Construction AI: the Lords ask who gets to switch AI off, and a big UK contractor decides the answer is still a person

On 3 September 2026 the House of Lords started debating something British law has never had: a power for the secretary of state to shut down a large AI system, or the data centre running it, in an emergency. The amendment sits inside the Cyber Security and Resilience Bill, tabled by the Liberal Democrat peer Lord Clement-Jones and co-sponsored across the benches by the Conservative Baroness Harding, the crossbencher Baroness Kidron and Labour's Lord Hunt. Clement-Jones called it a "vital safety net" to halt a runaway system "before it can compromise our critical national infrastructure". The government has rejected it. A spokesperson said Britain "cannot simply turn AI off", and that blocking a model here wouldn't stop it being built or misused somewhere else.

Both positions are reasonable, which is what makes it worth your time rather than a Westminster sideshow. And notice the phrase that keeps coming up: the data centre running it. Peers reached for construction to describe the last-resort scenario, because the physical off switch, if there ever is one, is a building most of us are pouring foundations for right now. That's the bit that makes this ours and not just the AI crowd's.

Strip the "runaway superintelligence" framing away, though, and the row is about something very ordinary: accountability. Who is responsible when an autonomous system does the wrong thing, and what's the mechanism to stop it? That question is already live on sites deploying agents that read safety data, check drawings or move a programme. The answer Parliament lands on will set the tone for the smaller, duller versions of it that your insurer and your principal designer will ask you next year.

For your board pack: the kill-switch debate is a proxy for a governance question you can answer this month without waiting for the bill. For every autonomous tool you run, name the person who can stop it and the person accountable for its output. If you can't fill in both boxes, you're not ready to deploy it.

While Parliament argues the principle, one of the country's biggest contractors quietly answered the practical version. Microsoft reported on 2 September 2026 that around 40 per cent of Kier's colleagues now use Microsoft 365 Copilot, and that Kier is in the early stages of building its own AI agents. One collects and organises safety data so the people overseeing it can spot risk areas earlier. Another is being trained on Kier's safety standards and linked to the photographs, equipment records and drone footage the firm gathers across its portfolio. To get people trying it, they ran "prompt-a-thons", pitting a team answering questions unaided against a team using Copilot.

The 40 per cent is a Microsoft-reported figure about its own customer, so hold it at arm's length. What I'd trust more is the rule Kier's COO Louisa Finlay put around the whole thing. She calls it "Human at the Helm", and it means a person stays responsible for any decision the technology produces or supports. That's not a marketing line, it's a decision about where liability sits, and it's the same decision the Lords are wrestling with three sizes up.

What that means on site is worth spelling out. An agent trained on your own standards is only ever as good as those standards, so the standards quietly become the product. If yours live in a PDF nobody's opened since 2019, a model pointed at them will confidently reproduce the gaps. The homework isn't the AI. It's writing down how you actually work, clearly enough that a keen, tireless, occasionally-wrong assistant can follow it, with a named person checking what it hands back.

Worth doing: before you buy a safety agent, open your safety standards and read them as if you were the machine. If they're vague, that's your first project, and it's one you needed anyway.

Put the two together and you get the shape of the autumn. At the top, peers and ministers argue over a last-resort power for models running critical infrastructure. On the ground, the reality is far more modest: RICS found earlier this year that only 19 per cent of construction firms use AI regularly, against 29 per cent in commercial property, with most still stuck in pilots. So the governance conversation is running well ahead of the deployment it's meant to govern. That's not a reason to ignore it. It's the reason to get your own house in order while the stakes are still small.

The through-line from Westminster to your site is a single sentence: when the clever thing gets it wrong, someone has to catch it, and someone has to be accountable. Kier's answer, a named human at the helm, costs nothing and works at any size. You don't need a Copilot rollout or a data centre to apply it. You need a rule that every agent's output is signed off by a person, and a note of who that person is. Write it into your method statements now, while the only thing at risk is a duff daily log, not a national grid.

Today's action: pick the one AI tool your team already uses without anyone really deciding to, and put a name against it: who checks it, who's accountable. That's your governance policy, started.