The Omnibus nobody asked for

The Record, written in October 2026. A retrospective. This piece covers November 2025, when the Commission published its Digital Omnibus proposal, and was written afterwards from the documents. Nothing here was published at the time.

On 19 November 2025 the Commission published a package it called the Digital Omnibus. Alongside changes to data rules and to the GDPR, it contained a proposal to amend the AI Act — including the timetable for high-risk systems, which at that point were still due to be regulated from 2 August 2026.

The reaction split cleanly. Industry called it pragmatism. Civil society called it deregulation by delay. Both were describing the same document, and both missed the more interesting question: why a Commission proposes to slow down its own flagship regulation nineteen months after passing it.

The answer was not in the AI Act

By the autumn of 2025, three things were true at once.

The standards bodies had not produced the harmonised standards that high-risk compliance was supposed to be built on. Without them, conformity assessment was a procedure with no agreed method. An association's supplier could not certify against a standard that did not exist.

Member states had not finished standing up the market-surveillance authorities that were meant to enforce any of it. A rule with no enforcement is a suggestion.

And the sector that had to comply — including every member association with a recruitment or credit-scoring tool in its stack — had spent the year discovering that the obligations were larger than the summaries had implied.

A deadline that cannot be met by the regulator is not a deadline. It is a liability. The Omnibus was the Commission admitting that in a document with a legal base.

What it proposed to change, and what it did not

The headline was the deferral: high-risk obligations pushed back, with application dates decoupled from the availability of standards. That last detail was the substantive one. The original design said, in effect, the rules bite when the standards exist. The Omnibus said, the rules bite on a date, standards or no standards. A deferral that removes a conditional dependency is not purely a loosening; it is also a promise that the goalposts will stop moving.

There were simplifications elsewhere: lighter documentation duties for smaller enterprises, adjustments to registration. Useful, unglamorous, and the part of the package that actually reduced cost.

What it did not change was everything that had already applied. The prohibitions from February 2025 stood. The literacy duty stood. The general-purpose model obligations from August 2025 stood. An association that read "Omnibus" and heard "the Act is being watered down" was reading a story about 2027 and hearing a story about today.

The strategic error it invited

Here is the thing about a deferral: it is the most reliable way to stop work.

Between November 2025 and May 2026, I watched AI governance programmes in associations quietly lose their sponsor. Not cancelled — deprioritised. The deadline had moved, the budget was tight, and the person who had been pushing had a conference to run. Nobody decided to stop. It stopped.

That is the real cost of a proposal like the Omnibus, and it is not a cost the Commission intended. The rules did not get weaker. The attention did. And attention was the scarce input in a sector where the actual work — an inventory, an oversight procedure, a sentence telling a member when a machine has decided something — takes weeks, not years.

What a serious association took from it

Two things, and they are the same thing.

First, that a moving deadline is a reason to finish the parts that do not move. The prohibitions and the literacy duty were not in the Omnibus. They were already law.

Second, that the date was never the deliverable. The deliverable was always the inventory: which systems make judgements about people, who is accountable for each, and what happens when one is wrong. That work does not expire when a deadline moves. It expires when the person who understood the systems leaves.

The Omnibus was published as a simplification. Read a year later, it was a test of whether an organisation's AI work was anchored to a date or to a reason. The ones anchored to a date stopped that winter. The others carried on, and did not need to care what the final text said.

---

The Record · ← The day the model became somebody else's problem · Agreement in May, law in July →

---

Running an association? The AI Act for associations — what your body must actually do sets out the whole thing on one page: whether you are a provider or a deployer, the four duties that reach a membership body, the dates that matter, and the five items of evidence a board will ask for.