Your AI Act deadline just changed

For 9 months, your compliance team has been working toward August 2, 2026, and last month, the EU moved that deadline 16 months out.

Most of the plan is now wrong.

Before we get into what changed, quick question:

Where is your plant on EU AI Act compliance?

👋🏻 I'm Leonardo Ubbiali. This week we are looking at the EU Digital Omnibus on AI deal, and what most compliance teams still have not noticed about the 16-month gift to industrial operators.

The Digital Omnibus on AI provisional deal landed in the early hours of May 7.

The Carve-out

Most of the coverage has been about the deadline shift but the bigger change for plants is the carve-out underneath.

If an AI component "merely assists users or optimises performance" without creating a health or safety risk, it is no longer in scope.

Process optimization, predictive maintenance, design assistance, engineering copilots no longer trigger high-risk obligations under the narrower definition.

Safety functions in machinery, worker biometrics, and employment decision systems still do.

This deal has a name on it.

Chancellor Friedrich Merz and Siemens CEO Roland Busch made the case at Hannover Messe in April.

Busch told Bloomberg the EU was treating industrial and machine data the same way as personal data, and that Siemens would direct most of its €1 billion industrial AI investment to the United States and China instead.

The EU moved.

Siemens has been pointing at Erlangen as the model.

The Erlangen Electronics Factory was announced at CES 2026 as the first blueprint for a fully AI-driven adaptive manufacturing site, with an AI Brain running the digital twin and pushing validated changes to the floor.

Nanjing followed in January as a WEF Lighthouse Factory: 50 AI applications, lead times down 78 percent versus 2022, time-to-market down 33 percent.

All of those tools sit outside the new high-risk definition.

Every compliance team I have spoken to in the last three weeks is still treating their full inventory of shop-floor AI tools as high-risk.

Almost none of it is, under the new definition.

Every shop-floor AI tool gets inventoried and re-mapped against the narrowed definition.

Once a tool is moved out of high-risk, the self-exemption gets documented and the system still gets registered in the EU database.

That registration obligation survived the deal.

The compliance budget earmarked for the August deadline needs a new home.

Move it in Q2 and the next 18 months go into building real capability, or leave it where it is and you spend that time on paperwork toward a deadline that has already moved.

Five things you can do this quarter

The problem: You need to know which of your shop-floor AI tools shipping into the EU still count as high-risk after the May 7 Digital Omnibus deal.

What you need: A list of every AI tool used in your plant or shipped in your machinery, what each one does, whether failure could create a health or safety risk, and which markets the tool ships into.

The Prompt (copy this):

❝

I'm a [YOUR ROLE] at a [FACILITY TYPE] manufacturer shipping into the EU. The following AI tools are in use across our plant and our shipped machinery: [LIST EACH TOOL, FUNCTION, AND WHETHER FAILURE COULD CREATE HEALTH OR SAFETY RISK].

Tell me: Which tools are likely OUT of high-risk under the May 7 EU Digital Omnibus carve-out for AI that "merely assists users or optimises performance"? Which are still in high-risk under Annex III or Annex I? What documentation do I need to defend a reclassification decision? What still needs to be ready by August 2, 2026 versus December 2, 2026 versus the new December 2, 2027 deadline?

What you'll see:

A reclassification map of your AI tools, a documentation checklist for defending each call, a deadline timeline for what still takes effect this year, and a contract update checklist for your AI vendors.

TRY IT