete milestones, and full transparency about how each relevant safeguard will be carried across.

Weaker protection for workers is never the answer

As trade unions, we are also concerned about the potential narrowing of the definition of a “safety component”. The AI Act currently classifies as high-risk only those AI systems intended to function as safety components and embedded in products subject to third-party conformity assessment. As AI becomes ever more embedded in functions such as obstacle detection, collision avoidance, and pedestrian recognition, any further restriction of this definition could leave systems that perform critical safety functions — such as the AI-supported forklift braking system described in our introduction — outside the high-risk category. That would allow products to be placed on the market, or put into use, without being subject to the AI Act’s minimum safeguards, resulting in weaker protection for workers.

It is true that AI is moving fast, and industry may argue that it develops so quickly that legislation safeguarding its use risks becoming outdated by the time it is adopted. For trade unions, however, this makes regulation more important, not less. The answer is not to leave safety unregulated but to build mechanisms that allow legislation to be updated quickly when necessary.

So, what happens next? The responsibility now lies primarily with the Commission. It must ensure that every relevant safeguard in the AI Act is fully and faithfully incorporated into the Machinery Regulation. The process needs a clear timetable, the involvement of trade unions and occupational-safety experts, and full transparency about how each requirement will be carried across. The European Parliament and the Council must then exercise their oversight powers to ensure that the final framework provides genuinely equivalent protection — with no omissions and no dilution.

This is not simply a technical exercise in moving requirements from one piece of legislation to another. It is a test of a fundamental principle: when new technology creates new risks at work, responsibility must remain with those who design, manufacture, and deploy it, not be transferred to the workers operating alongside it. Workers should not have to discover the limitations of an AI system in the seconds before an accident.

The Commission has promised protection consistent with the AI Act. It must now turn that promise into clear and enforceable rules. Simplification may change the regulatory route, but it must never change the destination: the highest possible level of safety for workers. AI may be moving fast; worker protection cannot be left behind.

  This post is sponsored by the ETUC