The UK military is “developing plans” to help protect the country’s infrastructure in a potential war, according to a minister. It comes as a Member of Parliament (MP) raised concerns about the integration of artificial intelligence (AI) into critical national infrastructure (CNI).
Protecting CNI from warfare has risen up the agenda in recent years due to Russia’s full-scale invasion of Ukraine in 2022 and the US and Israeli-led attacks against Iran, which spilt over into a wider regional conflict.
Hybrid warfare has also increasingly become a concern. These are actions short of a fully declared war, such as cyber-attacks, drones flying near or over sensitive sites, and the use of proxies to commit acts of sabotage.
Earlier in November, the NCSC warned that it is “highly likely AI malware will be used in a ransomware attack on UK critical national infrastructure (CNI)” and the head of the centre said it could happen within two years.
Recent incidents relating to the physical security of European rail networks include the sabotage of high-speed rail lines before the 2024 Paris Olympics opening ceremony and the arrest in Poland of a spy ring placing cameras along railway lines where a single line blockage could have halted military logistics to Ukraine.
On 28 October, Conservative MP Stuart Anderson asked the government “what engineering capabilities the armed forces maintain to rapidly restore damaged critical national infrastructure assets during national emergencies”.
Responding on 6 November, Ministry of Defence (MoD) parliamentary under-secretary of state Louise Sandher-Jones said: “It is the responsibility of Critical National Infrastructure (CNI) Lead Government Departments to set and enforce standards for the protection and resilience of infrastructure within their sector, in partnership with CNI operators in the private and public sectors.
“If there is a requirement for military support, the Military Aid to Civil Authority process is a mechanism through which Defence can be engaged.”
Sandher-Jones added: “As outlined in the Strategic Defence Review, it is important that CNI is protected from attack above and below the threshold of conflict.
“Under Home Defence planning, Cabinet Office and CNI Lead Government Departments are seeking to deliver measures to ensure CNI is safeguarded against a range of threats from hybrid to conventional conflict.”
She concluded by saying: “Defence is supporting these efforts by developing plans to ensure we align Defence resources and capabilities to the protection of the most important infrastructure in a conflict scenario.”
The capitalised word ‘Defence’ is shorthand used by the government to refer to the M0D.
Reflecting on the minister’s response, Anderson told NCE: “We must have the resilience of CNI baked into our defence and wider national security preparedness.
“I am concerned by the lack of action from the government on this since the Strategic Defence Review in 2025 – we are still waiting for the Defence Readiness Bill.”
He added: “It is obvious that the armed forces would play a crucial role supporting CNI in a crisis, but the Ministerial response provided no concrete examples of engineering capabilities or programmes within the armed forces that would meet this specific need.”
Meanwhile, on 27 October, former defence minister and current Labour MP Al Carns asked: “What assessment the government has made of the potential risks posed by the deployment of autonomous or agentic artificial intelligence systems in critical national infrastructure.”
He specified that this should include “the ability of such systems to take actions without direct human approval; the potential for unintended or unauthorised actions to disrupt essential services; and the adequacy of existing safeguards, monitoring and human oversight arrangements”.
Responding on 6 November, Cabinet Office minister of state Kanishka Narayan, often referred to as the AI Minister, said: “Recent incidents of AI models carrying out unsanctioned actions are a serious reminder of the risks AI capabilities pose.
“The government recognises the scale and severity of AI risks to CNI; our response incorporates cross-government AI risk management, world-leading research, and over £100M of investment in new monitoring and response bodies.
“The 2026 Budget committed £40M to a National Agentic Security Unit to detect and respond to misaligned agentic activity with monitoring mechanisms.”
Narayan continued: “We regularly assess the risks, including through the cross-government AI Risk Register, which assesses the full spectrum of AI risks that could impact the UK, including within CNI sectors.
“The AI Security Institute (AISI) and the NCSC regularly share relevant research and guidance online, as well as with relevant stakeholders within government. Due to national security concerns, I cannot provide more detailed information.”
Cybersecurity firm e2e-assure CEO Rob Demain spoke to NCE about the threat AI posed to CNI.
“Autonomous AI is being adopted in the systems that keep water, power, and transport running faster than the safeguards around it are being built,” he said.
“The main risk isn’t necessarily a malicious actor taking control of AI, or the AI being malevolent in and of itself. The risk is that an agent will take an action that looks sensible but is wrong, at machine speed, in an environment where one mistake can interrupt an essential service.
“If an AI agent doesn’t have the full context of an operational environment, what can seem like a sensible action – like shutting down a system if it notices an intrusion – can have physical, real-world consequences.”
He added: “The government’s answer should be practical. Agents should have only the access they need, firm limits on what they can touch, independent monitoring, and a reliable off switch that a human can use.
“The best way to test how an AI agent will behave in a real environment is to rehearse its actions in a digital twin – a replica of a real environment – to identify any unexpected actions before a real incident occurs. This oversight has to be designed in before deployment, not added afterwards.”
The project was expected to see a live test of a tool built by a company called Emerald AI, utilising Nvidia GPUs (graphics processing units), to “dynamically adjust energy consumption” of a data centre in late 2025.
Main image: A stock photo of a soldier with a drone.