1. AI ransomware warning: what the NCSC said on November 3
The United Kingdom's National Cyber Security Centre (NCSC) warned on Tuesday, November 3, that the country's critical national infrastructure should prepare for ransomware built around AI-powered malware by 2028. The agency, which belongs to the signals-intelligence service GCHQ, made the assessment in its annual State of the Cyber Nation report. It described the rise in hostile use of artificial intelligence over the next two years as highly likely.
The warning goes beyond one type of attack. According to the report, hostile actors who adopt AI are highly likely to expand both the scale and the efficiency of their operations. That covers ransomware, social engineering and the exploitation of software vulnerabilities. Social engineering means tricking people into handing over access, and ransomware locks systems or data until a payment is made.
On the day of the release, the NCSC also posted a short message on the social platform X announcing its annual review. It said the UK faces a moment of choice and that the future of cyber resilience is not set. The framing matters: the agency is presenting the coming years as an outcome that organizations can still influence, rather than a fixed forecast of decline.
2. What is the NCSC and why does its annual report matter?
The National Cyber Security Centre is the British government's technical authority on cyber threats. It sits within GCHQ, the country's signals-intelligence agency, and advises public bodies, companies and the public on how to defend networks. Its annual report is one of the main official yardsticks for how the threat picture is changing, so its wording is closely read by security teams, regulators and boards.
The phrase highly likely carries a specific weight in government intelligence language. Assessments are normally graded on a scale of probability, and a term at this end of the scale signals strong confidence rather than a loose possibility. That is why the agency's choice of wording attracted attention: it is a statement about expected direction, not an abstract caution.
Critical national infrastructure, in the British sense, covers the systems a society depends on to function, including energy, water, transport, health care and food supply. An attack that disrupts any of them reaches ordinary households quickly, which explains why the agency singles out this category when describing the stakes of AI-assisted ransomware.
For readers outside the security field, the report is useful mainly as a signal of priorities. When a national agency of this standing puts a date on a threat, boards and regulators tend to treat it as a reason to review budgets, and insurers and suppliers often follow by adjusting their own expectations of what counts as adequate protection.
3. Old weaknesses, new speed
NCSC chief executive Richard Horne delivered the message in prepared remarks for a speech launching the report. His central argument was that the danger comes less from novel attack techniques than from long-standing weaknesses that nobody has fixed. Artificial intelligence, in his account, has not invented a new way of breaking into systems.
What it changes is the rate at which existing flaws can be located and used. Horne said vulnerabilities that have been tolerated for years can now be found and exploited at an unprecedented scale. He warned of a real prospect of an unprecedented rise in the number of cyber incidents in the coming months, a notably short horizon compared with the 2028 date attached to AI-powered ransomware.
This distinction shapes the defensive advice. If the weak points are old ones, then much of the protection lies in basic maintenance and design choices that organizations have postponed, rather than in waiting for a new generation of security tools.
The point also reframes how organizations might think about risk. A company that has postponed patching an aging system has, until now, relied partly on the fact that finding the flaw required time and skill. If AI lowers that barrier, the protection that obscurity once offered shrinks, and the cost of delay rises accordingly.
4. The end of technology on the cheap
Horne used blunt language about the way many organizations have approached their systems. He said that for decades, people have collectively got away with doing technology on the cheap. The remark was aimed at the habit of treating security spending as a cost to be minimized, with the result that fragile systems stay in service because nothing has gone wrong yet.
He urged organizations to invest properly in their technology and to stop connecting systems to the internet when there is no need to. Each unnecessary connection widens the surface that an attacker, human or automated, can probe. Removing it is among the least expensive ways to shrink exposure.
Horne also addressed the Cyber Essentials scheme, the government-backed certification that sets out a baseline of security controls. He said organizations should treat it as a starting point rather than the end goal. In other words, passing the baseline should open the work of improving security, not close it.
He closed with a direct warning to leaders. Those who do not act, he said, may one day have to explain to investors, staff and customers why food is not on shelves, why the lights are not on and why businesses have gone under.
The argument is, at its core, an economic one. Cheap technology choices shift risk into the future, where it is paid for in disruption rather than in budget lines. Horne's remarks suggest the agency believes that the bill is about to arrive for organizations that have deferred it, and that customers and shareholders will expect an explanation when it does.
Reducing internet exposure is also practical advice for operators of older industrial equipment. Systems designed decades ago were often built on the assumption that they would sit on closed networks. Connecting them for convenience, without redesigning their protections, is exactly the kind of tolerated weakness the agency has in mind.
5. How much is the UK government spending to respond?
The government announced several funding and institutional measures in the week before the report. It committed £60 million, roughly $80 million, to the first phase of a national Cyber Shield. It also set aside £70 million, about $93 million, for a new National Exposure and Real-Time Vulnerability Centre, whose job is to identify weaknesses across critical infrastructure.
A third measure is a new AI Cyber Security Centre of Excellence in Manchester, to be run by the NCSC. Taken together, the three announcements show the state trying to build its own capacity for finding and closing weaknesses at the same speed that attackers are expected to find them.
The vulnerability centre in particular reflects the logic of Horne's argument. If AI lets attackers scan for flaws faster, defenders need continuous visibility of the same flaws, rather than periodic audits that are out of date by the time they are completed.
Money alone does not close the gap, and the agency's language makes that plain. Funding creates capacity, but the changes Horne describes, such as retiring unnecessary connections and treating baseline certification as a floor, depend on decisions taken inside each organization.
6. Are attackers already using AI?
Yes, and the evidence is mixed in its sophistication. In May, Google's Threat Intelligence Group said cybercriminals had used an AI model to find and weaponize a previously unknown flaw. The flaw let attackers bypass two-factor authentication in a popular open-source administration tool. It was described as the first confirmed case of AI-assisted zero-day development in the wild. A zero-day is a vulnerability that the software's maker has not yet had a chance to fix.
A separate study from the same month looked at more than 90,000 threads on cybercrime forums. It found that most criminals were using AI for spam and phishing rather than for sophisticated intrusions. That finding suggests the typical use today is cheap and high-volume, even if the most capable actors have begun to use it for something harder.
The two data points sit together uneasily, and the NCSC's forecast bridges them. Routine misuse is common now, the first genuinely advanced cases have appeared, and the agency expects the gap between them to narrow over the next two years.
Defenders are also expected to adopt AI. The same techniques that help attackers sort through large amounts of code for weaknesses can help defenders do that work first, which is part of the reasoning behind a dedicated AI-focused security centre run by the NCSC itself.
7. Governments and AI safeguards
The report arrived against a wider political discussion about how to govern advanced AI. In July, Foreign Secretary Yvette Cooper warned that governments must agree on safeguards before frontier AI systems transform warfare, crime and society. She wrote that the world cannot afford to wait for an AI equivalent of Hiroshima before it acts.
The NCSC's approach is more operational than diplomatic. Rather than calling for international agreements, it asks owners of critical systems to improve what they control directly: how systems are built, what they connect to and how much is spent keeping them secure.
8. What happens next
The near-term test is the period Horne described as the coming months, when the agency says a sharp rise in incidents is a real prospect. Organizations that run infrastructure will be judged on whether they have reduced unnecessary internet exposure and treated baseline certification as a floor.
The funded programs are at an early stage. The Cyber Shield has received money for its first phase only, and the vulnerability centre and the Manchester centre of excellence are new bodies still to show results. The 2028 date in the NCSC's assessment gives both government and industry a defined window in which to act before AI-powered ransomware against critical infrastructure becomes, in the agency's words, highly likely.