NCSC: AI-powered ransomware on UK critical infrastructure by 2028
NCSC annual review: AI-powered malware highly likely in ransomware against UK critical infrastructure by 2028. Attack vectors, data and what to do now.

On 3 November 2026 the UK National Cyber Security Centre (NCSC, part of GCHQ) presented its annual State of the Cyber Nation report, together with a separate assessment of how artificial intelligence will change threats over the next two years. Its sharpest forecast, reported by every outlet that covered the launch: it is “highly likely” that by 2028 AI-powered malware will be used in a ransomware attack on the UK’s critical national infrastructure.
NCSC chief executive Richard Horne spoke of a “real prospect of an unprecedented rise” in incidents in the coming months. The core of the report, however, is not a new class of attack: according to the NCSC, AI does not invent new techniques, but allows vulnerabilities that have been tolerated for too long to be found and exploited at a scale that was once unimaginable.
What the assessment says
- Ransomware against critical infrastructure: according to Infosecurity Magazine, the report names power stations, airports and hospitals as likely targets of financially motivated groups.
- States: Russia and China will “routinely” use AI in their operations by 2028. China is described as the “most sophisticated and sustained” cyber challenge and the best positioned to use AI against UK infrastructure; Russia runs hybrid campaigns that deliberately stay below the threshold of armed conflict; Iran keeps targeting the UK.
- How attacks change: AI will increase the speed, scale and efficiency of existing methods rather than autonomously carry out end-to-end attacks (a quote from the report carried by CSO Online). Ransomware, social engineering and vulnerability exploitation are the areas named.
- Worst case: The Register and Computer Weekly report the report’s pessimistic scenario, with more frequent and sophisticated attacks, harder attribution and the “realistic possibility” of an autonomous attacker, beyond its operator’s control, able to self-replicate.
The technical data
The report stresses that most attacks, state-backed ones included, do not use advanced AI but known, already patched vulnerabilities. The Record reports that even intrusions attributed to state actors often go through internet-facing applications and flaws for which a patch existed but was not deployed in time.
According to CSO Online, the NCSC estimates that up to 88% of attacks start from just four vectors, which map to well-known MITRE ATT&CK initial access techniques:
- phishing (T1566);
- exploitation of public-facing applications (T1190);
- use of valid accounts (T1078);
- external remote services such as VPNs and remote access (T1133).
Other figures quoted: 1,830 attack reports received in the past year, 80 of them nationally significant (Computer Weekly), and an estimated cost of £14.7 billion a year for UK organisations. On the pace of fixes, Horne noted that major vendors fix hundreds of vulnerabilities in a single update and that the Linux kernel fixed more than a thousand last month (The Record). For the PHP2Shell flaws, according to Infosecurity Magazine, the NCSC sent 110,000 proactive notifications and Early Warning alerts to UK organisations.
On tooling, The Register reports growing abuse of ARTEX, a Chinese open source AI red-teaming platform, repurposed in real attacks: this comes from that outlet alone, not from the report.
Operational technology and Europe
Horne focused on systems that control physical processes: electricity switching, sluice gates and dams, manufacturing lines, heating and refrigeration controls, where security investment has lagged. His example was last month’s attacks on refrigeration systems in French supermarkets. SecurityWeek also recalls the December 2025 attack on Poland’s power grid, attributed by the UK and EU member states to Russia’s FSB. CSO Online cites two NCSC alerts from 2026: in July on vulnerable routers exploited by Russian groups against energy and telecom networks, in August on increased targeting of OT systems.
For Italian and European companies the lesson is not British: the vectors are the same, and the measures the NCSC calls for match what NIS2 requires of essential and important entities (vulnerability handling, access control, business continuity, supply chain security). The same applies to healthcare, named among the likely targets.
The UK government’s response
The report also presents AI as a defensive tool: Project DROMOS, started as a four-week pilot on government code, found hundreds of potential issues in live environments, including critical vulnerabilities. Funding includes £60 million for the first phase of Cyber Shield (national-scale defensive AI agents, delivered in stages over five years according to The Register), £70 million for the NERV Centre (real-time exposure and vulnerabilities across critical infrastructure), £40 million for a unit dealing with AI agent security, and a 75% discount on Cyber Essentials certification for SMEs. According to the NCSC, the structural advantage remains with defenders, who own the environment and can instrument it, but only with investment and discipline.
What to do now
- Cut patching times on internet-facing systems, starting with already exploited vulnerabilities (CISA KEV catalogue) and end-of-life systems, to be replaced or isolated.
- Take off the internet what should not be there: management interfaces, OT controllers, refrigeration and HVAC systems, as the NCSC explicitly asks.
- Protect access: phishing-resistant MFA on VPNs, remote access and email, administrative privileges reduced to what is needed.
- Back up important data, kept separate from the network and verified with restore tests.
- Detection: monitor failed logins, logins from unusual locations and connections denied at the perimeter; with faster attacks, reaction time matters more than ever.
How ARXDOME helps
ARX Expose shows what an attacker sees from outside: exposed services, known vulnerabilities on public services and accounts involved in breaches. ARX Shield lists open vulnerabilities per machine, with already exploited ones highlighted, and lets you isolate a suspicious machine from the network. ARX Sentry collects failed logins, bursts of attempts and VPN errors in real time, while ARX Horizon weights vulnerabilities by actual and probable exploitation.
Last updated:


