NCSC Warns AI-Cyber Will Scale Attacks on UK Critical Infrastructure by 2028

NCSC Warns AI-Cyber Will Scale Attacks on UK Critical Infrastructure by 2028

The UK's National Cyber Security Centre (NCSC) says the country should prepare for a faster and more scalable cyber threat as nation-state operators adopt artificial intelligence. Its 2026 assessment does not describe a new class of attack happening today; it says familiar weaknesses, exposed services, and poor resilience remain the practical route into critical systems.

What the assessment says

The NCSC's 3 November 2026 annual review says threat actors are already incorporating AI tools into attacks, while none of the observed uses has yet created a novel class of attack. The agency's separate assessment of AI and cyber intrusion projects that, by 2028, China, Russia, and other nation states are highly likely to use AI-cyber routinely to pursue geopolitical goals.

The expected change is operational tempo. AI-assisted operators can search for weaknesses, generate tooling, process reconnaissance, and adapt their activity across more targets. The NCSC says attackers are likely to apply that capability against known vulnerabilities in legacy critical-national-infrastructure technology and against new attack surfaces created by the deployment of AI tools inside those environments.

Why defenders should act before the forecast arrives

The warning is not a reason to wait for an “AI attack” signature. The NCSC's incident-management findings reinforce a more immediate lesson: sophisticated actors still frequently gain initial access through known and preventable weaknesses. An AI-enabled operator can make an existing control failure more consequential, but patching, asset inventory, identity protection, segmentation, and monitoring remain the first line of defence.

For critical-infrastructure teams, the practical priority is to reduce the number of decisions an attacker can make after the first foothold:

• Maintain an authoritative inventory of internet-facing systems, legacy devices, suppliers, and the AI services connected to operational environments.
• Prioritise remediation of known exploited vulnerabilities and verify that fixes reached the vulnerable asset, not just the central patch system.
• Separate IT, operational technology, management, and AI development environments with explicit identity and network controls.
• Monitor for abnormal discovery, credential use, service-account activity, and rapid changes across many hosts; these signals matter whether the operator is human-led or agent-assisted.
• Test recovery and supplier-response procedures, including the loss of an AI service, identity provider, remote-management platform, or critical third party.

The NCSC's message is ultimately about resilience. A threat actor using AI does not make basic security controls obsolete; it reduces the time defenders have to notice and contain a failure. Organisations that close known gaps, limit blast radius, and practise incident response now will be better positioned when automated reconnaissance and exploitation become routine.

References

1. Artificial intelligence and cyber intrusion
2. NCSC Annual Review 2026: The cyber threat to the UK
3. NCSC Annual Review 2026: Incident management