AI is already inside your organisation – do you know what it’s doing with your data?

Fortinet is a Business Reporter client

From the NHS to the energy grid, critical infrastructure is rapidly adopting AI – but successfully balancing safety with efficiency means understanding how AI operates.

AI adoption at scale is real across UK industry, particularly in critical national infrastructure (CNI). For example, the NHS is rolling out AI assistants to 505,000 staff, designed to help clinicians draft documents and analyse data more efficiently so they can spend more time on patient care. AI is also changing the UK rail industry, reshaping how trains, stations and operational infrastructure are managed. In the energy sector, it is being used to forecast power demand, optimise supply and enable better predictive maintenance.

AI is clearly a valuable business opportunity. But as adoption moves from experimental to everyday work processes, new and significant risks are appearing. For example, in the US, a bank employee’s use of an unapproved AI tool to process customer information resulted in the bank having to make a formal data breach disclosure. Shadow AI can become a governance issue even without a conventional cyber-attack.

Organisations need to accept a new reality: they cannot manage AI if they do not know where and how it is being used. But the solution is not to stop AI adoption. Instead, CNI leaders adhering to good governance should be asking: “What are our employees doing with AI, and why?”

Visibility is the new necessity

You cannot build an effective AI policy from assumptions.

A policy that simply tells employees which tools they cannot use may look strong on paper but it won’t explain why people are looking for alternatives in the first place. If employees use AI to reduce repetitive work or analyse information, blocking their ability to use it will only encourage workarounds. The underlying demand will still be there.

Visibility changes that equation. If you understand which AI tools are being used, by whom, for what purposes and with what data, it is much easier to distinguish productive adoption from genuinely risky behaviour.

Visibility also makes policy more useful. Instead of issuing blanket restrictions, organisations can create clear boundaries around particular use cases and data types, enabling employees to decide what they can do safely and what they need to avoid.

Organisations must also take visibility seriously for compliance reasons. The UK’s Cyber Security and Resilience Bill requires organisations to be transparent about cyber-incidents affecting essential services. They must provide an initial notification within 24 hours and a fuller report within 72 hours. But unless they know how employees are using AI, they may be totally unaware that a cyber-breach has taken place.

Good insiders, bad insiders

Not every employee using AI is a malicious threat. Fortinet’s 2025 Insider Risk Report found that 77 per cent of organisations had experienced insider-driven data loss in the previous 18 months, yet only 16 per cent of incidents involved confirmed malicious intent.

Greater visibility changes how organisations can assess this insider risk. An employee who pastes sensitive information into an AI assistant while drafting a report may have made a serious security mistake but that does not necessarily make them a malicious actor. Unlike, perhaps, someone systematically downloading large quantities of information before leaving the organisation.

This distinction matters. The well-intentioned employee needs intervention at the point of risk – for instance, a warning when they are about to share sensitive data with a public AI tool. This is not the same as punishing every single breach of policy.

Intelligent technical controls

Education and cultural changes cannot, on their own, achieve total security. Technical controls are also needed. Data loss prevention and endpoint protection tools can be fine-tuned for AI workflows.

For instance, Fortinet’s AI security capabilities give organisations a live, risk-scored inventory of the AI tools in use, including what data flows into them. Sensitive data going into prompts is flagged, and strong defences against AI-related attacks such as prompt injection and data poisoning are provided. At the same time, people can be steered towards sanctioned, enterprise-grade AI services, rather than a “shadow AI” tool they have chosen, keeping data inside the organisation’s boundaries and out of external model training.

Accountable leaders

Good security should enable AI adoption, not obstruct it. But technology will only take an organisation so far. Ultimately, this is a leadership issue. A useful benchmark is the UK government’s Cyber Governance Code of Practice. This encourages clear lines of responsibility between security, IT and HR, so that when something is flagged, it is clear who owns the decision about what happens next and what should happen – training, escalation or policy change.

Most importantly, CNI boards must accept that accountability cannot sit with the security team. Leaders must step up to own the issue, while setting expectations in AI and everyone’s role in managing its risks.

Conclusions

Organisations that adopt AI successfully will understand the opportunities AI brings, but they will also address the challenges. Here are three actions that you can take to strengthen your organisation’s use of AI:

Fortinet delivers integrated data loss prevention and AI security across the entire digital infrastructure, from IT and cloud to the operational edge. It helps CNI organisations say yes to AI with confidence.

The organisations that benefit most efficiently from AI will be those with the clearest view of why and how AI is being used. Good visibility is what makes responsible innovation possible. For CNI organisations, that visibility can be the difference between AI being an unmanaged source of risk and being a transformative capability.

by John Stringer, Senior Director Product Management at Fortinet and Chris Parker, Director Government Strategy UK at Fortinet