Britain and the United States are creating a new defence partnership around artificial intelligence and autonomous systems, with an initial focus on protecting critical infrastructure including sea lanes and airspace. The UK government announced the agreement on 22 September as part of a wider package of defence and technology cooperation. The partnership brings together the UK’s Rapid AI Delivery Taskforce and the US Chief Digital and Artificial Intelligence Office.
The timing matters because the same announcement was accompanied by a concrete demonstration of autonomy at sea: a British uncrewed underwater vehicle fired a US heavyweight torpedo during trials. The government presents the partnership as a way to accelerate systems that can detect threats, support deterrence and work across both countries.
Critical infrastructure is moving into the AI security agenda
The official announcement says the collaboration will focus initially on systems intended to protect national infrastructure. Independent defence reporting describes the same partnership as an effort to develop interoperable AI and autonomous capabilities across the two countries.
Critical infrastructure covers assets where failure has consequences far beyond one organisation. Sea lanes, communications, energy systems, transport networks and airspace are not simply business systems. They are parts of national resilience.
That makes AI deployment in this setting different from ordinary automation. A mistake in an office workflow may waste time or money. A mistake in infrastructure protection can misidentify a threat, miss a real one or trigger a response with wider security consequences.
The partnership sits inside a much older cyber problem
LiveAIWire has previously covered Five Eyes warnings about AI-enabled cyberattacks. The Five Eyes cyber agencies say AI is accelerating the speed, scale and sophistication of cyber threats. The new UK-US partnership is the defensive side of the same race.
If AI systems can monitor more signals, coordinate autonomous platforms and respond faster than human teams alone, they could improve resilience. But increased autonomy also creates new attack surfaces. A defensive system that depends on models, sensors, software supply chains and machine-to-machine communication can fail in ways older infrastructure security programmes were not designed to handle.
That is why the word interoperable is important. Systems built jointly across two militaries have to exchange information and operate under compatible rules without creating ambiguity about authority.
Autonomy does not remove the accountability question
LiveAIWire’s analysis of AI in military strategy has already highlighted the difficulty of assigning responsibility when software influences high-consequence decisions. The new partnership does not eliminate that problem by focusing on infrastructure protection rather than offensive weapons.
A system that identifies a suspected threat still needs rules about what it may do next, which actions require human approval and how uncertainty is communicated. Those design choices become more important as the speed of the underlying system increases.
The government’s announcement emphasises protection and deterrence. It does not provide enough detail to assess the autonomy limits of future systems, and it would be premature to treat the partnership as evidence that operational safeguards have already been solved.
The undersea trial shows how quickly autonomy is becoming physical
The torpedo demonstration matters because it moves the conversation beyond software dashboards. An uncrewed underwater vehicle is a physical platform operating in a difficult environment where communication, navigation and target information can all become uncertain.
That does not mean the new AI partnership directly controlled the torpedo trial. The government presents the trial and partnership together as evidence of wider UK-US cooperation on next-generation defence technology. The distinction should remain clear.
LiveAIWire has also examined the growing policy interest in technical controls and remote restrictions around advanced AI systems. Defence autonomy raises the same underlying issue in a harsher setting: powerful systems need boundaries that still work when communications are degraded and decisions are time-sensitive.
This is an alliance story as much as an AI story
The partnership also reflects a strategic choice about who builds these systems together. AI capability increasingly depends on shared models, compute, data, sensors and standards. Defence alliances therefore have to decide not only what technology to buy, but how deeply they are willing to integrate the digital infrastructure behind it.
For Britain, cooperation with the United States can provide scale and access to capabilities that would be expensive to reproduce independently. For the United States, a closely integrated partner can provide operational environments, specialist expertise and allied deployment pathways.
The risk is that interoperability becomes dependence if one side cannot operate or audit critical systems without the other. That question will matter more as joint AI programmes move from announcements into deployed capabilities.
The next evidence should come from deployment, not promises
The UK-US announcement establishes direction rather than outcome. It tells us which institutions will work together and which kinds of infrastructure they want to protect. It does not yet tell us how much autonomy future systems will have, how they will be tested or what performance they will achieve in real conditions.
LiveAIWire’s reporting on Britain’s demand for kill switches and control mechanisms in AI agents offers a useful principle for watching this programme develop. The more authority a system receives, the more important it becomes to know how that authority can be constrained or withdrawn.
The partnership is therefore significant without needing to be described as a breakthrough. Two close allies are moving AI and autonomy deeper into the systems that protect physical infrastructure. The next question is whether the controls mature as quickly as the capability.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.