NCSC Expects AI-Driven Attacks on UK Infrastructure by 2028

Richard Horne, head of the UK’s National Cyber Security Centre, told an audience on Tuesday that Britain will likely face AI-enabled cyberattacks on its critical infrastructure by 2028. He was speaking as the NCSC published its annual report, according to SecurityWeek’s reporting of the Associated Press account.

What was said

Horne pointed to criminals and states, naming China and Russia, as increasingly using AI in their operations. He said AI-powered malware for ransomware attacks is a possibility, and that the technology will let “vulnerabilities that have been tolerated for far too long” be found and exploited “at a scale that was once unimaginable.”

The report calls Chinese activity the “most sophisticated and sustained cyber challenge facing the U.K.” It says Russia runs campaigns deliberately kept below the threshold of armed conflict, and Horne said Iran is attacking UK targets and individuals it sees as threats to its government.

Horne also cited the operational technology that runs electricity switching, water sluice gates, dams, manufacturing lines and refrigeration. He pointed to recent attacks on refrigeration systems in French supermarkets as an example of how easily civilian infrastructure can be hit. On Poland, the UK and EU member states have said Russia’s FSB was behind a December 2025 attack on the energy grid. Had it succeeded, 500,000 people could have lost power in winter.

On cost, the NCSC says malicious cyber activity costs UK organizations almost $20 billion a year. Jaguar Land Rover lost more than $260 million directly in 2025 after an attack that also hit suppliers.

The quieter line in the report

The headline is the 2028 prediction. The more useful sentence is further down: the report says AI is more likely to enhance and scale existing tactics and techniques than to find brand new vulnerabilities, while still being highly likely to raise the scale and efficiency of hostile operations.

Analysis: that framing matters for how organizations read the warning. If AI mostly speeds up the exploitation of flaws that already exist, the exposure is the backlog of unpatched, internet-facing and end-of-life systems, not some novel class of attack that needs novel defenses. Horne’s phrase about vulnerabilities tolerated for too long reads as a statement about that backlog.

The week’s news supports the point without proving it. The Associated Press account notes that governments and companies last week scrambled on urgent updates after two critical flaws, found with AI assistance, turned up in a programming language that underpins 70% of websites. The flaws could in theory allow unauthenticated access to sites. The pattern is faster discovery against software that is already everywhere, which is a patching-speed problem.

What the NCSC asks for

The advice is plain. GCHQ director Anne Keast-Butler said every UK organization needs to act now, and the NCSC listed keeping devices up to date, limiting administrator access, and backing up important files and data. Horne said leaders who have “got away with tech on the cheap” will have to explain empty shelves, dark buildings and failed businesses if they do not invest.

Nothing here is new, and that is the argument. The controls are old and dull, and the timeline for needing them just got a date attached.

Related coverage: PHP2Shell Hits Government Network with AI Post-Exploit Tools and AI Agents Ran INC and Gentlemen Ransomware Intrusions. More in the threat intelligence hub.