AI Embedded in Regulated Products Gets Until August 2028

The AI Act deferral agreed this summer has two halves, and the second matters most to anyone who builds physical products. Alongside the move of stand-alone Annex III systems to 2 December 2027, obligations for high-risk AI embedded in products already regulated under Annex I have been deferred to 2 August 2028, a twelve-month extension.

Annex I is the list of existing EU product legislation that already governs safety: the Machinery Regulation, the Medical Devices and In Vitro Diagnostic Regulations, the Lifts Directive, the Radio Equipment Directive, the Toy Safety Directive, pressure equipment, and others. Where AI performs a safety function inside one of these products, or is itself a safety component, the AI Act's high-risk duties layer on top of the sectoral rules rather than replacing them.

That layering is the reason the longer deferral was granted. A manufacturer in this position is not writing one conformity assessment but reconciling two regimes, often with the same notified body, and the sectoral standards being revised to accommodate AI requirements are on their own timelines. Machinery manufacturers in particular face a Machinery Regulation transition running in parallel.

The staggered structure produces an awkward middle period. Between December 2027 and August 2028, a stand-alone system performing a given function faces the full high-risk regime while the same function embedded in a CE-marked machine does not. Suppliers who sell both a boxed product and an integrated module will find their two lines governed differently for eight months.

Two dates that were previously one now need tracking separately, and which one applies depends on how a product is sold rather than what its software does.