Construction AI: OpenAI offers Washington a 5% stake in the frontier, and Gateway 2 approvals climb again to 77%

The Financial Times reported on Thursday 2 July that OpenAI has proposed handing the US government a 5% equity stake, worth roughly $42.6bn at the $852bn valuation it closed in March. Sam Altman's pitch, as corroborated by CNBC, Forbes and Bloomberg, is modelled on the Alaska Permanent Fund, the sovereign wealth vehicle Alaska set up in 1976 to bank its oil surplus, and the proposal imagines Anthropic, Google and Meta ceding similar stakes into the same pot. The discussions are described as conceptual, and would likely need an act of Congress, so hold this loosely. But the direction of travel is hard to miss.

Because it doesn't arrive in isolation. On 25 June, the White House's Office of the National Cyber Director asked OpenAI to stagger the release of GPT-5.6, the model I covered last Tuesday, so that government-approved customers got it during a preview window before the general public. That request sits under the executive order signed on 2 June, which gives Washington up to 30 days of pre-release access to frontier models on a voluntary basis, and OpenAI agreed. So within a fortnight the US government has delayed a flagship model's public launch and been offered a seat on the cap table of the company that makes it. Regulator, customer, and now possibly shareholder, all at once.

What that means on site is less abstract than it sounds. If your estimating assistant, your document checker or your golden thread tooling runs on a frontier model, and increasingly it does, then the release cadence of your software now has a political layer in it. A model your vendor promised for Q3 can spend a month in a government preview tier first. I'm not saying that's wrong, the cyber logic for reviewing models that can find software flaws on their own is defensible. But UK firms should stop assuming the model market is a normal commercial one. It's becoming something closer to an aviation supply chain: strategically managed, politically sensitive, and prone to schedule slips that no procurement clause of yours will touch.

For your board pack: List which of your AI tools depend on US frontier models, and ask each vendor one question: what happens to our service if your model provider's next release is delayed or tiered? A vendor with no answer is a vendor with no plan.

A dated update to last Thursday's brief, where I reported the BSR's approval rate reaching 75%. The regulator's latest figures, published on 3 July and covered by Construction Management, show the improvement holding for a second period. In the 12 weeks to 28 June 2026 the BSR made 368 Gateway 2 decisions with a 77% approval rate, 284 cases, against 358 decisions and 75% in the period to 30 May. Those approvals represent 6,544 residential units, and there are 138 new-build and conversion applications live in the system totalling 30,393 units. Acting chief executive Charlie Pugsley says existing-building remediation approvals have hit 85% over the last 12 weeks, already past the 65% minimum target for the year.

There's texture in the detail worth having. London accounts for 57% of decisions, internal refurbishment work has been split into its own category and now makes up the majority of the live caseload, and the programme is running on 152 internal and 493 external staff, with more recruitment planned to cut determination times. And for the first time we have a Gateway 3 number: 277 applications at the final hurdle before residents move in, with the regulator saying it is working with applicants towards approvals rather than rejections. That last stage is where the golden thread either exists as structured, retrievable information or it doesn't, and 277 schemes are about to find out which.

So, two consecutive periods of more decisions and higher approval rates. I hedged last week about whether one good quarter was a trend; I'll commit now, the remediation plan launched in April is working, at least on throughput. What it hasn't yet fixed is determination time on the older complex cases, and an 18-week target for non-complex applications is still pencilled for March 2027 rather than tomorrow. But if you shelved a higher-risk building scheme in 2025 because the gateway felt like a lottery, the person doing your programme risk register should be looking at these numbers, not last year's.

Today's action: If you have a Gateway 3 submission in the next six months, pull a sample of your handover information this week and test whether someone outside the project can actually find fire-stopping records in it. The regulator is signalling cooperation; meet it with evidence.

Friday's brief covered NG Bailey creating a chief AI officer and Kier's data appointments. A Construction Wave piece published on 2 July shows how far the same wave has spread. Laing O'Rourke's AI subsidiary Akordi, launched in 2023 for the strategic application of AI in infrastructure, is advertising for an AI construction optimisation lead, candidates with design or digital experience wanted. Turner & Townsend has gone further up the ladder, appointing Emma Gilthorpe, the former Royal Mail chief executive, as chief transformation officer with an explicit brief to support the firm in taking up AI and technology. Her colleague George Mokhtar, head of digital for real estate, says the opportunity goes well past productivity, into predictability, quality and speed of delivery on major programmes. Vendor-adjacent optimism, yes, but from a firm that answers to clients rather than software buyers.

What we've found watching this run of appointments is that the titles matter less than the reporting lines. A chief AI officer who reports to the chief executive can stop a bad deployment; an innovation lead buried three layers down mostly writes newsletters. RICS's acting president-elect Maureen Ehrenberg made the same point in the piece, firms need leadership that will champion responsible use, and a 2025 RICS report found plenty of optimism on cost control and risk management sitting alongside an honest admission that most organisations weren't ready. Ready, in practice, means someone accountable when the tool is wrong.

The comparison I keep reaching for is the health and safety director, a role the industry once treated as overhead and now treats as non-negotiable. The comparison only goes so far, AI risk won't kill anyone on site tomorrow, but the governance shape is the same: a named person, board access, and the authority to say no. Smaller firms don't need the salary line, they need the name. Someone whose job it is to know what the tools do, what they send home, and when they're wrong. That's what it's about.

Practical bit: Write one sentence in your management meeting minutes this month: "X owns AI adoption and governance." If you can't fill in X, that's the gap, not the tooling budget.

Put the three items together and you get the shape of mid-2026. Upstream, the frontier is being folded into the state, with Washington delaying releases and being offered equity, so the models underneath your tools now move to a partly political timetable. Downstream, the UK regulator that gates your higher-risk work is visibly speeding up, two periods running. And in the middle, firms are responding the right way, by putting a named person in charge of the technology rather than letting it seep in through personal ChatGPT accounts and hope.

The standing discipline doesn't change: date your facts, name your owners, and verify announcements against documents. The BSR numbers are checkable against the regulator's own data. The OpenAI stake is conceptual until Congress moves. Treat each accordingly.

Worth doing: Add a quarterly line to your risk register for model-supply disruption, the same way you carry one for materials lead times. It felt paranoid for steel in 2020 too.