UK NCSC warns AI ransomware is highly likely to target infrastructure by 2028

The UK’s National Cyber Security Centre assessed that AI-powered malware is highly likely to feature in a ransomware attack on the country’s critical national infrastructure by 2028. Its November 3 annual review and separate AI assessment call on infrastructure operators to confront ageing systems and weak maintenance alongside emerging attack capabilities.

In his launch speech, NCSC chief executive Richard Horne said AI accelerates the discovery and exploitation of longstanding weaknesses. He highlighted unsupported systems, excessive access and procurement driven by price. The same investment problem extends to technology controlling electricity switching, water gates, dams, manufacturing, heating and refrigeration.

The retrospective figures cover incidents across the cyber threat landscape. Between September 1, 2025 and August 31, 2026, the NCSC received 1,831 attack reports. After triage, 337 incidents required support; 80 of those were nationally significant.

The annual review says attackers have begun incorporating AI tools, but observed uses enabled no novel class of attack. Automating defence faces additional constraints. In a September 21 explanation, the NCSC’s Dave Chismon noted that defenders must show patches or firewall changes will not disrupt business and obtain approval, while attackers can concentrate on technical success.

The separate intrusion assessment explores baseline, pessimistic and optimistic paths to 2028. An uncontrolled, self-replicating attacker appears in the pessimistic scenario. These are explicitly plausible possibilities for planning, rather than predictions or accounts of completed events.

Reading the assessment and planning defence

The phrase “highly likely” uses the government’s PHIA probability yardstick, corresponding to approximately 80–90%. Official guidance says most intelligence judgements lack quantitative data; verbal ranges help avoid misleading precision. This is an assessed likelihood, not a measured attack probability for an individual company.

The same guidance separates probability from analytical confidence. Confidence concerns the soundness of the evidence and how stable a judgement may be as circumstances change. A high likelihood therefore does not establish certainty about every supporting fact or identical exposure across organisations.

Research questions remain behind the national defence plan. The NCSC’s October 20 AI problem book makes human–AI collaboration a cross-cutting theme: which tasks suit different autonomy levels, where judgement remains essential and how oversight survives greater scale. It calls for comparisons with human-led approaches to establish where AI reduces risk or introduces new operational problems.

The NCSC’s announcement cites investments announced by the Chancellor the previous week: £60 million for Cyber Shield’s first phase and £70 million for the National Exposure and Real-Time Vulnerability Centre, or NERV. NERV will use AI to identify vulnerabilities across infrastructure and the economy. A Manchester centre is intended to develop government AI-security expertise.

Chismon’s September explanation distinguishes AI analysis supporting a human from actions that change systems. It assesses risk through the scope affected, business criticality, testing and recoverability, alongside the power of an automated action to alter systems. Cyber Shield initially targets detection and blocking, within a five-year programme.