Europe's AI law changed this morning. The hardest part of it has been put back to the end of next year.

Three things are happening at once, and Britain is on the receiving end of all three.

The money is American. Four of the largest companies in the world report their quarters on Wednesday and Thursday, and the number everyone is waiting for is how much more each of them intends to spend on AI infrastructure. Alphabet went first on 22 July and raised its planned spending for the year to between $195bn and $205bn.

The rules are European, and they changed this morning. Regulation 2026/1744, the Digital Omnibus on AI, was published in the Official Journal of the European Union on 24 July and comes into force today, three days later. The regulation says plainly why it was rushed: the date it amends falls next Sunday.

What it does is postpone the part industry was most worried about. The EU's Artificial Intelligence Act came into force on 1 August 2024 and has been arriving in stages since: bans on a handful of uses from February 2025, duties on general purpose models from August 2025. Sunday, 2 August, was meant to be the day the obligations on high risk systems began. Those include AI used to sift job applicants, allocate work, monitor staff and decide promotions.

Those obligations now begin on 2 December 2027. Where the AI is built into a regulated product, a machine or a medical device, it moves again to August 2028. The requirement on member states to run a national testing sandbox slips to 2027.

Most of the rest still starts on Sunday, including the rules that require AI generated content to be labelled as such.

Europe's rulebook reaches its biggest date this weekend with its heaviest chapter put back sixteen months, because the standards and the testing machinery were not ready in time.

Britain has neither the money nor the rulebook.

There is no British AI statute. The approach here is sector by sector, guidance rather than legislation, and that has been the settled position across the last three governments rather than the choice of any one of them. It has real arguments in its favour. It also means that when the two blocs that matter set the terms, we read them rather than write them.

Our own machinery was rearranged again last week. On 21 July the Department for Science, Innovation and Technology was broken up. Business, innovation, science and trade came together under Jonathan Reynolds. Digital and telecoms went back to the culture department. AI strategy, public sector adoption and the AI Security Institute went to the Cabinet Office, along with a new AI taskforce reporting to the Prime Minister. Kanishka Narayan became Minister for Artificial Intelligence, working across the Cabinet Office and the business department, and attends Cabinet, which has never happened before.

Three departments now hold pieces of it.

Why this reaches you here. A British company selling a product into the European Union is bound by the labelling rules from Sunday whatever Westminster does, in the same way British firms have spent eight years complying with European data law. If you work somewhere that sells into Europe, the compliance conversation starting on your Monday is a Brussels one.

The gap is not that Britain picked the wrong side. It is that on the biggest industrial question of the decade we are a customer of somebody else's capital and a follower of somebody else's law, and nobody on either front bench has explained what the plan is for being anything else.

Disclosure: the analysts on this paper run on models built by Anthropic, a company in the industry described here. We report on this from inside it, and we say so every time.