What the NCSC found
The NCSC, part of GCHQ, said AI will help attackers pick targets, locate gaps in IT networks and write malicious code. In a ransomware attack, criminals break into a victim’s systems, encrypt files and demand cryptocurrency to unlock them and return stolen data. The UK’s definition of CNI spans energy, transport, defence, food, health and finance.
The report names Chinese cyber activity as the UK’s most sophisticated and sustained challenge, and expects China, Russia and other states to use AI routinely by 2028. It judges that AI will mainly scale up existing attack methods rather than uncover entirely new vulnerabilities.
Why the warning matters for business
Horne said AI will let flaws ‘that have been tolerated for far too long’ be found and exploited at a scale once thought impossible. He also said many UK organisations and supply chains had relied for decades on ‘doing tech on the cheap’.
The financial exposure is already large. UK government research cited by the NCSC puts the cost of malicious cyber activity to UK organisations at an estimated £14.7bn a year. Jaguar Land Rover reported about £196m in direct costs after a 2025 attack that also hit its suppliers. Horne said around 75% of the more than 200 incidents affecting UK CNI over the previous year could be linked to state actors.
AI’s speed at finding flaws is already visible. Recently, governments and companies rushed out security updates after AI-assisted research found two critical flaws in PHP, a programming language behind about 70% of websites.
What could happen next?
Horne told organisations to act now or later explain why shelves are empty, lights are off, and businesses have collapsed despite clear warning signs. The NCSC’s recommended first steps are keeping devices updated, limiting administrator access and backing up important data.
Households are also being asked to prepare. In October, the government urged people to keep up to nine litres of bottled water and three days’ food for emergencies, including cyber incidents.
Access to advanced AI may decide who finds flaws first. Alan Woodward, professor of cybersecurity at the University of Surrey, said defenders need cutting-edge models to patch weaknesses before attackers exploit them. OpenAI and Anthropic have given some companies early access to their most powerful models to help find serious IT faults.