The EU Omnibus Carve-Out: Less Duplication, Not Less Urgency

The EU’s Digital Omnibus on AI has been described by many as a delay to the AI Act. For some companies, that may be true. For machinery and robotics manufacturers, the picture is more complicated.

The key point is this: the EU Machinery Regulation date has not moved. It still applies from 20 January 2027. What has changed is how AI-related safety obligations for machinery are expected to be handled. Instead of applying directly through the AI Act for certain embedded AI systems, those requirements are expected to be routed through the Machinery Regulation by delegated acts.

That may reduce duplication. But it does not remove the work.

The carve-out changes the route, not the destination

Before the Omnibus agreement, an AI-enabled machine could potentially fall under both the Machinery Regulation and the AI Act high-risk framework. That raised the possibility of overlapping conformity routes, duplicated technical documentation, and multiple assessments. 

The carve-out is intended to simplify that. AI systems embedded in machinery are expected to be removed from direct AI Act application, with AI-specific health and safety requirements introduced instead through the Machinery Regulation. 

That sounds cleaner, and in many ways it is.

But for manufacturers, it also means the Machinery Regulation becomes even more important. It is no longer just the machinery safety framework. For AI-enabled machinery, it becomes the primary route for demonstrating that intelligent, adaptive, or autonomous behavior has been addressed safely.

Why this is not a reason to wait

The difficult part is timing.

The Machinery Regulation application date remains fixed at 20 January 2027, while detailed AI-related safety requirements under delegated acts are still expected to follow.

That creates an uncomfortable planning window. Manufacturers cannot wait for every interpretive detail before starting their compliance work. The core safety questions are already clear enough to act on:

• What AI functions can influence safety?
• How does the machine behave when the AI is wrong, degraded, or unavailable?
• What evidence will show that risks have been identified and reduced?
• How will updates, learning, or behavioral changes be controlled after placing on the market?

Those questions do not disappear because the legal route has changed. If anything, they become more central.

The real issue is evidence

AI-enabled machinery creates risks that are not always visible through traditional testing alone.

A robot may behave correctly in one environment and differently in another. A vision system may perform well under normal lighting but degrade under glare, dust, or obstruction. A decision-making function may be safe most of the time, but fail in edge cases that were not considered during design.

That is why the technical file matters so much.

Manufacturers will need to show not just that the machine works, but that they understand the limits of its behavior. They will need evidence that AI-related hazards have been considered, tested, documented, and controlled across the machine lifecycle.

This is where the carve-out should not be misunderstood. It may simplify the regulatory pathway, but it does not simplify the engineering burden.

How Intertek can support

Final thought

The Omnibus carve-out should be seen as a clarification, not a pause.

For machinery manufacturers, the message is simple: the route may be changing, but the destination is still safe, compliant machinery by January 2027.

Waiting for perfect certainty is risky. Starting with the safety case now is the better path.