Four dates. One gap.

Four dates now govern AI safety in European machinery. Three of them moved in the last two weeks, and the press coverage described the change as simplification.

27 July 2026. Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force after publication in the Official Journal on 24 July. Among its effects: machinery moved out of Section A of Annex I of the AI Act and into Section B. In practice, robots and other machinery are no longer directly governed by the AI Act's high-risk technical obligations. The AI Act's transparency obligations were not deferred and took effect on 2 August 2026, but the technical requirements that would have applied to an autonomous machine's safety functions were lifted out.

20 January 2027. Regulation (EU) 2023/1230 replaces Machinery Directive 2006/42/EC. There is no overlap period. Machines and safety components whose safety functions rely on fully or partially self-evolving machine learning behaviour fall under Annex I Part A, which means notified body assessment — EU type-examination or full quality assurance — becomes mandatory. Self-declaration is no longer available for those categories.

A note for anyone working from a downloaded copy: the regulation as first published says it applies from 14 January 2027. A corrigendum in OJ L 169 on 4 July 2023 replaced fourteen dates throughout the act, including that one, moving the date of application to 20 January 2027. The original drafters counted from adoption; the corrigendum re-anchored everything to entry into force. If your compliance calendar came from a 2023 PDF, it is six days wrong in the wrong direction, and your notified-body designation date and penalty-notification date are wrong too.

2 December 2027. The AI Act's standalone high-risk obligations under Annex III, postponed from 2 August 2026.

2 August 2028. The deadline for the delegated act that will write AI-specific health and safety requirements into the Machinery Regulation. That act does not exist yet. It has not been drafted.

Now read them in order.

From 20 January 2027 until 2 August 2028 — eighteen months — an autonomous machine placed on the European market is regulated by a Machinery Regulation whose text was finalised before the AI provisions were written, and is no longer directly regulated by the AI Act that was supposed to cover its learned behaviour. The notified body assessment is real and mandatory from January. What that body assesses against, where AI is concerned, is a delegated act due nineteen months later.

This is not a loophole. Nobody hid it. It is a scheduling gap, published in the Official Journal, and it follows from two reasonable decisions taken separately: give machinery a single conformity route rather than two, and give the Commission time to write the AI requirements properly. Both are defensible. The interval between them is the problem.

Eighteen months is longer than most robotics companies take to ship a product.

It is worth being precise about what this does not mean. Machines are not unregulated in that window. The Machinery Regulation's general safety objectives still apply, and they are not weak — risk assessment, protective measures, the whole Annex III essential health and safety requirements set. A manufacturer who ships an unsafe autonomous machine in February 2027 is liable in exactly the way they would have been in January. What is missing is the specific technical layer: the AI-particular requirements that would tell a notified body what "adequate" looks like for a self-evolving safety function, and would tell a manufacturer what to build toward.

There is a second timing problem underneath. The Commission's own Guide to the application of the Machinery Regulation, whose first editorial group meeting was held in January 2025, is expected at the end of 2026 — that is, at best a few weeks before the regulation applies, and after every manufacturer has already had to decide what to build. The harmonised standards list has been amended twice in 2026, and there remain requirements with no cited standard behind them.

The composite picture: a regulation that applies in January, an application guide that arrives around the same time, a harmonised standards set with acknowledged gaps, and the AI-specific requirements deferred to a delegated act with a 2028 deadline.

I could not have read any of that four months ago. I would have taken the press release at face value; it says simplification, and simplification is a good thing. Learning to read Annex I Part A, the difference in legal force between a Directive and a Regulation, what "substantial modification" does to your obligations, why a notified body is not a certification body, and what a corrigendum can quietly do to your deadline — that was the point of the last six weeks.

The fine print is where robot safety actually lives.