AI, Cybersecurity and Digital Defence: Who Is Paying to Protect Britain’s Digital Future?

1. The problem: Britain's digital dependence is a national security issue

Cybersecurity is no longer simply a matter of protecting office computers or preventing individual email accounts from being hacked. Digital systems underpin financial transactions, healthcare, transport, energy, communications, local government and many of the services on which everyday life depends. A serious cyber attack can therefore become a physical, economic or public safety problem.

The National Cyber Security Centre (NCSC), part of GCHQ, reported 204 nationally significant cyber incidents during the year covered by its 2025 Annual Review, compared with 89 in the previous review period. Eighteen incidents were categorised as highly significant. The NCSC cautions that its incident figures do not represent every cyber attack affecting the UK, because not all incidents are reported or handled by the organisation.

The report also identifies ransomware as a major continuing threat. Ransomware is malicious software used to disrupt systems, steal information or prevent access to data, often in an attempt to extort payment. The effects can spread beyond the organisation directly attacked, disrupting suppliers, customers and essential services.

Healthcare provides a particularly serious example. The NCSC's review reported that the ransomware attack affecting pathology services provider Synnovis caused substantial disruption to healthcare in London. The incident was associated with costs of £32.7 million and directly contributed to at least one patient death, according to the review. This demonstrates why cyber resilience must be treated as part of public safety rather than merely an IT expense.

AI enters this picture in two ways. It can help defenders identify suspicious activity and investigate vulnerabilities. It can also help attackers make existing techniques more efficient, increase the volume of malicious activity and produce more convincing attempts to deceive people.

Source: NCSC Annual Review 2025: Incident Management; Engineering Resilience Against Critical Loss.

2. What has the Government announced?

Budget 2026, published on 28 October 2026, sets out an additional £154 million for cyber defences through three main commitments: £70 million for a National Exposure and Real-Time Vulnerability Centre, £60 million for the first phase of a National Cyber Shield, and £24 million to expand the public-sector AI defence programme.

These amounts add up to £154 million. They are not three estimates of the same investment, and they should not be counted again on top of the £154 million headline figure. The Budget also identifies other related initiatives, including a £7.5 million second phase for the Cyber-AI Hub at Queen's University Belfast and a new Centre for Cyber and AI Excellence in Manchester.

The National Cyber Shield: £60 million

The National Cyber Shield is intended to develop AI-enabled capabilities that automate elements of cyber defence. The Government says the first phase will be developed in partnership with the UK cybersecurity industry. The package includes a £25 million AI and Cyber Security Challenge Fund to support innovation, together with work to scope a regulatory deployment sandbox.

A deployment sandbox is a controlled environment in which organisations can test new technology under defined conditions before wider adoption. In this context, it could help establish whether AI-enabled security tools are reliable enough to use in critical infrastructure, public services and industry.

The challenge fund could give British cybersecurity companies opportunities to develop tools that might otherwise struggle to attract early funding. However, the published announcement does not by itself establish how many awards will be made, what the selection criteria will be, which organisations will receive funding or what results each project must demonstrate.

The NERV Centre: £70 million

The National Exposure and Real-Time Vulnerability Centre is intended to use AI tools to identify cyber vulnerabilities across critical infrastructure and the wider economy. A vulnerability is a weakness in software, hardware, configuration or operating practice that may be exploited to gain unauthorised access or disrupt a system.

The concept addresses a longstanding security problem: organisations often have more weaknesses to investigate than their teams have time to address. A national capability could help identify important exposures, prioritise remediation and improve the overall picture of Britain's digital vulnerabilities.

Yet discovering a vulnerability is only the beginning. Someone must verify the finding, determine whether the weakness is exploitable, assess the potential consequences, coordinate with the responsible organisation and ensure that the problem is fixed safely. A detection system that generates more alerts than security teams can investigate may increase workload without materially improving protection.

Public-sector AI defence: £24 million

The Government plans to expand its public-sector AI defence programme to strengthen technology used by local authorities, hospitals and schools. These organisations hold sensitive information and depend on systems that may include a mixture of modern cloud services and older technology.

The investment could support earlier detection, faster incident response and more consistent security practices. However, the Budget announcement does not establish how the £24 million will be divided among councils, hospitals, schools or central government, nor does it provide a complete public breakdown of delivery milestones.

The value of the programme will depend on whether it improves actual security outcomes, rather than simply purchasing new software. Staff training, reliable backups, timely software updates, secure access controls and tested recovery plans remain essential, regardless of whether AI is introduced.

Sources: HM Treasury, Budget 2026; Government announcement on the AI investment package.

3. Why is AI being used in cyber defence?

Traditional cybersecurity already uses automation, statistical analysis and machine learning. The new funding is therefore not about introducing artificial intelligence into a previously entirely manual field. It is about developing and expanding capabilities that can analyse large quantities of information, adapt to changing patterns and potentially automate more complicated security tasks.

For example, a security team may receive thousands of alerts from computers, cloud services, network equipment and user accounts. An AI-supported system could help identify which alerts are connected, highlight suspicious behaviour and prioritise investigations according to potential impact.

Another application is vulnerability management. AI tools may help examine software configurations, analyse technical reports, identify relationships between weaknesses and recommend remediation steps. More advanced systems could coordinate several defensive actions, provided that their permissions and decision-making processes are appropriately controlled.

These capabilities could be valuable where attackers operate continuously and defenders face limited resources. But the quality of the results depends on the data available, the system's design, the accuracy of its analysis and the organisation's ability to act on its findings.

AI should therefore be treated as a way to strengthen a security programme, not as a substitute for a well-managed security programme.

4. The other side of the threat: AI can help attackers too

The NCSC's assessment of AI and cyber threats identifies the enhancement of existing attacks as an important near-term concern. Its 2025 Annual Review describes hostile actors using large language models to support reconnaissance, process stolen information, improve social engineering and assist vulnerability research.

Social engineering involves manipulating people into revealing information, transferring money or taking actions that compromise security. AI can make fraudulent messages more fluent, help criminals personalise communications and support impersonation attempts. These uses do not require a completely new form of cyber attack; they can make familiar methods easier to scale.

AI can also create new defensive challenges when organisations connect models to email, databases, code repositories or other operational tools. These connections can expand the number of ways an attacker might attempt to manipulate a system or gain access to sensitive information.

That does not mean every AI system is inherently insecure or that every reported attack is AI-generated. Attribution requires evidence. Security reporting should distinguish confirmed AI involvement from suspected use, and should not attribute an incident to AI merely because an attacker used sophisticated techniques.

The NCSC's assessment is that AI will continue to make elements of cyber intrusion more effective and efficient, increasing pressure on defenders. Its findings support investment in cyber resilience, but they do not prove that every proposed AI defence will be effective.

Source: NCSC Annual Review 2025: The Cyber Threat to the UK; Impact of AI on Cyber Threat from Now to 2027.

5. Who manages and oversees the investment?

The Department for Science, Innovation and Technology is responsible for the Government's wider technology policy, while the NCSC is the UK's technical authority for cybersecurity. The NCSC operates as part of GCHQ and works with government, industry, academia and international partners.

The announced NERV Centre is intended to sit within the NCSC. The National Cyber Shield is described as a partnership with the UK cybersecurity industry. The public-sector AI defence programme is intended to strengthen the technology estate used by public bodies.

These delivery arrangements create different accountability questions. A national vulnerability centre needs clear rules for validating findings and communicating them to affected organisations. A challenge fund needs transparent selection and monitoring of grant recipients. A public-sector programme needs evidence that participating organisations have improved their security and can maintain those improvements after the initial funding.

The Budget announcement establishes the intended programmes and headline amounts, but does not supply a complete set of public implementation contracts, annual spending profiles or independent outcome evaluations. Those details will be important in determining whether the announced commitments are delivered as planned.

Security considerations may legitimately restrict the publication of operational details. They should not prevent the Government from reporting aggregated expenditure, procurement principles, performance measures and independent assessments of effectiveness.

6. The UK cybersecurity industry: jobs, investment and regional opportunity

The Government's Cyber Security Sectoral Analysis 2026 provides useful context for the economic opportunity surrounding the new investment. It estimates that the UK cybersecurity sector generated £14.7 billion in revenue and approximately £9.1 billion in gross value added in 2025, employing around 69,600 full-time-equivalent workers across 2,603 firms.

The report also records £184 million raised across 47 investment deals in dedicated cybersecurity firms during 2025, a reduction from £206 million across 59 deals in 2024. Small firms accounted for £84 million of the 2025 investment total, while micro firms attracted £43 million.

These figures show that Britain already has a substantial cybersecurity industry. They also indicate that investment conditions can change from year to year. The new challenge fund could support innovation, but it should not be presented as the sole cause of future industry growth or employment.

London remained the largest regional destination for cybersecurity investment in 2025. The Government's plans for a Centre for Cyber and AI Excellence in Manchester and continued support for the Cyber-AI Hub at Queen's University Belfast could help broaden regional capability. Whether they succeed will depend on sustained investment, employer demand, training pathways and the ability of local businesses to participate.

Smaller firms may have valuable specialist expertise but limited resources to meet complex procurement requirements. The challenge fund should therefore make its application process proportionate, explain selection criteria clearly and ensure that security standards do not unnecessarily exclude innovative small businesses.

Source: Cyber Security Sectoral Analysis 2026, DSIT.

7. The costs that headline figures do not reveal

The £154 million commitment provides a headline measure of the Government's planned cyber-defence investment, but it is not a complete estimate of the lifetime cost of operating these capabilities. AI-enabled security requires computing, software, specialist personnel, integration with existing systems, maintenance, training, incident response and ongoing testing.

There may also be costs associated with false positives, incorrectly prioritised vulnerabilities, service interruptions during remediation and the need to replace or reconfigure systems that cannot work with the new tools. Where systems become dependent on external AI providers, future subscription prices and changes to contractual terms may affect long-term affordability.

The Government should publish a costed implementation plan covering initial expenditure and ongoing operating costs. It should distinguish grants and contracts from internal staff costs, and identify which resources are newly funded rather than transferred from existing budgets.

The £25 million AI and Cyber Security Challenge Fund is included within the £60 million National Cyber Shield allocation. It must not be counted as an additional £25 million on top of the £60 million. Likewise, the £154 million headline total already includes the £70 million, £60 million and £24 million components.

Value for money should be judged by comparing the programme with realistic alternatives. If an inexpensive software update or improved access control addresses a vulnerability, an AI system may be unnecessary. If AI substantially reduces the time needed to detect and investigate complex attacks, it may justify its additional cost.

The relevant comparison is the total cost of achieving a defined security outcome, not simply the purchase price of a tool.

8. What could go wrong?

False alarms and missed threats

AI detection systems can flag harmless activity as suspicious or fail to recognise genuine attacks. Too many false alerts can overwhelm security teams, while missed detections create false confidence. Evaluation should measure both outcomes under realistic conditions.

Unsafe automation

An AI agent may be authorised to isolate devices, block network traffic, change configurations or initiate other defensive actions. These capabilities can accelerate incident response, but an incorrect action could disrupt a hospital service, disable a business system or block legitimate users.

High-impact actions should therefore have clearly defined permissions, safeguards against cascading failures, logging and recovery procedures. Human approval should be required where the potential consequences justify it, while carefully tested automation can be used for lower-risk and reversible tasks.

Attackers targeting the AI system

AI security tools may themselves become targets. Attackers could attempt to manipulate the data on which they rely, exploit vulnerabilities in connected services or misuse the permissions granted to automated agents. A defence system that can take actions across an organisation needs particularly strong access controls and monitoring.

Dependence on a small number of suppliers

Concentrating security capability in a small number of AI providers can create commercial and operational dependencies. Price changes, outages, contractual disputes or changes in product availability could affect the ability to maintain protection.

Contracts should address continuity, access to logs, data protection, incident notification, portability and exit arrangements. Where practical, organisations should retain alternative response procedures that do not depend entirely on a single AI service.

Unequal protection

Large organisations generally have greater resources to employ cybersecurity specialists and procure advanced tools. Smaller businesses, charities, schools and local services may struggle to maintain basic protections. A programme that benefits only well-resourced organisations could leave significant gaps in the national security picture.

Public funding should therefore be evaluated not only by the capabilities it develops, but also by whether it helps raise the security baseline across organisations with fewer resources.

9. What are the alternatives to AI-led defence?

The first alternative is to invest in established cybersecurity fundamentals. Timely software updates, multi-factor authentication, strong access controls, secure backups, network segmentation and tested recovery plans can prevent or limit many attacks. These measures are not glamorous, but they address weaknesses that AI cannot automatically fix.

The second alternative is to strengthen the human cybersecurity workforce. Skilled analysts, incident responders, system administrators and security engineers are needed to interpret alerts, investigate incidents and maintain infrastructure. AI may reduce some repetitive tasks, but it does not eliminate the need for qualified staff or clear responsibility.

The third approach is to improve shared services and coordinated vulnerability management. Better information-sharing, common security standards and support for smaller organisations could help reduce widespread weaknesses without requiring every organisation to purchase its own advanced AI platform.

These approaches can be combined. AI may be most effective when it supports reliable infrastructure, trained people and established security practices. The Government should fund AI where it addresses a defined gap, rather than treating it as a replacement for conventional defence.

10. A practical action plan

Immediate priorities: the first 12 months

• DSIT and the NCSC: publish an implementation plan for the £154 million package, including delivery milestones, accountable leads and an annual spending profile.
• NERV Centre leadership: define how vulnerabilities will be verified, prioritised, communicated and tracked through to remediation.
• National Cyber Shield programme leaders: establish challenge-fund criteria that reward measurable security improvements rather than novelty alone.
• Public-sector technology leaders: assess existing cyber weaknesses, backup arrangements and recovery plans before introducing additional AI tools.
• Independent evaluators: establish baselines for detection accuracy, response times, false alerts, operational disruption and total cost.

Medium-term priorities: years two and three

• Conduct independent tests of AI-enabled tools against conventional security methods and publish suitable aggregate findings.
• Measure whether participating public bodies remediate vulnerabilities faster and recover more effectively from incidents.
• Expand training and collaboration through regional cyber and AI centres, with clear routes into employment for new entrants and career changers.
• Review the challenge fund to determine whether supported businesses deliver usable technology, attract further investment and improve security outcomes.
• Test supplier continuity, data protection, access controls and recovery arrangements through regular exercises.

Long-term priorities: beyond year three

• Continue funding tools that demonstrate sustained benefits and discontinue or redesign projects that repeatedly fail to meet agreed targets.
• Integrate successful capabilities into wider national cyber resilience arrangements rather than leaving them as disconnected pilots.
• Maintain human expertise, conventional security controls and contingency procedures alongside AI-enabled automation.
• Ensure smaller organisations and regions outside the largest technology centres can benefit from training, shared services and proportionate support.

11. How should success be measured?

The Government should report measurable outcomes rather than relying on announcements, procurement totals or the number of AI tools deployed. The NERV Centre should be assessed on the proportion of priority vulnerabilities verified, the time taken to notify affected organisations, the proportion remediated and the reduction in exposure to known weaknesses.

The National Cyber Shield should be evaluated on detection accuracy, the time needed to investigate incidents, the proportion of alerts that lead to useful action, false-positive rates and the safety of automated responses. Independent tests should examine performance under changing conditions, including attempts to manipulate the tools themselves.

For public-sector defence, measures should include recovery time after simulated incidents, the proportion of critical systems covered by tested backups, remediation of priority vulnerabilities and the ability of essential services to continue operating during disruption.

Financial reporting should compare actual expenditure with budget, include ongoing operating costs and explain material differences from the original plan. Grant evaluations should identify which projects produced useful results, which failed and what was learned.

Results should be reviewed regularly, with annual public reporting and independent assessments at major milestones. Security-sensitive details may need to remain confidential, but the Government should still publish enough information for Parliament and the public to assess whether the investment is working.

12. Accountability: who answers if it fails?

Responsibility is distributed across government, technology suppliers and the organisations deploying AI. The NCSC provides technical guidance and national cyber incident support. DSIT oversees relevant technology policy and funding. Public-sector organisations remain responsible for managing their own systems, suppliers and operational risks.

Where private companies develop or supply publicly funded technology, contracts should establish who is responsible for testing, reporting incidents, correcting defects and maintaining security. Public bodies should not assume that buying an AI product transfers their accountability for the services they operate.

Parliamentary scrutiny and appropriate financial audit should examine whether money was spent as intended and whether the programme delivered its stated objectives. Independent technical evaluators should assess whether the tools are effective and whether their limitations have been accurately described.

The Government has also indicated that it will consult on tackling insecure technology at source and considering where liability for cyber breaches should sit. This is an important policy question, but a future consultation should not be treated as a law already in force.

Accountability must cover failure as well as success. When an AI system misses an attack, produces an unsafe response or creates a new vulnerability, there should be a documented process for investigation, correction and learning.

13. What remains unanswered?

The Budget identifies the main funding commitments, but several details remain necessary for a complete assessment. These include the annual spending schedule, procurement arrangements, challenge-fund award criteria, the precise scope of the NERV Centre and the distribution of support among public-sector organisations.

Further evidence is needed about how the Government will compare AI-enabled defence with conventional alternatives, how it will test automated actions that could disrupt essential services and how it will manage dependence on external technology suppliers.

The programme's public value will also depend on whether smaller organisations benefit, whether cybersecurity skills develop outside established centres and whether the new capabilities measurably reduce risk rather than simply increase the volume of monitoring.

These gaps do not prove that the programme is ineffective. They identify the information needed to judge its design, implementation and results as the work progresses.

Conclusion: stronger cyber defence requires evidence, not technology alone

The UK's £154 million cyber-defence package reflects the importance of protecting critical infrastructure, public services and the wider economy against evolving digital threats. The National Cyber Shield, NERV Centre and public-sector AI defence programme could strengthen detection, prioritisation and incident response if they are implemented effectively.

But AI is not a substitute for secure systems, trained staff, clear accountability and tested recovery plans. It can improve the speed and scale of defensive work, while also introducing new dependencies, errors and opportunities for attackers. The public benefit will depend on how well those trade-offs are managed.

The most important next step is a transparent implementation framework that separates announced commitments from actual expenditure and measures the results against credible baselines. Independent testing, clear responsibility and support for organisations with fewer resources should be built into the programme from the outset.

Britain's digital future will not be protected simply by funding more advanced technology. It will be protected when that technology demonstrably reduces vulnerabilities, limits disruption, supports essential services and gives the public confidence that those responsible can explain what went wrong when defences fail.