The UK's next AI bet should be AI verification
AISI is one of the UK's few pieces of international leverage in the AI race. More bets are needed, and AI verification technologies should be the next one
A bet on AI verification technologies can give the UK economic and diplomatic leverage. This is an early, open goal for the Burnham government on AI.
Summary
AISI is the Crown Jewels of the UK Government’s AI work (and arguably the most important thing happening anywhere in the UK on AI).
The UK should make more bets like AISI. Each one is a shot at economic growth, a piece of international leverage, and an opportunity to secure our future.
The next bet should be AI Verification: the tools that enable international coordination over the development and deployment of AI. Verification underpinned nuclear agreements, in Reagan’s words: “trust, but verify”; AI deals need this too.
AI companies and their employees are now calling for these technologies to be built, and there is a window of opportunity to be the ones building them.
There is an open door for the UK to lead. UKG (via AISI) can start by collaborating on existing verification projects before launching its own work.
I run Amodo, an R&D company in Sheffield working on AI verification (our Northern R&D cluster is open for collaboration!). In this piece, we’re talking our own book, but we have chosen this book because we think it’s vitally important for the world and a real opportunity for the UK.
Intro
In May 2026, under the headline “Inside the British Lab Hunting for Dangers Lurking in A.I.”, the New York Times spent 1250 words raving about the UK’s unique government capability in AI. Talking about the UK Government’s AI Security Institute, the NYT wrote:
“Now, the institute’s work is becoming a blueprint for other governments as concerns about A.I. safety grow. [...] With many governments lacking the technical understanding to police the technology and reliant on big tech firms to self-regulate, the institute may offer a different path to which A.I. experts bring real technological know-how into government decision-making.”
As AI capabilities continue to advance, the UK has the opportunity to make more bets like AISI. Each one is a shot at economic growth, a piece of international leverage, and an opportunity to secure our future.
Make early bets
AISI’s roots lie in the Bletchley Park AI Safety Summit. In November 2023, the UK government convened industry, academia, third-sector, and governments to talk about the emerging risks and opportunities from AI. The Bletchley Declaration, signed by 28 countries and the EU (including the US and China), was a prescient document outlining how we ought to orient to the development of AI.
The foresight of the UK Government to convene that meeting and set up the AI Security Institute is now paying dividends. As AI models continue to improve, AISI has become the trusted partner to AI developers in evaluating dangerous capabilities.
Not only did this move provide international leverage (AISI were the only non-American government organisation to receive access to Mythos for safety testing), but AISI has helped grow the UK’s AI industry.
AISI drew technical talent from American AI developers to serve a tour of duty in the UK government. After government, this talent can support British AI startups.
London has a growing AI safety, security, and governance ecosystem (Apollo Research, Conjecture, plus a host of non-profit research organisations)
The UK has a growing AI cybersecurity industry
In January, Faculty AI was sold to Accenture for ~$1B. Faculty’s work included projects with the AI Security Institute
The UK Government moved early and quickly with AISI. The next frontier is AI Verification.
The state of play three years on from Bletchley
2026 has seen a marked change in how AI developers and expert researchers are talking about AI. This summer has been characterised by the AI companies making unprecedented statements on the need for coordination over the development of AI.
This is because Anthropic and OpenAI both consider themselves to be on the precipice of Recursive Self-Improvement (AI systems that build themselves). In the midst of a competitive race to build Artificial General Intelligence, they both say that they are worried that they are going to build AI systems that they cannot control. OpenAI’s recent announcement that, in testing, one of their AI systems broke out of its testing environment and hacked the company Hugging Face is a small example of the kinds of things these companies are worried about.
The CEOs of both Anthropic and Google DeepMind are on the record saying that they would like to proceed more slowly, if only it were possible:
Demis Hassabis: “it would be good to have a bit of a slightly slower pace than we’re currently predicting, even my timelines, so that we can get this right societally. But that would require some coordination.”
Dario Amodei: You know, I said before, I prefer Demis’ timeline. I wish we had five to 10 years, you know, so it’s possible he’s just right and I’m just wrong, but assume I’m right and it can be done in one to two years. Why can’t we slow down to Demis’s timeline? [...] the reason we can’t do that is because we have geopolitical adversaries building the same technology at a similar pace. It’s very hard to have an enforceable agreement where they slow down and we slow down.
World Economic Forum
The most recent and clearest call has been from the employees of the AI companies themselves. In an open letter on the 28th of July 2026, more than 1000 AI company employees wrote:
“We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.”
Herein lies the opportunity:
“today, the world lacks the technical and governance tools to deliberately pace frontier-wide progress”
If we move now, the UK can be the one to build the technical tools and diplomatic levers needed to deliberately pace AI development. And these aren’t just tools for pacing the frontier; they’re a set of primitives that can enable the control and safe diffusion of advanced AI systems.
AI will be globally transformative, and if the UK moves now on verification technologies, we have a layer of the AI stack. Within just a few weeks, the UK can have an important piece of international leverage in AI, and start a new industrial sector.
What are these “verification technologies”?
The AI companies are racing, and they’re worried that in racing they’re not implementing the safeguards they’d like and that humanity needs. Whilst US regulation could solve the race happening between OpenAI, Anthropic, and Google DeepMind, it won’t have the same effect on international (e.g. Chinese) AI developers1. We therefore need international coordination, and this likely requires tools to verify that everyone is sticking to a deal. Verification technologies enabled extraordinary international coordination on nuclear development, as Reagan famously put it: “Trust, but verify”.
These rules could be about “pacing the frontier”, implementing agreed safeguards, monitoring deployment, or ensuring end users get what they’re paying for (e.g. the model being served is the one the end user expects).
“AI verification” tools enable us to make claims about AI development and deployment provable, rather than trusting assertions and documentation. This piece won’t go into too many details, but you can read more about some of these technologies at www.ai-verification.com or get a taste in this table:
How does this help the UK?
Economic development
AI verification can be part of the UK’s portfolio of AI bets. As AI eats more of the economy, we need to make sure the UK has companies with exposure to the upside of frontier AI capabilities:
AI verification technologies may ultimately be deployed in every single AI data centre in the world. Companies supplying these technologies may become very valuable.
In August 2026, any new AI verification project will immediately be world-leading. Globally, there are fewer than three AI verification R&D projects with more than five full-time staff. Like AISI in the early months, this work is early and first movers gain disproportionate leverage
The UK needs orthogonal AI bets. We aren’t training frontier models, haven’t been able to build large data centres, and don’t have any particular leverage in the AI supply chain. Moving early to seed a new field will help us bridge this gap.
International leverage
The US and China develop frontier models, Taiwan has TSMC, the Netherlands has ASML, and South Korea has SK Hynix.
The UK needs international influence, and AI verification is one way to have a seat at the table2. Moving early on these technologies will ensure that the UK plays a key role in international discussions over the development, deployment, and governance of frontier AI.
Securing humanity
Not all government actions are for economic development or international leverage. The Bletchley Park AI Safety Summit was not just held because there was an opportunity, but because the world needed to start talking about the safety and security implications of frontier AI.
The impacts of frontier AI are not confined to a single nation, so we must seek international coordination. AI verification technologies might be what enables this coordination.
Why the UK is well placed to do this
AI verification suffers from both a first-mover problem and the bystander effect.
Despite >1300 AI lab employees calling for these tools to be developed, AI developers describing these tools in blog posts, and papers with leading government, industry, and academic figures calling for this technology as early as 2020, there are fewer than a dozen projects worldwide, and by our estimates only ~15 full-time engineers in the world working on this!
The UK can overcome this first-mover problem. In particular, AISI’s incredible global reputation will help build field credibility and momentum.
It would be cheap to start this work.
We estimate that, to date, less than £15M has been spent on technical AI verification R&D focused on international coordination (it could be as low as £5-10M). Even a modest £10M project would immediately put the UK right at the lead of AI verification.
AI verification combines AI knowledge with security knowledge; UKG excels at this
AI verification R&D combines both knowledge of frontier data centre hardware and machine learning techniques (AISI), with a need for cybersecurity expertise (NCSC). These government teams mean that the UK is world-class at AI combined with security.
Early development can and should happen in public, but when it comes to red-teaming, this work requires nation-state-level offensive and defensive cybersecurity experience.
Work happening within the UK Government has the essential benefit of ensuring that there is internal talent who really understands AI verification. World-leading capabilities get the UK in the room on negotiations and enable UK policymakers to make the right decisions.
International reputation
AI verification aims to ultimately be an international endeavour. The UN, G20, plus new dedicated international fora will all play an important role in international AI governance – and the UK is well positioned to be a key node in each of these.
This work plays into the industrial and academic capabilities of the UK
Work can lean on the UK’s strong cybersecurity and engineering capacity. The Cambridge Cluster still contains incredible chip design knowledge, all of which supports hardware-enabled AI verification work. Particularly relevant projects include CHERI and LowRISC.
AI Growth Zones as testbeds
AI Growth Zones can host applied R&D including verification experiments.
In 1988/1990 the UK ran a series of mock weapons inspections, all on its own sensitive sites. These were designed to test whether they could prove compliance without harming the confidentiality of sensitive data.
Data centres in the UK’s AI Growth zones, plus the UK’s national AI capacity (e.g. Isambard) can be used for similar AI verification experiments. This should be combined with national security red-teaming.
This sounds great, what should the UK do?
Here are some initial ideas for the AI Minister, Kanishka Narayan, and the teams around him. We’re happy to help develop any of these ideas further and provide more context on AI verification work happening both in the UK and internationally.
The UK can immediately make both technical and political progress. There are easy first steps on both fronts, plus more ambitious goals that can be achieved by the end of the year.
Technical
Central goal: Start an AI verification team in AISI. This should deliver AI verification R&D, coordinate and fund UK AI verification groups, host joint AI verification experiments, lean on NCSC for cybersecurity expertise, and be the point of contact for international discussions on AI verification.
Immediate steps:
#1 As a first step, AISI should collaborate with existing AI verification projects.
AISI would act as an outside reviewer, giving feedback on the verification pilots being developed (e.g. the inference verification pilot we’re building in Sheffield). As a stretch goal, AISI could red-team the AI components of the verification scheme, seeing whether they can disguise training workloads and inference workloads.
This would both move forward the state of AI verification and be a quick way for AISI to learn about current AI verification techniques. This project could be done without funding any outside groups and could lead to a paper on the verification schemes being developed (the first such public piece with any government collaborator).
#2 Talk to the AI model developers about their priority AI verification projects.
Both Anthropic and OpenAI have expressed interest in AI verification. AISI can use their existing relationships to discuss priority projects that the AI model developers would like to see completed. These priority projects should lead into a research agenda at AISI and be the start of joint verification experiments with the AI model developers.
#3 Run early grant rounds on AI verification.
AISI should aim to seed a wider field of AI verification organisations. This should include non-profits, for-profits, and academic groups. Whilst there are some existing AI verification organisations, more are needed, and early grants help to seed a field that AISI will later draw on for R&D.
Such a grant round might be done in collaboration with other government organisations (e.g. a joint fund with ARIA (Nora Ammann’s programme Safeguarded AI)), or philanthropic funders.
#4 AISI should start an AI verification team, modelled on the successes of their Red Team and Alignment Team.
AISI is one of only a few organisations in the world with close ties to national security cyber expertise (their expert colleagues in NCSC), and in-house AI talent. AISI can spin up an AI verification R&D team and immediately accelerate the development of these technologies.
#5 Run Joint AI verification experiments (with industry and governments)
AISI should host a series of joint verification experiments to red-team AI verification techniques. These should bring together experts from industry and governments to test the latest AI verification tools.
The Joint Verification Experiment hosted by the US and Soviet Union was key to the development and adoption of nuclear verification technologies.
Diplomatic
Central goal: Lay the foundations for international coordination over AI development and deployment, including international agreement on AI verification tools. Early international collaborations should include joint verification experiments and early work to draft AI verification standards.
Immediate steps:
#1 At the UN General Assembly, the UK should discuss international AI governance, including the need for cooperation over the development of AI verification tools.
#2 The UK should start multilateral discussions on AI verification technology development. This should bring together governments, AI model developers, and AI verification researchers to discuss joint technical blueprints and AI verification research programmes.
The goal should be to establish a series of ambitious research programmes, plus international verification experiments to happen in 2027.
#3 At the AI summit in Geneva in 2027, the UK should ensure that AI verification is one of the central topics discussed.
Conclusion
There is an open goal for the UK Government. AI developers are sending up the demand signals, and if the UK moves now, then we can secure another point of leverage in the international AI race.
If you’re interested in this, please reach out: tom.milton@amododesign.com
We’re happy to support with more specific memos, share info about a world-first AI R&D cluster we’ve built in the North of England, or help you explore how the UK Government can start work on AI verification.
Unilateral moves to slow down by the US or even just one US AI developer could considerably slow the pace of AI development. If an AI developer did want to slow the pace of AI development to implement stronger safeguards then it may be sensible for them to do this even without international coordination.
Though we need more than one point of leverage. HMG should pursue many in parallel