The UK National Cyber Security Centre has published an assessment of how AI-enabled intrusion tools could change the cyber threat facing the country up to 2028. The assessment judges that the scale and speed of attacks on UK systems is highly likely to increase by 2028. Analyst reading: the sharpest exposure sits in legacy critical national infrastructure technology and in the new attack surface created when AI tools are deployed inside that infrastructure. The assessment examines current AI-enabled threat trends and projects the threat to 2028 through three scenarios, which it describes as plausible possibilities rather than predictions.
What the assessment judges
The NCSC uses its probability yardstick throughout, so its wording carries weight. It judges that by 2028 it is highly likely Russia, China and other nation states will routinely use AI-driven cyber capability to pursue geopolitical goals, with the UK facing an increase in the scale and speed of sophisticated attacks. It also judges it almost certain that many systems will remain vulnerable to AI tools that identify and exploit weaknesses, and that attackers display less caution about losing access once an intrusion is discovered, because replacing an access path costs them little.
Two judgements are the most specific for infrastructure operators. The first is that attackers are likely to apply AI-assisted techniques against known flaws in legacy critical national infrastructure technology, as well as against new weaknesses introduced by deploying AI within those environments. The second is that it is highly likely AI-powered malware will have been used in a ransomware attack against UK critical national infrastructure by then.
Three scenarios, one common driver
The assessment builds three scenarios. The baseline case judges that the global threat has almost certainly increased, with defensive AI shortening the window between a flaw being discovered and being patched. The pessimistic case assumes capability spreads faster than defences can absorb it. The optimistic case assumes development slows, allowing the UK to replace ageing systems and build a workforce that handles AI securely.
The pessimistic scenario is the one defenders should read closely. It describes a growing set of less controlled actors, including criminal groups and novices, as frontier AI spreads, and what the assessment calls a realistic possibility: an automated actor, using frontier AI, that is no longer under its operator’s control. It also describes attribution becoming harder, since proxies and noise make it difficult to tell who is responsible for an intrusion. The NCSC describes each scenario as a plausible possibility in the medium term, not a prediction.
Where the exposure sits
Three areas stand out. First, known flaws in legacy critical national infrastructure technology, which the assessment judges attackers are likely to target. Second, the new attack surface created when AI tools are deployed inside critical infrastructure. Analyst reading: each integration adds access paths, credentials and permissions that an attacker could abuse. Third, espionage and prepositioning. The pessimistic scenario says AI-enabled espionage makes it easier for threat actors to remain undetected. Analyst reading: that raises the chance that access already exists in networks that look clean.
The optimistic scenario depends on a workforce that understands both AI and security well enough to deploy it safely. The baseline scenario says defensive capability in UK critical national infrastructure lags behind offensive capability.
What defenders should do first
The assessment does not prescribe these steps. The priorities below are analyst guidance, drawn from its judgements, in rough order of urgency.
Analyst guidance: start with an inventory of legacy and internet-reachable systems in operational environments, and rank them by how quickly they could be exploited and how hard they are to patch. The assessment judges attackers are likely to target known flaws in this legacy technology.
Analyst guidance: shorten the path from vendor fix to tested deployment. The assessment says internal patching and testing processes will need to evolve to keep pace with faster vulnerability discovery, so the test-and-deploy stage needs the same attention as detection.
Analyst guidance: treat every AI tool you deploy as part of your attack surface. Review its permissions, the credentials it holds, the data it can reach and the interfaces it exposes, as you would for any privileged service.
Analyst guidance, drawing on the assessment’s concern about prepositioning: assume some intrusions may already have happened quietly. Run threat hunts for persistence and for unusual credential use, particularly in environments where long-term access would be most valuable to an espionage actor.
Analyst guidance: invest in training for staff who deploy and run AI systems, so that secure configuration is routine rather than an afterthought.
What remains unknown
The assessment does not say when any of the scenarios will materialise, which sectors will be hit first, or how quickly defensive AI will mature. It also states that the prospect for surprise is high, given how quickly AI development is changing. Analyst reading: readers should treat its scenarios as planning inputs rather than as a timeline.
Frequently Asked Questions
Will AI-enabled cyber attacks on the UK increase by 2028? The NCSC judges it highly likely that the scale and speed of attacks on UK systems will increase by 2028. It reaches this through three scenarios, which it describes as plausible possibilities rather than predictions.
Which UK systems does the NCSC see as most exposed to AI-enabled intrusion? The assessment judges attackers are likely to apply AI-assisted techniques against known flaws in legacy critical national infrastructure technology. It also points to new weaknesses created when AI is deployed inside that infrastructure.
What should critical infrastructure defenders do first about AI-enabled threats? The assessment itself does not prescribe steps, but analyst guidance is to inventory legacy and internet-reachable systems and shorten the path from vendor fix to tested deployment. Treating every deployed AI tool as part of the attack surface and hunting for quiet persistence are also recommended.
Contains public sector information licensed under the Open Government Licence v3.0 (UK NCSC).