Britain’s National Cyber Security Centre says attackers are likely to use AI-enabled capabilities to exploit known vulnerabilities in legacy technology at scale across UK critical national infrastructure by 2028. The assessment puts the emphasis on a familiar security problem—unpatched or difficult-to-replace systems—rather than on a sudden arrival of independently operating cyber weapons.
A Guardian report published November 3 described the outlook more specifically, attributing to the NCSC a view that AI-powered malware was highly likely to have been used in a ransomware attack on UK critical national infrastructure by then. The agency’s publicly available June material supports the broader 2028 warning, but does not use the ransomware formulation. That leaves an important boundary around what has been independently documented: the NCSC has publicly forecast AI-assisted exploitation of known legacy weaknesses, while the ransomware scenario is reported as an NCSC assessment by the Guardian.
Known weaknesses, accelerated by AI
In a June 17 speech to the Royal United Services Institute, NCSC chief executive Richard Horne said the agency assessed that AI-enabled cyber capabilities would likely be used against known legacy-technology vulnerabilities across critical national infrastructure by 2028. The agency’s published speech transcript is significant because it identifies the expected route to harm: adversaries using increasingly capable tools to make better use of security gaps that already exist.
That is a more concrete threat model than claims that AI will simply make hacking effortless. Reconnaissance, identifying likely weak points and producing or adapting malicious code are all tasks that can be accelerated when an attacker has software able to process large quantities of technical information. The Guardian reported that the NCSC expected AI to help with those activities. The likely consequence is not that every attack becomes novel, but that established techniques can be applied more quickly, more broadly or with less specialist effort.
Legacy technology is especially exposed because replacement can be costly and disruptive, and because a system may sit within a long chain of suppliers, operators and connected networks. The NCSC’s public wording does not identify particular industries, targets or products. It refers instead to critical national infrastructure as a whole, avoiding a claim that any individual service has been singled out.
Ransomware remains a plausible concern because it turns access to a network into an immediate disruption and extortion problem. But “AI-enabled exploitation” and “AI-powered ransomware” should not be treated as interchangeable labels. The first can describe AI helping an attacker locate or exploit a weakness; the second describes a particular kind of criminal operation. Nor does either phrase establish that malware will be fully autonomous.
An existing pressure, not a distant hypothetical
The warning arrives against a substantial existing workload. In its June account of Horne’s speech, the NCSC said it had managed more than 200 incidents affecting UK critical national infrastructure and its supporting ecosystem in the year to May 2026. About 75% were believed to be linked to state actors.
Put on a simple annual average, more than 200 cases amounts to at least roughly 17 incidents a month, though incident volumes will not be evenly distributed and the figure includes the infrastructure-supporting ecosystem as well as infrastructure itself. The number does not measure successful ransomware events or prove that AI was involved. It does show that the NCSC’s forecast is being made in an environment where essential systems and their surrounding suppliers are already regularly targeted.
The agency also said in June that it was seeing attackers discuss repurposing newly released AI-powered security tools on cybercrime forums. That detail is a reminder that commercial and defensive advances can diffuse quickly. A tool built to help security teams inspect systems or identify flaws can be useful to an intruder if it reduces the time needed to understand a target.
The defensive test is recovery as well as prevention
The most direct implication of the NCSC’s assessment is operational rather than futuristic. Organizations responsible for vital services need a clear account of old systems, the internet-facing services around them, and the known vulnerabilities they cannot immediately eliminate. Where replacement is not feasible, compensating controls, segmentation and close monitoring become more important because they can reduce the chance that a single compromise spreads through an organization.
Recovery capability is equally material in a ransomware scenario. A security program that focuses only on detecting an initial intrusion can still fail if it cannot restore systems and services after an attacker has gained privileged access. The public NCSC material does not prescribe a sector-by-sector response in the documents cited here, but its focus on known vulnerabilities directs attention toward basics that are frequently less glamorous than new AI products: patching where possible, isolating fragile systems and preparing to operate through an incident.
The June assessment also complicates a popular framing of AI security as a contest between offensive and defensive algorithms. AI can help defenders process technical data and find flaws, as the Guardian noted in reporting comments from University of Surrey cybersecurity professor Alan Woodward. Yet defensive advantage depends on whether organizations can act on those findings before an attacker does. In infrastructure environments with aging equipment and constrained maintenance windows, discovering a weakness is not the same thing as being able to remove it quickly.
The NCSC’s public benchmark is therefore less a prediction of a science-fiction attack than a deadline for reducing a stockpile of familiar risks. By 2028, its assessment says, AI is likely to make exploiting those risks feasible at greater scale across systems whose failure can affect services beyond the network that was first breached.
