Last week, UK AI Minister Kanishka Narayan announced the government would take steps to boost Britain’s resilience to AI-driven risks. New measures will include advancing the government’s cyber defence and considering how to use legislation to mandate serious AI incident reporting.
In the face of US and Chinese dominance and growing global concerns about an AI crisis, the UK is one of many countries seeking to build agency and security over AI. Last week at Chatham House, Canada’s AI Minister Evan Solomon outlined his country’s vision for secure, democratic technology and called for coalitions among likeminded countries to build and govern it.
Solomon’s speech came after a group of 20 countries and the EU also issued a joint statement calling for measures to control frontier AI models. Led by Norway and Finland, the signatories are all, arguably, ‘middle powers.’ They include democracies like Australia and Canada and non-democracies like Singapore and the United Arab Emirates.
The UK was noticeably absent from this joint statement. But UK prime minister Andy Burnham did use his UN General Assembly address to commit UK leadership to a global agreement on AI. And AI will be front-and-centre of the UK G20 presidency next year.
The joint statement is just a call to action. Yet it may be a promising indication that middle powers are creating the tools and networks they need to coordinate on safe, secure AI. The UK can strengthen its middle power status on AI through driving coordination while also carving out a leading role on AI security.
Welcome to the club
Coordination between middle powers first requires defining them. The challenge is that who counts as a middle power is contested.
One argument would be that ‘AI middle powers’ is simply a term for everyone except Washington and Beijing. The state of play is sobering. The US and China dominate up to 90 per cent of compute. Middle powers cannot compete with their capital expenditure on AI. Most countries are dependent on US and Chinese technology.
But middle powers like the UK are not constrained. To build agency over the AI stack, they have options. Some choose to align with a superpower or hedge between them. Others specialize, dominating a strategic foothold in the supply chain.
However, there is another option: coordination. For most countries, the secure, safe buildout and maintenance of the AI stack is already an exercise in inescapable interdependence. This cooperation extends far beyond flashy frontier models. There is a case to be made that middle power status itself is forged through coordination.
Recent months have seen a shift in the discourse on sovereignty: to build agency and shape technology markets, middle powers realize they cannot go it alone. Many EU states reiterated their commitment to European AI sovereignty at last week’s Dublin AI summit.
But the competitiveness, security and prosperity that sovereignty might deliver are endangered if countries can’t work together to mitigate AI-driven risks. Agency and readiness demand coordination extending beyond dialogue and declarations.
Countries in the Global South like India and Brazil have known for years that agency requires coordination, and have created open digital infrastructure like payment systems and digital IDs.
This imperative is now felt by others: last month, Canada’s Cohere and Germany’s Aleph Alpha joined forces to become ‘the first transatlantic sovereign AI solution’, while the two countries committed £300 million to a non-profit AI safety start-up.
Some middle powers are concerned about their readiness after a summer of hype (some warranted, some inflated) over AI’s potential to cause a global crisis.
Many realize their access to mostly American proprietary models such as Claude or GPT-5 is not guaranteed, while some are concerned about dependence on Chinese open-weight models. Accurate information about security risks and impacts is hard to come by: whether in the AI system or its applications. In the face of this uncertainty, Canada has led calls for a Technology Stability Board modelled on the post-crash Financial Stability Board (FSB).
Team GB
The UK is playing a balancing act. It is working bilaterally with middle powers, announcing partnerships with Canada on AI security and Germany on technology investment this month. It has diplomatic capital: it brought the US and China together for the world’s first AI Safety Summit in 2023, while King Charles convened AI executives at Dumfries last month.
But the UK has arguably been wary of wider middle power coordination on AI. At the same time, Burnham’s government has pursued a close relationship with the US, announcing an AI and defence deal last month. The US and UK signed a Technology Prosperity Deal for the UK’s AI infrastructural buildout last year, which has since been criticized for its lack of transparency and appears to have not been widely implemented.
The UK has an advanced AI ecosystem. Yet, alone, this buys Britain little. Granted, the US is a powerful partner on technology, but it is a volatile one: UK resilience and sovereignty are not guaranteed by alignment with the US.
A high-profile case in point: the UK’s AI Security Institute (AISI) is world-leading for evaluating models and their risks. But having previously been granted access to US frontier models, concerning reports have circulated that US companies might be under pressure from the White House to not send their models to AISI for testing immediately.
This summer featured multiple U-turns over whether anyone outside of the US would have immediate access to Anthropic’s latest models. The question is whether the UK, EU and other middle powers could have negotiated for access as a bloc – and whether the UK’s alignment with the US is at odds with coordinating with its counterparts.
Coordination of the willing
Optimism about a middle power alliance on AI is often misplaced. This is because coordination does not happen organically. Diplomatic infrastructure – information, institutions, and networks – is needed. This helps countries coordinate, especially in a crisis: like a future supply chain shock or revoked access to a widely used model.
Building diplomatic capacity on AI is gathering momentum. G7 finance ministers recently committed to using simulation exercises to build resilience. For a year, Chatham House has worked with AI Safety Asia to test diplomatic infrastructure through simulated crises to stress-test solutions, like risk monitoring and technical interventions. Findings can be scaled up to multilateral institutions.
The UK is well-placed to advance this diplomatic infrastructure. Its G20 presidency offers an opportunity for middle power coordination on AI. The UK government should use the various summits in 2027 – including the G20 Geneva AI Summit and UN Global Dialogue – to build consensus among different middle powers that can underpin a global agreement.
Specifically, the UK should use its diplomatic capital and technical authority to broker agreement on issues underpinning stability and trust. These could include risk monitoring, independent evaluations, incident reporting and notification, with trusted standards building reliability.
In the short term, the UK should sign the middle powers’ statement on AI security. Using the G20 to design an FSB-style arrangement could even pave the way for some mandatory testing and serious incident information-sharing.
US-China challenges
This does not come without challenges. US-China competition has prevented internationally binding rules on AI, although November’s planned bilateral talks on AI may move the needle on issues like an incident hotline.
Washington could impose retaliatory measures if middle powers come together on safety regulation that it perceives as unfriendly or to set conditions on market access. Meanwhile, other middle powers may be tempted by China’s AI offer, which is backed up with new multilateral initiatives such as the World AI Cooperation Organization, launched in Shanghai.
Despite these obstacles, coordination between middle powers – in regional or issue-based groupings, and on bargaining, building or securing AI – is flourishing. The UK can be a leader on AI security and advance an ambitious multilateral agenda on AI. But first, it needs a middle power strategy.