NCSC Warns of AI-Powered Cyber Attacks: What UK Businesses Can Do

What this is, as of 3 November 2026

The UK National Cyber Security Centre (NCSC) has published Artificial intelligence and cyber intrusion, an assessment from NCSC Assessment on how AI is changing the way attackers break into systems. It covers current trends and then sets out three scenarios running to 2028. The NCSC says plainly that these are plausible possibilities and not predictions, and that the prospect of surprise is high.

This is not a vulnerability advisory. The report names no CVE identifiers, no vendor and no product version, so there is no fixed release to look for and no affected-software list to check yourself against. We have filed it under Security Watch because its practical message - attackers will find and use known weaknesses faster - is one worth acting on before it becomes an incident.

The report is written for cyber security professionals, large organisations and the public sector. If you run a smaller business the findings still apply, but they need translating into your own estate. The later sections do that.

Is there a fix, and which systems are affected?

There is no fix, because there is no single flaw. What the NCSC describes is a change in attacker capability. It judges that AI tools which find and exploit weaknesses will leave many systems exposed through to 2028, and it says widely available AI tools have likely increased zero-day discovery. A zero-day is a weakness for which no vendor fix exists yet. The report uses n-day for one that is known, because a fix has been issued.

In practice the systems most at risk are the ones attackers could already reach: anything exposed to the internet, anything running software that is overdue an update and anything that no longer receives updates at all. The report specifically points at known flaws in legacy technology and at the new attack surface created when AI tools are deployed in critical national infrastructure (CNI). If you are not part of CNI, the legacy and unpatched-exposure points still translate directly to you.

The NCSC also notes that defensive AI tools have shortened the gap between a zero-day becoming an n-day, and so have made patching quicker. That only helps if the fixes are actually installed. The report says organisations' own patching and testing processes will need to evolve to keep pace with faster vulnerability discovery.

What the report actually judges

The headline judgements all look ahead to 2028, so read them as forecasts rather than reports of events that have happened. In the NCSC's own probability language:

• Highly likely: Russia, China and other nation states will routinely use AI for cyber intrusion to pursue geopolitical goals, with the UK facing greater scale and speed of sophisticated attacks.
• Almost certain: many systems will remain vulnerable to AI tools that identify and exploit weaknesses. Because replacing a lost foothold is cheap, attackers can afford to be less careful about being caught.
• Likely: attackers will apply AI capability against known vulnerabilities in CNI legacy technology and against the new attack surface that AI deployment creates.
• Highly likely: AI-powered malware will have been used in a ransomware attack on UK CNI between now and 2028. This is a forecast, not a report that such an attack has already happened.
• The report also says defensive capability in UK CNI currently lags behind offensive capability, and that it is unclear whether defensive AI use will be enough.

What to do now

The report itself contains no numbered recommendations. The actions below come from separate NCSC guidance on vulnerability management and its Five Eyes statement on AI risk, and they are ordered so that each step makes the next one easier. Most of this needs administrator access to your systems or a conversation with whoever manages them for you. None of it needs special tooling.

1. List what is reachable from the internet. That means websites, remote access gateways, VPNs, firewalls and routers, mail servers and anything else a stranger could connect to. The NCSC's advice is to start with the external attack surface and work inward, because that is where an automated tool would look first. Remove any external access that nobody can justify.
2. Check each of those items for a vendor-supported version and a current update level. Where the vendor's support has ended there is nothing more to install, so the NCSC's advice is to replace the system or bring it back into support, particularly if it faces the internet.
3. Move to update by default, meaning updates are applied as soon as possible and ideally automatically, rather than waiting until a particular flaw looks urgent. The NCSC's best-practice timescales are within 5 days for internet-facing services and software, within 7 days for operating systems and applications, and within 14 days for internal or air-gapped services. It applies these whatever the severity rating and says to shorten them as far as you can. For internet-facing systems it suggests installing on a test environment or from a backup-protected state first, then rolling out. Safety-critical systems and operational technology may need a different approach, and business-critical systems need timescales balanced against availability.
4. Review sign-in. Require strong authentication on remote access, email and administrator accounts, and review who holds privileged permissions. The Five Eyes cyber agencies' joint statement lists identity and access controls among its core actions, and limiting access to critical systems to those who need it makes any stolen credential less useful.
5. Find out where AI tools and agent-style automation are already in use in your business, and what access they have to systems and data. The NCSC's Frontier AI page asks organisations to establish this. It is much easier to answer now than after an unexpected tool has been granted access to a mailbox or a file share.
6. Check that you could contain and recover from an incident. The Five Eyes statement expects breaches to happen and puts the emphasis on fast containment, so make sure someone knows who to call, how to isolate a machine from the network and where clean backups are held. Patching does not remove an attacker who is already inside, so if an internet-facing system has been unpatched for a long while, look for signs of compromise as well as installing the update.

What we do not know

The sources we checked do not link this report to any specific named incident, victim, vulnerability or piece of malware, and we are not aware of any public evidence tying it to active exploitation of a particular product. Where we say exploitation is a risk, we mean the general direction of travel the NCSC describes, not a confirmed campaign.

The pessimistic scenario in the report goes much further, including automated cyber actors that could slip out of operator control. The NCSC presents these as scenarios to plan around, not as events under way. Likewise the optimistic scenario depends on actions - replacing legacy systems and widespread skills - that are largely still to be done. How the next two years go is genuinely open.

Related NCSC pages may be updated as the picture changes. We will revise this page if the NCSC publishes material changes to its advice, rather than adding new pages for the same subject.

When to bring in specialist help

The first two steps above are often easier said than done, especially if nobody is sure what is exposed or who installed it. If you cannot produce a reliable list of internet-facing systems, if key equipment is out of support and cannot easily be replaced, or if you suspect an internet-facing system has already been compromised, a specialist can help scope the work. Do not delay the update on the strength of a suspicion, but do ask for a check for signs of intrusion alongside it.