The UK's National Cyber Security Centre has put a date on a worry many defenders already had. In its Annual Review 2026, published on Nov. 3 under the title State of the Cyber Nation, the agency judges that AI-powered malware is "highly likely" to feature in a ransomware attack on UK critical national infrastructure (CNI) between now and 2028.

This is a forecast, not a report of an attack. No incident of this kind is described in the review. What the NCSC has done is set out how it expects AI to change the criminal ransomware business over the next few years. Its main conclusion is fairly plain: the danger is not some exotic new kind of attack. It is the familiar attacks, carried out faster, in greater numbers and against the same old unpatched systems.

Below we cover what the assessment actually says, what it does not say, and what organisations should do about it.

What did the NCSC actually conclude?

The review's AI threat assessment rests on a few core judgements. The NCSC says AI is "almost certain" to act as a force multiplier for attackers. It also says uptake will be uneven, with some groups adopting it much faster than others. The agency judges that AI's biggest effect is to speed up existing tactics, techniques and procedures (TTPs, the methods attackers already use) rather than to create new ways in. And it stresses that humans remain essential to running attacks. AI is described as an accelerant, not an autonomous replacement for criminal operators.

On ransomware specifically, the NCSC says cybercriminals' adoption of AI will "highly likely" increase the scale and efficiency of ransomware operations. It singles out one route: open-source AI tools originally intended to assist security researchers, which it expects criminals to repurpose to automate their existing workflows.

The headline judgement on CNI follows from that. AI-powered malware is "highly likely" to appear in a ransomware attack on UK critical national infrastructure by 2028.

Is the target power stations and hospitals, or CNI in general?

The coverage differs a little here. Infosecurity Magazine reported that financially motivated hackers are expected to use AI-powered malware against UK power stations, airports and hospitals by 2028. The NCSC's own assessment page refers to critical national infrastructure in general. Those sectors are reasonable examples of CNI, but readers should not treat the review as naming specific sectors as confirmed targets.

Why does patch speed sit at the centre of this?

The most practical judgement in the review concerns vulnerability discovery. The NCSC says AI-assisted vulnerability discovery is "highly likely" to give attackers a chance to target systems before they are patched. That leaves defenders in a race to shorten their patch timelines.

NCSC CEO Richard Horne put it in concrete terms at the review's Nov. 3 launch event in London. As reported by Infosecurity Magazine, he said the risk is not that AI has invented new ways to attack, but that vulnerabilities "tolerated for far too long" can now be discovered and exploited at scale, and he pointed to end-of-life systems kept running for "just one more year." Horne warned of a "real prospect of an unprecedented rise in the number of cyber incidents."

That framing matters for anyone sizing up their own exposure. An AI-boosted attacker is not mainly dangerous because of a brilliant new exploit. The danger is that it can find the forgotten VPN appliance, the unsupported server or the unpatched edge device faster than an overstretched IT team can.

What does the worst case look like?

According to The Register, the review includes a pessimistic scenario for 2028 that raises the risk of autonomous attackers. The NCSC says it "is a realistic possibility" that such an entity could self-replicate and conduct cyberattacks at will. That is a different, more tentative judgement than the "highly likely" one about AI malware in ransomware. Human-directed criminals using AI tools are the central expectation. Fully autonomous attack agents are the tail risk.

Who else is in the picture?

The review also covers nation-state actors. The NCSC judges that China is "highly likely" the state actor best positioned to use AI against UK CNI. Infosecurity Magazine reported that Russia and China are both expected to use AI routinely. Those attributions are the NCSC's assessments as reported. The review does not tie them to any specific incident.

Two other findings are relevant to day-to-day security work. First, the NCSC warns of a widening gap between organisations that can adopt AI for defence and those that cannot. Second, it judges that "shadow AI" is "highly likely" to leave an organisation open to exploitation, because staff moving company data into unsanctioned AI tools means the organisation loses visibility of how that data is handled.

What is the government doing about it?

Around the review, the government announced funding and support measures. Infosecurity Magazine dates the funding announcements to Oct. 24, and the Cyber Essentials discount to the review event. Figures are given as each outlet reported them.

  • Cyber Shield: Security Journal UK, The Register and Infosecurity Magazine all report £60 million for a national Cyber Shield. Infosecurity Magazine describes it as the first phase, including a £25 million Challenge Fund. The Register describes Cyber Shield as an agentic cyber defence system to be delivered in stages over five years.
  • NERV Centre: The Register, Security Journal UK and Infosecurity Magazine report £70 million for the NERV Centre.
  • National Agentic Security Unit: The Register and Infosecurity Magazine report £40 million for this unit, to respond to uncontrolled AI bots.
  • AI Cyber Security Centre of Excellence: Security Journal UK, The Register and Infosecurity Magazine report a new centre in Manchester.
  • Cyber Essentials discount: The Register reports a 75 percent discount on the first Cyber Essentials certificate for micro, small and medium-sized firms.

The Register also reported on Project DROMOS, which began with a four-week pilot analysing government code repositories and, per the NCSC, uncovered hundreds of potential security issues. It is an early example of defenders using automated discovery against their own software before attackers do.

Horne also called for an end to "tech on the cheap", meaning technology bought or built without security designed in. That argument points as much at vendors and procurement teams as at the IT staff who inherit the results.

What It Means for Defenders

No new vulnerability or patch comes with this review, so there is no fixed version to install. The NCSC's assessment does, however, point to clear priorities.

Who is affected: UK critical infrastructure operators are the explicit focus. But the reasoning applies to any organisation running internet-facing systems, and it applies most to those on the wrong side of the NCSC's AI "divide": small businesses, under-resourced public bodies and suppliers to larger targets. Federal contractors and CMMC shops outside the UK should read it as a well-sourced forecast of where the ransomware ecosystem is heading in general.

Shorten patch timelines: This is the NCSC's core point. If AI shrinks the gap between a flaw being found and being exploited, then monthly or quarterly patch cycles for internet-facing systems become a real liability. Rank exposed edge devices, remote access services and public-facing applications first.

Retire or isolate end-of-life systems: Horne named long-tolerated vulnerabilities and end-of-life systems as the risk. Take an inventory of anything that no longer gets security updates. Then either replace it, or cut it off from the internet and from your most important networks.

Know your attack surface before someone else maps it: AI-assisted discovery works in both directions. Regular external asset discovery and vulnerability scanning of your own estate gives you the same view an attacker would build. The government's DROMOS pilot applies the same idea to its own code.

Get a handle on shadow AI: Find out which AI tools staff are using, what data they send to them, and whether those tools have been approved. Put in place a usage policy and an approved list of tools.

Use the Cyber Essentials discount: UK micro, small and medium-sized firms that have not yet certified can, according to The Register, get 75 percent off their first certificate. Cyber Essentials covers basic controls such as patching, secure configuration, access control and malware protection. Those are the same basics that AI-accelerated attackers will be probing.

Keep ransomware fundamentals current: The NCSC judges that AI speeds up existing techniques rather than inventing new ones. So the established defences still apply: offline and tested backups, multi-factor authentication on remote access, network segmentation and a rehearsed incident response plan. The difference is that there will be less time to get them in place.

Sources