What the NCSC actually said
Some coverage rounds this off to "you have just over a year." The primary documents don't say that. NCSC Assessment's report, Artificial intelligence and cyber intrusion, looks at AI-enabled threat trends through 2028 using three scenarios: baseline, pessimistic and optimistic. The NCSC says the scenarios are plausible possibilities, not predictions. It adds that AI capability is changing quickly, so the prospect of surprise is high.
The four headline judgements, in the NCSC's probability language:
- Highly likely: by 2028, Russia, China and other nation states will routinely use AI-cyber capability for geopolitical goals. The UK would face more sophisticated attacks, at greater scale and speed.
- Almost certain: by 2028, many systems will still be vulnerable to AI-cyber tools that find and exploit weaknesses. Attackers will worry less about losing access because replacing it costs so little.
- Likely: by 2028, attackers will point AI-cyber capability at known vulnerabilities in CNI legacy technology. They will also target new attack surface created by AI tools deployed inside CNI.
- Highly likely: between now and 2028, AI-powered malware will have been used in a ransomware attack on UK CNI.
The report also says it uses the Professional Heads of Intelligence Assessment (PHIA) probability yardstick. On that scale, "highly likely" is roughly 80–90%. FourWeekMBA likewise describes it as a band of about 80% to 90%. That is a calibrated judgement, not a scheduled event.
It's about old weaknesses, found faster
The most useful line comes from NCSC chief executive Dr Richard Horne's launch speech. He says the coming rise in incidents is not because AI has invented new ways to attack, but because long-tolerated vulnerabilities can now be found and exploited at a scale once unimaginable.
For admins, that shifts the picture. You don't need to fear a mystical new class of malware. You need to fear your unpatched, forgotten and over-exposed systems being found much sooner.
The report's glossary supports this. A zero-day is exploited before a fix exists. It becomes a "known" or n-day vulnerability once the vendor ships a fix. Exploiting n-days depends on finding systems that haven't been updated. In the baseline scenario, the NCSC says defensive AI tools may shorten the zero-day-to-n-day window and speed up patching. It adds that this only helps if the tools are properly and widely configured. An organisation's internal patching and testing processes will also have to evolve to keep up with faster vulnerability discovery.
Put simply, faster discovery on both sides makes your patch pipeline the bottleneck.
The NCSC's recommendations
Horne urged organisations to make different choices now. The NCSC's announcement lists these:
- Fix underinvestment in technology. Address the failure to invest adequately in the systems you run.
- Don't connect things to the internet unnecessarily. Reduce exposure for systems that have no need for it.
- Treat Cyber Essentials as the starting point. It is not the end goal. Build further resilience according to your size, complexity and risk.
- Use what already exists. The NCSC says existing tools, standards and frameworks are evidenced to protect against both conventional and AI-powered attacks.
A practical reading for a Windows-heavy estate (my interpretation, not an NCSC checklist):
- Build or refresh an accurate asset inventory, including unsupported operating systems and applications.
- Find anything with an internet-facing management interface, such as RDP, remote admin consoles or VPN appliances, and ask whether it needs to be there.
- Check that your test-and-deploy cycle can handle a faster stream of fixes.
- Plan the replacement of legacy systems you can't patch quickly.
Nation-state context
The NCSC's baseline scenario names state-sponsored actors, such as China and Russia, as the greatest threat. In his speech, Horne is quoted saying China is highly likely to have the intent and be best positioned to use AI against UK critical national infrastructure. The report also suggests Russian actors have adapted open-source AI models rather than building their own from scratch.
The pessimistic scenario goes further. It describes proliferation of frontier AI to actors ranging from states to novice hackers, and even a self-replicating automated actor beyond its operator's control. That is a scenario, not the NCSC's expectation. The optimistic scenario describes the opposite: CNI sectors replacing ageing legacy systems, backed by workforce education. The NCSC frames the outcome as depending on choices made now.
Government money announced alongside
The warning came with funding announcements. The government is putting £60 million into the first phase of a national Cyber Shield, and £70 million into a National Exposure and Real-Time Vulnerability (NERV) Centre at the NCSC. It is also setting up a new AI Cyber Security Centre of Excellence in Manchester.
A separate GOV.UK press release adds detail:
- The Cyber Shield will use agentic AI and includes a £25 million challenge fund for UK AI and cybersecurity innovators.
- The NERV Centre is meant to build and maintain a national picture of vulnerabilities across CNI and the wider economy.
- There is £40 million for a National Agentic Security Unit to monitor and respond to uncontrolled bots, and £24 million to enhance public-sector cyber defences.
These are announced investments and intended functions, not capabilities that are already running.
Help for smaller UK businesses
The same release covers Cyber Essentials, the baseline the NCSC wants organisations to build on:
- Discount: a 75% discount for eligible micro, small and medium-sized businesses. They would pay between £80 and £125 plus VAT.
- First Step Fund: first come, first served, for businesses not certified in the last four years. It runs until March 2027 or until the funding is used up. The government estimates about 6,000 new businesses could benefit.
- Supplier grant: worth up to £7 million until March 2029. It offers fully funded first-time certification and tailored Cyber Advisor support to eligible UK SME suppliers of Cyber Resilience Pledge signatories. The government estimates about 8,000 businesses could benefit.
- Insurance: the release says certified organisations with turnover under £20 million also receive up to £25,000 of cyber liability insurance.
Eligibility conditions apply, so check them before budgeting around any of this.
A fair note of skepticism
Any national agency warning about future AI attacks is also making a case for funding and attention. The NCSC's own caveats are worth taking seriously: the scenarios are not predictions, and the report admits the prospect of surprise is high. Fourweek's coverage also points out that the NCSC's count of nationally significant incidents fell in the latest year. That is a reminder that forecasts and current incident counts are not the same thing.
Still, the advice is sound whatever happens with AI. Inventory your estate, patch faster, shrink your exposure and replace what you can't defend. None of that depends on a 2028 prophecy being right.
Bottom line
- The NCSC judges AI-powered malware in a ransomware attack on UK CNI "highly likely" between now and 2028. That is a probability judgement, not a certainty or a dated event.
- The core message is that AI speeds up the discovery and exploitation of old weaknesses. It has not invented new attack methods.
- Admins should focus on legacy systems, internet exposure, patch and test speed, and treat Cyber Essentials as a floor rather than a ceiling.