Close-up of a server rack with glowing blue lights in a data centre

The AI attack window is two years: what CAIOs should take from the NCSC's State of the Cyber Nation

On Tuesday 3 November 2026 the National Cyber Security Centre published its State of the Cyber Nation report with a blunt forecast for Chief AI Officers. AI-powered malware is highly likely to be used in a ransomware attack on the UK's critical national infrastructure by 2028. Hostile actors are also expected to use AI to scale social engineering and vulnerability exploitation within the next two years.

That is not a distant science-fiction risk. It is a board timeline. CAIOs who treat AI as a product programme only will miss the point. The same technology that is changing customer service, coding and decision support is changing how attackers find and exploit the weaknesses that organisations have lived with for years.

What the NCSC is actually saying

The NCSC is clear that AI has not invented brand-new ways to attack. The danger is speed and scale. Weaknesses that teams have tolerated for too long can now be discovered and exploited far faster than before. In his launch speech, NCSC CEO Dr Richard Horne called this a "Sliding Doors" moment for the UK: a short window in which leaders either invest in foundations or explain later why they did not.

His phrase for the old habit was "tech on the cheap". End-of-life systems kept for one more year. Upgrades pushed back as a cost saving. Access granted because it was convenient. Services bought on price alone, without asking whether the underlying technology was resilient. Monitoring skipped until something goes wrong. Those are not niche IT complaints. They are the foundations AI workloads and AI-enabled attacks both sit on.

Horne also stressed that Cyber Essentials is a starting point, not the finish line. Organisations need to build more resilience according to their size, complexity and risk. The same day, government announced a Cyber Essentials First Step Fund that cuts the cost of first-time certification for eligible micro, small and medium businesses by 75%, with a separate route for pledged organisations to fund supplier certification for free.

Why this belongs on the CAIO agenda

A Chief AI Officer's remit usually covers models, data, use cases and value. The NCSC report pushes a harder question into that remit: are the systems your AI programmes depend on actually fit to carry that weight?

Three links matter for CAIOs.

1. AI programmes inherit weak foundations. If your organisation is still running unsupported systems, wide standing privileges or poorly monitored suppliers, AI will not paper over that. It will make the exposure easier for someone else to find.

2. Attackers will use AI against your AI estate. Social engineering, credential theft and vulnerability hunting all get cheaper and faster with AI assistance. That hits the identity, API and data paths that AI products rely on.

3. Defence also runs on AI investment. The same government package that sits behind the Annual Review includes funding for a National Exposure and Real-Time Vulnerability (NERV) Centre, the first phase of a national Cyber Shield, and an NCSC AI Cyber Security Centre of Excellence in Manchester. Those are national tools. Boards will still expect their own AI leaders to know where exposure sits inside the firm.

Five moves CAIOs can make this quarter

Put foundational hygiene on the AI roadmap. Treat unsupported systems, identity hygiene, patch debt and supplier access as blockers to AI scale, not as someone else's backlog. If a model or agent cannot run safely on the current estate, that is an AI governance issue as much as a security one.
Map every AI system to the assets it can touch. For each production use case, list the data stores, APIs, SaaS tools and human workflows it can reach. If you cannot name them, you cannot defend them.
Set external limits on agents. Do not rely only on prompts or model-side guardrails. Decide what agents may read, change and call, and put those limits in the wider control plane.
Rehearse AI-amplified incident paths. Tabletop a ransomware or social engineering scenario where reconnaissance and exploit writing are accelerated by AI. Ask how fast finance, legal, communications and the AI team would lock down model access and data exports.
Brief the board in plain language. Use the NCSC's 2028 CNI timeline as the frame. The ask is not "more AI tools". It is urgent investment in the technology foundations that AI both needs and will stress.

What to leave out of the panic

This is not a call to freeze every AI project. The NCSC is equally clear that AI is one of the best chances the UK has to find and fix vulnerabilities faster. The Linux Kernel Project, for example, was cited as patching more than a thousand vulnerabilities in a month. The choice is between using AI to harden foundations and letting adversaries use it first.

It is also not a substitute for the Cyber Security and Resilience Bill work already under way. That Bill strengthens duties for essential services, managed service providers and others. The Annual Review is the threat picture that explains why those duties, and board investment, cannot wait.

The bottom line for CAIOs

By 2028 the UK's critical infrastructure is expected to face ransomware that uses AI-powered malware. The organisations that will cope are the ones that stop treating security basics as optional and stop treating AI leadership as separate from the estate the models run on. The NCSC has named the window. The question for every CAIO is whether their board has opened it yet.