Briefing: The Leadenhall Project – Do not let anyone say ‘AI told you so’

Although 2026 may have started with political risk as the primary flashing blob on the UK’s risk register, 2025 concluded with former Google chief executive Eric Schmidt issuing a viral warning to “pull the plug” of artificial intelligence (AI) if it begins to improve outside of human control.

AI is certainly UK general insurance’s (UKGI) trouble-bubble – but what remains when this unsustainable growth bubble bursts will be even more impactful than the internet.

So, how can market firms shape their strategies for this economic nuclear weapon?

Let’s consider the landscape.

We should all be excited and petrified about AI in equal measure. To most c-suites, my sympathies. I fear anyone confident in their strategy is not nearly paranoid enough.

One thing is clear – it simultaneously holds true that AI is undoubtedly a bubble and that it is also only going to improve. It will dually impact digital and physical industries across economies. You only need to look at engineering firm Boston Dynamics’ new robot – which motor manufacturer Hyundai has allegedly ordered 10,000 units of – to see the impact of AI technology.

Factories and makers of tangible goods can assess the impact of AI without worrying about becoming stuck as a captive application layer. There will likely be other robot companies to switch out to and engage – or even humans to bring back should robot companies inflate their pricing unreasonably.

Service-based businesses might want to think twice, however. Underestimate the uniqueness of human expertise and experience at your peril. While human staff are often the bulk cost of any organisation, they are far easier to negotiate with than a multitrillion pound AI monopoly.

This is where the AI bubble is headed. Once an industrialised trait is lost, it is hard to get back – services will be no different.

It may unsettle you to keep hearing that AI is in a bubble.

Cynical? Did you know Microsoft accounted for 20% of AI company NVIDIA’s revenue and ChatGPT creator OpenAI now owns 10% of chipmaker AMD?

An October 2025 article published by Yale School of Management, entitled This is how the AI bubble bursts, reminded readers that AI is estimated to have generated 1.1% of the US’ headline gross domestic product (GDP) growth.

These musing may sound like a future Michael Lewis book, with a catchy title of AI told you so or something similar. This is not outside the realm of possibility – we have all asked AI and it will list the industries it expects to learn and do in due course.

It cannot be ignored that we are in the AI bubble because this environment will happen regardless of what we believe on an individual or firm basis. Businesses just need to plot a good course in order to sail the current and survive if the bubble pops.

Cat and mouse costs

Any user of AI tools can vouch for their power. AI is taking over functions. Tools like Bolt and Lovable allow products and prototypes to be coded in hours. Many engineers will rightly caution that these apps are often clunky on code volume, act randomly, have security issues and may not scale. Fair.

And business model prompts use credits – the amount of digital calories AI will use to execute requests. If an AI totally wrecks your prototype, then there is often no recourse – just the option to spend more credits to fix the problem. The professional indemnity market may need to turn its attention to that long-tail risk sooner rather than later.

With hundreds of billions being spent on new data centres and most big AI businesses losing billions themselves, the economics surrounding AI resemble an underpriced game of cat and mouse. This is not sustainable for transformation budgets.

Indeed, the Massachusetts Institute of Technology reported in August 2025 – within its The GenAI Divide: State of AI in Business 2025 study – that 95% of 52 enterprises collectively investing $30bn to $40bn (£22bn to £30bn) in AI projects did not generate a return on investment. Now imagine if credits were 10 times this cost.

Strategy stress test

So, time for firms to stress test their strategies to navigate the AI storm that lies ahead.

Let’s begin with the assumption that UKGI businesses have already or are currently operating protocols that are protecting intellectual property, to ensure company data is not AI training fodder. That was the first warning over a year ago – and we trust that AI companies really are walling off our data, right? OK.

The second thing to do is analyse. Do not pick winners in each use case. Let evolution avail. Deploy multiple tools. Ensure roving engineers perform numerous ‘codetopsies’ on projects to gauge the incremental costs, benefits and capabilities of each. This will help firms spot signals of viability leaps.

Next, avoid reliance. Bake in plans ensuring that AI and its credit rivers are interchangeable. Ensure that you own the interface or middle layer, so that you are not hamstrung by the ‘bubble debt’ – the fictional variant of technical debt.

Many in insurance too often mistake technology as a cost, rather than an asset. This can be avoided by compartmentalising AIs to conduct niche functions and wisely training your own specific tools. It is self-evident that one firm alone cannot hope to match the capabilities of Anthropic, Google and OpenAI.

The Leadenhall Project proposal

That final point leads me to the biggest strategic question – does AI present the prospect that businesses are at risk of becoming some kind of enslaved application layer? Maybe. US big tech has some enviable financial firepower.

If AI is set to become to economies and their industries the equivalent of nuclear weapons to nation states, perhaps we need an AI linked Manhattan Project for our sector, mirroring the mentality of the World War II US government research programme that created the first nuclear bombs.

I propose the Leadenhall Project.

UKGI does not need an everything AI – it needs an AI for risk value chains.

Suspend disbelief and consider how a mutualised, sector-owned AI could be the perfect hedge and help defend against the future input cost uncertainty generational AI tools represent.

Shall we have a symposium? Sadly, talk is not free – let’s just price up the conversation in credits as we go.

Awaiting prompts.