Summary

• The National Cyber Shield receives £60 million, including a £25 million challenge fund for British AI and cybersecurity companies.
• A separate £70 million National Exposure and Real-Time Vulnerability Centre will focus on weaknesses affecting critical infrastructure and the wider economy.
• A regulatory sandbox remains under consideration, while delivery structures, automated decision-making limits and operational targets have yet to be disclosed.

The UK government has committed £130 million to two artificial intelligence cybersecurity programmes intended to strengthen national defences and improve the identification of vulnerabilities across critical infrastructure, public services and businesses.

The package, outlined by HM Treasury on 24 October, allocates £60 million to a National Cyber Shield using agentic AI and £70 million to a separate National Exposure and Real-Time Vulnerability Centre, known as NERV.

The Cyber Shield investment includes a £25 million challenge fund intended to support British AI and cybersecurity developers. Ministers are also considering a regulatory sandbox that would allow selected technologies to be evaluated before wider deployment across sensitive operating environments.

Although the government has identified the funding and broad objectives, it has not published detailed delivery structures, operational dates or measurable performance targets for either programme. The commitments form part of a wider AI investment package that also covers model safety, domestic computing capability and technology skills.

National exposure requires more than scanning

Cybersecurity teams already use automated tools to discover vulnerabilities, detect insecure configurations and identify software that requires updating. The difficulty is interpreting those findings across organisations whose technology systems vary considerably in age, importance and accessibility.

A weakness affecting an isolated test system may present limited risk, while a similar vulnerability within equipment connected to essential services could require urgent attention. Determining the difference depends on understanding where systems are exposed, what they support and which safeguards are already in place.

The proposed NERV Centre is intended to use AI to identify weaknesses affecting critical infrastructure and the wider economy. It could potentially provide a broader view of exposure than assessments conducted independently by individual organisations, although the government has not explained precisely how the centre will obtain or process the information required.

One possibility would involve analysing reports from participating organisations, while another could include direct technical assessments undertaken under defined authorisations. The announcement does not establish which arrangements will be adopted, and those alternatives have different legal and operational implications.

Information about exploitable vulnerabilities is itself sensitive. A central programme must protect technical findings, restrict access to detailed records and establish procedures for informing the organisations responsible for affected systems.

Discovering a vulnerability also does not guarantee that it can be repaired immediately. Industrial operators may depend on specialist equipment that cannot be interrupted without affecting production or essential services, while changes to established systems can require extensive testing.

The value of a national assessment capability will therefore depend on the quality of its findings, the timeliness of responsible notification and whether organisations can act on the information provided.

Agentic defence introduces new operating risks

The National Cyber Shield is intended to make greater use of AI systems capable of carrying out multistep tasks through connected software tools.

Within a cybersecurity environment, such agents could assist with analysing alerts, correlating information from different systems and preparing recommendations for technical teams. Depending on their permissions, they might eventually perform limited defensive actions within predetermined operating boundaries.

Those applications differ substantially in risk. Producing an investigative summary is not equivalent to changing a firewall rule, disabling a user account or isolating a computer that supports an important business process.

Automated defensive action can reduce response times, but mistakes may interrupt legitimate services. An agent acting on inaccurate information could also compound an incident by making several inappropriate changes before an operator identifies the problem.

Organisations consequently need to establish which actions remain advisory, which can be performed automatically and when human approval is required.

Such systems must also be protected against attempts to influence their behaviour through the information they investigate. Malicious instructions embedded in websites, documents or incident records can create risks when an AI agent treats untrusted material as an authoritative instruction.

Testing these controls requires realistic scenarios involving incomplete information, unexpected system behaviour and attempts to manipulate the tools used during investigations.

The proposed regulatory sandbox could provide an environment for examining those questions, but the government describes it as a possibility under consideration rather than an established facility accepting participants.

Challenge funding targets domestic suppliers

The £25 million challenge fund within the Cyber Shield allocation is intended to support UK developers building AI-assisted security technologies.

Government challenge programmes can create opportunities for smaller companies to demonstrate products against practical public sector requirements. They may also allow procurement teams to examine approaches that would otherwise struggle to compete with established suppliers.

However, a successful demonstration is not equivalent to evidence that a product can operate reliably within critical infrastructure. Developers need clearly defined technical requirements, realistic testing conditions and access to organisations capable of evaluating security performance.

Commercial arrangements will also affect whether successful participants can turn an initial government-supported project into a sustainable business. Intellectual property rights, procurement procedures and the availability of follow-on customers all influence the wider benefit of public investment.

The government’s broader announcement includes an additional £120 million for the AI Security Institute and separate funding for domestic chip design capabilities. Those initiatives may complement cybersecurity development, but they are not part of the £130 million assigned to the Cyber Shield and NERV programmes.

The distinction matters because evaluating advanced AI models, developing semiconductor skills and deploying operational cyber defences involve different institutions, technical requirements and measures of success.

Resilience depends on implementation

Critical infrastructure organisations operate systems that can be difficult to secure through conventional enterprise methods. Equipment may remain in service for many years, depend on specialist suppliers or require maintenance arrangements that limit when changes can be introduced.

Automated assessment tools must account for those constraints. A vulnerability report may accurately identify a technical weakness while proposing a response that would be unsafe or impractical in the affected environment.

Useful national coordination therefore requires cooperation between cybersecurity specialists, operational engineers and the organisations responsible for maintaining essential services.

Potential measures of success include the accuracy of findings, the time taken to notify affected operators, the proportion of serious vulnerabilities addressed and the reliability of automated defensive processes under operational conditions.

These outcomes cannot be inferred from the investment amounts alone. Funding creates the opportunity to develop systems, but measurable protection depends on implementation, effective oversight and the ability to act on identified risks.

The government has outlined the two programmes ahead of its Budget, with detailed procurement, delivery and governance arrangements still to be established. Those decisions will determine whether AI can improve national cyber resilience without creating additional dependencies or unacceptable operational risks.