NCSC: what AI-powered attacks mean for UK organisations
Key takeaways
- The NCSC says AI-powered malware will highly likely feature in a ransomware attack on UK critical national infrastructure by 2028
- The warning is about old, tolerated weaknesses being found and exploited faster, not about brand-new attack types
- Cyber Essentials should be the starting point, not the end goal, with resilience built up according to size, complexity and risk
What does the NCSC mean when it says AI-powered attacks will hit UK critical infrastructure within two years, and what should an organisation do about it? The short answer: the National Cyber Security Centre, part of GCHQ, expects AI-powered malware to feature in a ransomware attack on the UK's critical national infrastructure by 2028, and says the weaknesses AI will exploit are ones organisations already know about.
What the warning actually is
The claim comes from the NCSC's State of the Cyber Nation report, published on Tuesday 3 November. Its central assessment is that, within the next two years, AI adoption by hostile actors is "highly likely" to increase the scale and efficiency of their operations, including ransomware attacks, social engineering and vulnerability exploitation.
Ransomware is malware that locks or encrypts an organisation's data and demands payment for its return. Social engineering is the manipulation of people, through convincing emails or calls, into handing over access. Vulnerability exploitation is the use of flaws in software or configuration to get in.
Computer Weekly reports that the NCSC's assessment says Russia and China will routinely use AI-powered cyber tools to achieve their geopolitical goals. It also reports a worst-case scenario in which frontier AI, meaning the most capable general-purpose models, could self-replicate and conduct attacks at will against UK targets.
The detail that matters most is what the NCSC says AI is not doing. In his launch speech, chief executive Dr Richard Horne is set to say that the expected rise in incidents is "not because AI has invented new ways to attack us". It is because "vulnerabilities that have been tolerated for far too long can now be discovered and exploited at a scale that was once unimaginable."
How it works: the old lock, the faster burglar
An analogy helps. Imagine a street where half the houses have a back door that has never quite closed. For years, burglars had to walk the street and try each handle. That limited the damage, because attention and time were scarce. Now give the burglars a tool that tries every handle on every street in an afternoon. The doors have not changed. The cost of finding them has collapsed.
That is the NCSC's argument. Legacy systems, unpatched software, and equipment that was connected to the internet for convenience are the loose back doors. AI lowers the effort of finding and using them.
The report singles out one area in particular. Computer Weekly reports that organisations will need to secure the technology controlling physical infrastructure, including electricity switches, dams, manufacturing lines, and heating and refrigeration controls, where investment in security and resilience has been less of a priority. Dr Horne's stated stakes are blunt: "when food isn't on shelves, lights aren't on, businesses have gone under", leaders will have to explain themselves to investors, shareholders, employees, customers and suppliers.
Who is affected, and what government is doing
The direct concern is critical national infrastructure, but the NCSC addresses organisations across the UK, and the article is tagged as written for small and medium sized organisations. Dr Horne calls this "a moment of choice".
The government response is partly financial.
- £60 million for the first phase of a national Cyber Shield
- £70 million for a new National Exposure and Real-Time Vulnerability (NERV) Centre, to identify vulnerabilities across critical infrastructure and the economy
- a new AI Cyber Security Centre of Excellence in Manchester, run by the NCSC
It is also partly legislative. The source points to the Cyber Security and Resilience Bill and a forthcoming National Cyber Action Plan. IT teams in regulated sectors should read our explainer on NIS2 and the UK Cyber Security and Resilience Bill to see how the existing network and information systems regime is expected to change.
The NCSC also frames AI as an opportunity, not only a threat. The report describes AI as a critical chance to improve collective cyber resilience, which is the logic behind funding AI for defence.
What the limitations are
The warning has boundaries worth stating plainly.
First, it is a forecast. "Highly likely" is NCSC probabilistic language, and the 2028 date describes a window, not a scheduled event. The worst-case scenario of self-replicating frontier AI is exactly that: a worst case, not the central assessment.
Second, the advice is unglamorous. There is no new product to buy. Dr Horne's asks are to invest adequately in technology, avoid connecting technology to the internet unnecessarily, and treat Cyber Essentials as the starting point for security rather than the end goal. Organisations should then build further resilience according to their size, complexity and risk. That is hard in practice because it competes for budget with everything else, which is precisely the "tech on the cheap" habit the NCSC is criticising.
Third, the sources give no breakdown of how many organisations are exposed or how many incidents to expect. The prediction is directional. Planning has to rest on the principle that known weaknesses become cheaper to exploit, not on a forecast number.
What this means in practice
For IT teams, the practical reading is that the baseline matters more, not less. Dr Horne says Cyber Essentials should be the starting point for cyber security, not the end goal. For the difference between the basic and audited tiers, see Cyber Essentials vs Plus: cost, renewal and requirements.
Several steps follow directly from the report's framing:
- Inventory what is internet-facing. The NCSC explicitly calls out technology that is unnecessarily connected. If a system does not need to be reachable, disconnect it.
- Fix known weaknesses first. If AI makes discovery cheap, the old backlog becomes the attack surface.
Organisations should know how they would respond if ransomware hit before it does.
For US readers, the relevance is indirect: the NCSC's assessment names Russia and China as routine users of AI-powered cyber tools for geopolitical ends.
Where this is heading
The funding announced, from the £60 million Cyber Shield to the £70 million NERV Centre, signals that government intends to fight AI-assisted discovery with AI-assisted defence, and to map exposure across the economy rather than leave it to individual firms. The Cyber Security and Resilience Bill and the National Cyber Action Plan are the next things to watch, because they will turn the NCSC's advice into obligations. Meanwhile, the NCSC's own message is that the cheapest time to close a known gap is before something faster than a human starts looking for it.
Frequently asked questions
What does the NCSC say about AI-powered attacks by 2028?
The NCSC's State of the Cyber Nation report says AI-powered malware will highly likely be used in a ransomware attack on UK critical national infrastructure by 2028, as hostile actors use AI to increase the scale and efficiency of their operations.
What does "tech on the cheap" mean?
It is Dr Richard Horne's phrase for years of underinvesting in technology and security. He says the era of getting away with it is over, because AI will expose those weaknesses at unprecedented speed and scale.
Does the NCSC say AI is creating new kinds of attack?
No. Dr Horne says the risk is not that AI has invented new ways to attack, but that long-tolerated vulnerabilities can now be discovered and exploited at a scale once thought unimaginable.
What should organisations do first?
The NCSC urges adequate investment in technology, removing unnecessary internet connections, and treating Cyber Essentials as the starting point rather than the end goal, then building resilience according to size, complexity and risk.