What did the NCSC actually warn about?
The National Cyber Security Centre published its annual assessment, State of the Cyber Nation, on 3 November 2026. The report reviews the agency's tenth year, covering 1 September 2025 to 31 August 2026. Its most consequential judgement concerns the two years ahead.
The NCSC says it is 'highly likely' that between now and 2028, AI-powered malware will have been used in a ransomware attack on UK critical national infrastructure. Under the UK definition, CNI covers energy, transport, defence, food, health and finance. The agency's warning puts power stations, airports and hospitals in scope.
Richard Horne, the NCSC's chief executive, framed the finding as the product of neglect rather than a sudden leap in attacker skill. 'For decades, we have collectively got away with doing tech on the cheap,' he said in remarks released with the report. He also warned of 'a real prospect of an unprecedented rise in the number of cyber incidents' in the coming months.
The wording matters. The report does not identify a target, a date or an attacker. It sets a two-year window and a confidence level, then locates the cause in underinvestment and ageing technology.
The NCSC is part of GCHQ, the UK's signals intelligence agency, and its assessments use a standard probability scale. That makes this a policy document as much as a technical one, written for boards and operators of essential services as well as security teams.
The same scale is used throughout. The report applies 'highly likely' both to the 2028 malware judgement and to its assessment that China is best positioned to use AI against critical infrastructure.
The launch release also lists the government's response: new funding for a Cyber Shield and a vulnerability centre, a security bill and a national action plan. Those measures are discussed below, alongside what the report says about their limits.
Is AI creating new attacks, or speeding up old ones?
The report's answer is mostly the second. AI is 'unlikely to fundamentally change the goals of hostile actors', it says, but 'highly likely to increase the speed, scale and efficiency' of their operations.
The NCSC describes AI as a force multiplier across the attack lifecycle, including reconnaissance, social engineering, vulnerability discovery and coding for exploit development. The biggest impact is expected to come from accelerating existing tactics rather than inventing new attack vectors.
Adoption is uneven. Skilled operators gain most from models, while weaker actors gain less. Human operators 'highly likely remain essential' for directing, validating and adapting attacks, and the report says losing control of an agentic system would probably hinder a threat actor rather than help it.
The report is blunt about how most attacks still work. For large organisations, 88% of the incidents the NCSC handled came from four initial access routes: public-facing applications, phishing, misuse of valid accounts and external remote services. Known vulnerabilities and stolen passwords remain the main way in.
That continuity is the report's central irony. It compares 2026 with 2015, when Sandworm attacked Ukraine's power grid, the US government suffered the Office of Personnel Management breach and TalkTalk was hacked. The adversaries and their basic methods have not changed as much as the technology around them.
The report warns against both complacency and tunnel vision. 'We cannot spend our time obsessing about the latest AI-driven threat, but equally we can no longer remain complacent,' it says. Attacks on programmable logic controllers are cited as evidence that the gap between modernised and lagging organisations is widening.
Why does the report single out China, Russia and Iran?
China is described as 'the most sophisticated and sustained cyber challenge facing the UK'. The report says China highly likely has both the intent and the best position to use AI for operations against critical infrastructure and other strategic targets. Its frontier model programmes are expected to sharpen vulnerability identification and targeting.
The report also references sanctions. In December 2025, the UK sanctioned two Chinese technology companies, i-Soon and Integrity Technology Group, over malicious cyber activity. Integrity was also assessed by the FBI to be responsible for activity attributed to the Flax Typhoon group.
Russia is labelled the 'most audacious' cyber threat. The NCSC says Russian campaigns deliberately fall below the threshold of armed conflict. It cites an attack on Poland's energy grid in December 2025, attributed to Russia's FSB, which could have left 500,000 people without electricity.
The report points to joint advisories as evidence of coordination. In April 2026, the NCSC and 15 international partners published guidance on tactics used by China-linked actors. In July 2026, it and 18 partners published methods used by FSB Centre 16, which was exploiting vulnerable routers.
Iran-linked groups are cited for a different reason: a willingness to cause indiscriminate disruption at scale. The common thread is not one campaign but shared access to commercial intrusion tools and similar routes into poorly maintained systems.
For UK organisations, the practical implication is that the threat is not only criminal. State actors and ransomware gangs often exploit the same unpatched services, even when their goals differ.
What would an AI-assisted ransomware attack look like?
Ransomware already follows a familiar sequence: gain access, encrypt systems, demand cryptocurrency and threaten to leak stolen data. The NCSC expects AI to lower the cost and raise the volume of each stage, particularly for criminals targeting less resilient organisations.
The report also warns that open-source AI tools built for security researchers are likely to be repurposed by criminals. That, it says, would cut operating costs while increasing attack volumes against weaker targets.
The refrigeration incidents show how digital access becomes physical disruption. On Monday 2 November, supermarkets in Ireland, Sweden, Estonia, Belgium and other European countries reported fridge units being warmed remotely. The Guardian reported temperatures above 50C in affected units.
The pattern is not new. In early October, hundreds of French supermarkets were hit overnight, spoiling fresh and frozen food estimated to be worth millions of euros. The BBC reported that the attacks exploited flaws in remote refrigeration controls that researchers had flagged months earlier.
The UK was not untouched. Horne told BBC Radio 4's Today programme that the NCSC knew of 'a couple of instances where vulnerable technology had been exploited in the UK', and that staff had worked through the night to alert organisations. The BBC reported that some members of the British Association of Independent Meat Suppliers had been affected as collateral.
Horne's wider point is that the same exposure applies to electricity switching, sluice gates, dam operations, manufacturing lines and heating controls, as the Associated Press reported. A compromised IT system is no longer only a data problem.
That said, the report's own evidence suggests most ransomware still succeeds through ordinary weaknesses. Its case studies describe known vulnerabilities, stolen credentials and social engineering rather than autonomous AI campaigns. Sophistication tends to appear after initial access, not before it.
What has cybercrime cost the UK so far?
The report estimates that malicious cyber activity costs UK organisations about £14.7 billion a year, based on government research. It notes that the figure is roughly equivalent to the size of the UK's cyber security sector.
Jaguar Land Rover's 2025 incident is the case study. The carmaker reported about £196 million in direct cyber-related costs, with disruption spreading to suppliers and retail operations. Suppliers reported operational and workforce pressure as production schedules slipped. The report links the fallout to UK GDP growing just 0.1% in the third quarter of 2025.
The agency's incident figures give a sense of the workload:
Those numbers cover only what was reported. The report cautions that incident reporting does not follow a clean year-on-year trend, and that many events never reach the agency.
Ransomware remains 'highly disruptive', the report says, with extortion, data theft, operational disruption and long recovery timelines often arriving together. Preparation, particularly logging and incident response planning, is said to determine outcomes.
- •1,831 reports of cyber attacks affecting organisations
- •337 incidents that required NCSC support
- •80 incidents classed as nationally significant
- •8,500 organisations notified of suspected compromises through the Early Warning service
What is the government doing about it?
The NCSC's launch release lists measures announced by the chancellor the previous week. They include £60 million for the first phase of a national Cyber Shield, £70 million for a National Exposure and Real-Time Vulnerability Centre, and an AI Cyber Security Centre of Excellence in Manchester run by the NCSC.
Cyber Shield is the most concrete. It is jointly delivered by the NCSC and the Department for Digital, Culture, Media and Sport, and aims to build a national agentic cyber defence ecosystem over five years. The first operating stage focuses on detection and blocking, which the report says is more amenable to automation than autonomous remediation.
Project DROMOS shows what that looks like in practice. A four-week pilot used frontier models to review government code repositories; the programme then expanded to selected environments and surfaced hundreds of potential issues, including critical vulnerabilities, with human oversight. It is the defensive mirror of the AI-assisted vulnerability discovery the report warns about.
Two policy vehicles are still in progress: the Cyber Security and Resilience Bill and a National Cyber Action Plan. Security minister Dan Jarvis said the annual report shows 'cyber threats are growing more sophisticated', and pointed to the Cyber Shield and AI centre as the response.
Supply chains are part of the plan. The government's Cyber Resilience Pledge includes a commitment to roll out Cyber Essentials across supply chains, on the reasoning that a weak supplier can expose the organisations that depend on it.
Horne's asks of organisations are less dramatic. He wants proper investment in technology, systems taken off the internet when they do not need to be connected, and Cyber Essentials treated as a floor rather than a finish line. Leaders who skip those steps may have to explain 'when food isn't on shelves, lights aren't on, businesses have gone under'. He told the BBC that the NCSC intervenes where it can, but that organisations ultimately have to make the choice themselves.
Defenders also need access to frontier models, according to Alan Woodward, professor of cybersecurity at the University of Surrey. He told The Guardian that AI can help find and patch flaws, but that defenders will lose the race without access.
The sums are modest against the damage. The £130 million in new funding is a fraction of the £14.7 billion annual cost estimate, although the two measure different things: capital for specific capabilities versus an economy-wide loss. The ratio still explains why the report treats the problem as an investment gap.
How strong is the warning, and what should you watch next?
The confidence language is deliberately calibrated. 'Highly likely' sits below 'almost certain' in the NCSC's scale, and the assessment is a judgement about probability rather than a schedule. The agency is not saying a specific plant will be hit in a specific month.
The claim is also hard to test. The report does not define 'AI-powered malware', and attributing AI involvement after an attack is difficult. It notes that proxies and commercially available tools complicate attribution. A future incident could be described as AI-assisted without public evidence either way.
The scenario chapter is explicit about its status. It imagines a 2028 in which frontier AI is widely available, attribution is harder and an automated actor operates beyond its owner's control. The report says the purpose is not to predict the future but to show the cost of inaction.
It also concedes that the picture is incomplete. In that scenario analysis, it says the situation is likely to be worse than currently known, with many variables still to be discovered.
Three things are worth watching. First, whether the Cyber Security and Resilience Bill and the National Cyber Action Plan turn the assessment into duties and deadlines. Second, whether the NERV Centre and Cyber Shield become operating services rather than funding lines. Third, whether AI-assisted vulnerability discovery produces a patch wave that smaller organisations cannot absorb.
The international picture matters too. The report argues that no country can secure the digital future alone, and treats the Five Eyes partnership as the cornerstone of UK cooperation on standards, attribution and joint guidance.
For organisations, the checklist is unchanged and unglamorous: patch known flaws, remove unnecessary internet exposure, limit administrator rights, keep offline backups and adopt passkeys or multi-factor authentication. The report's argument is that these basics, not exotic AI defences, will decide how much damage the next incident causes.