The UK government will spend £130m on artificial intelligence cyber defence, including a National Cyber Shield that uses agentic AI to protect critical infrastructure, as part of a wider package of more than £600m for AI to be set out in this week’s Budget.
The plans were announced in a statement on Saturday, ahead of Wednesday’s Budget. The government said the funding is intended to strengthen the UK’s defences as AI tools make it easier to find and exploit weaknesses in digital systems, while also giving British AI companies and researchers new support.
The cyber allocation is split between two main projects: £60m for the National Cyber Shield and £70m for a National Exposure and Real-Time Vulnerability Centre. The government also confirmed £120m for the AI Security Institute and a new NCSC AI Cyber Security Centre of Excellence in Manchester.
Key facts at a glance
- £130m for AI cyber defence, split between £60m for the National Cyber Shield and £70m for the NERV Centre.
- £25m of the National Cyber Shield money will form a challenge fund for UK AI and cybersecurity companies.
- A regulatory sandbox will be scoped for agentic security tools across critical infrastructure, the public sector and industry.
- £120m for the AI Security Institute, which tests advanced AI models.
- A new NCSC AI Cyber Security Centre of Excellence in Manchester will train government cyber and AI specialists.
- £60m more for the Sovereign AI Fund, with a new focus on physical and industrial AI such as robotics.
- £60m for Sovereign AI’s R&D procurement scheme.
- £18m for a UK Advanced Chip Design Academy.
- £30m for local AI partnerships with mayoral authorities.
- £23m over three years for the AI Economics Institute, plus £10m a year to study how AI will reshape jobs.
- £15m for an AI Challenge Prize, starting with health.
- £165m for biosecurity, including a £100m AI Biosecurity Lab supported by OpenAI and Anthropic with up to £200m in model credits.
National Cyber Shield and a challenge fund for UK firms
The National Cyber Shield receives £60m. Within that total, £25m will be used as a challenge fund for UK AI and cybersecurity companies. The aim is to draw on domestic expertise in machine learning, threat detection and automated response, rather than relying only on tools built overseas.
The government will also scope a regulatory sandbox for agentic security tools across critical infrastructure, the public sector and industry. Agentic AI systems can pursue goals, call other tools and chain actions together with limited human input. That makes them potentially useful for monitoring networks and responding to attacks at machine speed, but it also raises questions about accountability, unintended actions and the security of the AI systems themselves.
A sandbox would allow companies and public bodies to test such tools under regulatory supervision before wider deployment. The statement did not say how the sandbox would be governed, which sectors would take part first, or what safeguards would be required. It also did not specify when the National Cyber Shield tools would be deployed.
The challenge fund could give smaller UK firms a route into government cyber work, but the package does not set out how contracts will be awarded or how quickly products will move from testing into operational use. Those details will matter for companies hoping to benefit from the spending.
NERV Centre to map vulnerabilities at speed
The National Exposure and Real-Time Vulnerability Centre, or NERV Centre, receives £70m. It will use AI tools to find cyber vulnerabilities across critical infrastructure and the wider economy. The centre is intended to improve exposure management: identifying weaknesses before attackers can exploit them and helping organisations prioritise patches and mitigations.
Critical infrastructure spans energy, water, transport, health, finance and telecommunications, all of which depend on complex supply chains and legacy systems. AI can scan large volumes of code and network data faster than human teams, but it can also produce false positives or miss context-specific risks. The effectiveness of the NERV Centre will depend on how its findings are validated, disclosed and acted on by operators.
The statement did not say when the tools will be deployed, or how far they will act without human oversight. Those omissions are significant because automated vulnerability discovery and response can create risks if systems are patched or isolated without adequate checks. Real-time vulnerability data can also be sensitive if it reveals unpatched weaknesses in national infrastructure.
AI Security Institute and a Manchester skills hub
The AI Security Institute, which tests advanced AI models, receives a further £120m. The funding extends work on model evaluation, safety testing and the security of AI systems. As AI models become more capable and more widely embedded in public services and industry, testing is seen as one layer of defence against misuse and accidental failure.
A new NCSC AI Cyber Security Centre of Excellence in Manchester will train government cyber and AI specialists. The centre is intended to build skills across the public sector, where demand for people who understand both cybersecurity and AI is growing. Training government specialists could help departments assess risks, procure tools and respond to incidents more effectively.
The AI minister, Kanishka Narayan, said last week that AI tools had found 500 previously unknown flaws in parts of the public sector. That figure illustrates both the scale of hidden weaknesses and the potential of AI-assisted scanning. It also underlines the challenge of fixing vulnerabilities once they are found, especially in fragmented public sector systems.
The Institute’s expanded role comes as governments and regulators try to keep pace with fast-moving model development. Testing alone cannot guarantee safety, but it can provide evidence about model behaviour and help buyers set conditions before deployment.
Recent attacks sharpen the focus on resilience
The plans follow a week in which the PHP2Shell flaws pushed government websites across Europe offline. The incident served as a reminder that widely used software components can create cascading failures across borders and sectors. Governments are increasingly concerned that AI could shorten the time between vulnerability discovery and exploitation.
Ciaran Martin, former head of the National Cyber Security Centre, said: “This week we’ve had a stark reminder that AI exposes some of our digital world’s weaknesses and brings new ways of exploiting them to those who’d do us harm. So we need to padlock our systems.”
Narayan has previously said that testing AI alone is not enough and that Britain must harden its defences. The Budget package reflects that view by combining cyber defence spending with support for domestic AI capability, research and skills.
The announcement also reflects a wider shift in government thinking: AI is being treated both as a tool for defence and as a source of new risk. That dual framing helps explain why the funding covers threat detection, model testing, workforce training and industrial policy in the same package.
Chips, start-ups and jobs
The rest of the package backs British AI companies and research. An additional £60m will go to the Sovereign AI Fund, which will add a focus on physical and industrial AI, such as robotics. A further £60m will support Sovereign AI’s R&D procurement scheme, intended to help public bodies buy and test new AI tools.
£18m will fund a UK Advanced Chip Design Academy. £30m will support local AI partnerships with mayoral authorities. £23m over three years will go to the AI Economics Institute, plus £10m a year to study how AI will reshape jobs. £15m will establish an AI Challenge Prize, starting with health.
The largest single item in the wider package is £165m for biosecurity. That includes the £100m AI Biosecurity Lab announced on Saturday. OpenAI and Anthropic will support it with up to £200m in model credits. The involvement of major AI developers reflects a growing focus on biological risks and the role of advanced models in both creating and defending against them.
Narayan said in the statement: “We cannot afford to rely entirely on technology built overseas.” The sovereign AI measures are designed to support UK companies, researchers and infrastructure, though the package also depends on partnerships with foreign AI developers for the biosecurity lab.
The funding is spread across cyber defences, model testing, skills, chips, local partnerships, economic research and biosecurity. The government has framed the package as a response to both the opportunities and risks of AI. The statement did not give a detailed timetable for the cyber projects, and it left open how far agentic tools would be allowed to act without human oversight.
Source: TNW | Government-policy News