In a speech published by the Foreign, Commonwealth and Development Office on 23 September 2026, the Foreign Secretary used a UN Security Council session in New York to argue that advanced AI is now a question of state responsibility as well as technological progress. The central claim was that private developers may build frontier systems, but governments remain responsible for protecting the public from serious harm. (gov.uk) The address treated the issue as one for collective security rather than a narrow innovation debate. It presented AI as a technology with major public benefits, while warning that the same systems could affect cyber security, biosecurity, national security, financial stability and the future of work if safeguards do not keep pace. (gov.uk)
A notable feature of the speech was its rejection of an industry-only settlement. The Foreign Secretary said warnings from technology founders could not be ignored and argued that governments cannot transfer their basic public protection duties to companies, even where firms possess the deepest technical expertise. (gov.uk) This matters because it places the UK on the side of more active public oversight of frontier AI. The speech did not announce a Bill or consultation, but the direction is clear: ministers are positioning AI governance as a matter for democratic accountability, international coordination and institutional capacity inside the state. That is an inference from the policy line set out in the published text. (gov.uk)
The framework itself was organised around three tests: safety, transparency and resilience. On safety, the speech called for frontier models to be rigorously tested and for an assurance system around model development so that public authorities can judge whether systems are safe enough to deploy or release. (gov.uk) On transparency, the speech argued that governments need visibility into what leading AI companies are doing and noted that major developers have already made commitments to provide that access. On resilience, the emphasis shifted to consequence management: stronger cyber defences, protection for critical infrastructure, harder financial systems and public support for labour-market disruption linked to AI adoption or misuse. (gov.uk)
In practical terms, this reads as a case for pre-deployment testing, structured disclosure and broader contingency planning across essential systems. For frontier developers, the policy signal is that voluntary safety statements alone are unlikely to satisfy ministers if governments do not also have access to evidence, testing outcomes and the ability to assess misuse risks for themselves. That is a policy reading of the speech rather than a separate legal announcement. (gov.uk) For regulators and departments, the speech broadens the AI file well beyond digital policy. By placing child safety, jobs, cyber security, biosecurity, national security and financial stability in the same frame, the government is signalling that AI assurance will need cross-Whitehall coordination rather than a single-sector response. (gov.uk)
The UK case was also tied to institutional capability. The speech pointed to the AI Security Institute as the government's standing vehicle for testing, research and expert advice, and said it is working closely with the United States and leading AI laboratories. Official material describes the institute as a research body inside government focused on understanding advanced AI risks, developing mitigations and informing policy. (gov.uk) That institutional history matters. The government launched the world's first state-backed AI Safety Institute in November 2023 and renamed it the AI Security Institute in February 2025, explicitly sharpening its focus on national security and criminal misuse alongside broader model safety. (gov.uk)
The speech also sat inside a wider diplomatic timetable. On 22 September 2026, the Prime Minister told the UN General Assembly that AI would be placed at the centre of the UK's G20 Presidency the following year, with the aim of bringing governments together around common principles and standards. The government has separately confirmed that the UK will host the G20 Leaders' Summit in Manchester on 27 and 28 November 2027. (gov.uk) Read alongside that announcement, the Foreign Secretary's remarks look less like a standalone intervention and more like an attempt to set the UK's negotiating position early. The intended formula is not a simple choice between growth and guardrails, but an effort to make public confidence, state access and international coordination the basis on which further AI deployment becomes politically sustainable. That reading is inferred from the two speeches and the Summit announcement. (gov.uk)
There was also an equity argument in the address. The speech said common standards should work for the Global South as well as the Global North, suggesting that the UK wants AI governance to be framed not only as risk control for advanced economies but as a rules question with distributional consequences across different states and markets. (gov.uk) For businesses, public bodies and policy teams, the immediate takeaway is straightforward. The UK is using multilateral forums to push a model of AI governance built around testing, visibility and resilience, while keeping open the case for further domestic action if safeguards prove insufficient. The speech was not a legislative package, but it was a clear statement that the government sees frontier AI as a matter of public authority, not private discretion. (gov.uk)