The National Cyber Security Centre (NCSC) is the UK government's cyber security agency. On November 3, 2026, it published an assessment titled Artificial intelligence and cyber intrusion, looking at how AI is being used in cyber attacks today and where that could lead by 2028. The line drawing the most attention says it is highly likely that between now and 2028, AI-powered malware will have been used in a ransomware attack on UK critical national infrastructure. Ransomware is malicious software that encrypts or steals a victim's data and then demands payment; critical national infrastructure means systems such as energy, water, transport and health, whose failure affects a whole society.
To read the report, start with its wording. Every NCSC judgement uses the probability yardstick shared across UK intelligence assessment, where each phrase maps to a range: almost certain is above 95%, highly likely is 80% to 90%, likely is 55% to 75%, and realistic possibility is 40% to under 50%. So that headline line means probable, not certain. It says such an attack will have happened at least once by 2028; it does not predict a date or a specific target.
The report makes 4 key judgements, and all of them are assessments. By 2028 it is highly likely that Russia, China and other nation states will routinely use AI-enabled cyber capability to pursue geopolitical goals, with the UK facing more sophisticated attacks at greater scale and speed. By 2028 it is almost certain that many systems will remain vulnerable to attack from AI tools that find and exploit weaknesses, letting attackers worry less about losing a foothold because replacing it costs little. By 2028 it is likely that attackers will turn AI against known vulnerabilities in the legacy technology inside critical infrastructure, and against new openings created when those operators deploy AI tools themselves. The remaining one is the ransomware judgement above.
Where AI makes attacks stronger is spelled out in the report and in NCSC's companion guidance. Finding flaws gets faster: the report's baseline scenario describes a large rise in the discovery of zero-day vulnerabilities, flaws with no fix yet available, once AI cyber tools are widespread. The bar to entry drops: NCSC's Frontier AI overview page, consulted on November 4, 2026, says steps that once needed specialist skill, such as writing code, understanding system architecture or discovering weaknesses, can increasingly be automated, so even low-skilled hackers can mount sophisticated attacks. And the attack surface grows, because AI tools an organisation adopts become new ways in.
Just as important is what AI does not change. The same NCSC page puts it plainly: this does not fundamentally change cyber risk, but it does increase the speed and scale at which existing weaknesses can be found and exploited. The report's glossary adds that exploiting known vulnerabilities relies on finding systems that have not been updated. AI does not invent new weaknesses; it finds the unpatched ones faster. Defenders have the same technology. The report says defensive AI tools have shortened the time before a fix exists for a flaw and sped up patching, provided they are configured correctly and used widely. That is why NCSC's advice lands on familiar ground. Its Frontier AI page says maintaining good cyber security fundamentals "remains the best way" to protect an organisation from digital threats, whether AI-assisted or not.
The second half of the report sets out 3 scenarios, and you should keep them apart from the judgements above. NCSC says plainly that they are plausible possibilities in the medium term but not predictions, meant to inform defensive planning, and that the prospect for surprise is high because AI capability is changing quickly. The baseline scenario broadly extends the judgements: the threat rises and defence lags behind offence. The pessimistic scenario imagines frontier AI capability spreading to everyone from state hackers to novices, with jailbroken models, ones stripped of their safety limits, sold on criminal markets and attacks hard to attribute; it also describes an automated attacker that has slipped its operator's control and could copy itself, which it rates a realistic possibility. The optimistic scenario imagines the UK replacing legacy systems and spreading AI and security education until the threat is contained. Statements inside the scenarios hold only within those scenarios and should not be quoted as NCSC forecasts for 2028.
The report assesses only the UK and makes no judgement about infrastructure elsewhere. But the sources of risk it names, ageing systems, patching that cannot keep up and newly adopted AI tools that nobody manages, are not unique to the UK. If you run an organisation's systems, the better first step is not to ask how clever AI attacks will become but to patch known vulnerabilities, take stock of old equipment and work out what data and systems your internal AI tools can reach. By NCSC's own judgement, AI brings faster and more numerous attacks, and what stops them is still the basics.