NCSC Annual Review 2026: what it says about AI and cyber threat

The short answer. The National Cyber Security Centre's 2026 Annual Review says AI is almost certainly acting as a force multiplier for attackers. It speeds up the things they already do, like reconnaissance, phishing, finding vulnerabilities and writing exploits, rather than inventing new kinds of attack. The NCSC also judges that the gap between well-defended and poorly defended organisations will almost certainly widen, and that staff using unapproved AI tools ("shadow AI") is highly likely to increase exposure to attackers. For a UK SME, the practical message is to patch faster, know which AI tools your people are using, and stop being the easiest target in your sector.

The assessment, The impact of AI on cyber threat, was published on 3 November 2026 and covers 1 September 2025 to 31 August 2026. It's written for security professionals and the public sector, but most of it applies directly to smaller businesses. This article goes through the seven key judgements and then sets out what to do about them.

How to read the NCSC's language

The NCSC writes its judgements using the UK intelligence community's probability yardstick, so the wording is deliberate. "Almost certain" is the strongest level of confidence it uses, at 95% or above. "Highly likely" sits at roughly 80 to 90%. When the NCSC says something is almost certain, it isn't hedging.

The seven key judgements

1. AI is a force multiplier, not a new weapon (almost certain)

AI is improving several stages of an attack: reconnaissance, social engineering, vulnerability discovery and exploit coding. Its main effect is to speed up and scale existing tactics, techniques and procedures, not to create new attack routes. The NCSC also notes that adoption among threat actors is uneven. Some groups use AI heavily and others barely at all.

What it means for you. The attacks you'll face look like the ones you already know about: phishing emails, fake invoices, password attacks and unpatched systems. There will just be more of them, they'll be better written and they'll arrive faster. The defences that worked before still work. The cost of skipping them goes up.

2. Humans are still in charge of the attacks (highly likely)

The NCSC describes what it sees as human-machine teaming rather than autonomous attacks. Human operators are still needed to direct AI, check what it produces and adapt it. Skilled attackers get the most out of it: the better the operator, the bigger the gains. The NCSC also points out that losing control of an agentic AI tool mid-operation would likely get in the way of an attacker's goals, which is one reason they keep humans in the loop.

What it means for you. Fully automated "AI hacking" isn't the main risk right now. The risk is capable criminals who can now do in an afternoon what used to take a week. That same lesson applies to your own AI use: the NCSC's point about agentic AI needing human oversight is just as true for the AI agents you deploy in your business.

3. AI is finding vulnerabilities faster (highly likely)

AI can find flaws in large codebases and spot misconfigurations that have gone unnoticed. Not every flaw it finds can be exploited straight away, but the NCSC says defenders face a race to shorten patch deployment timelines before attackers build working exploits.

What it means for you. The time between a vulnerability being published and it being used against you is shrinking. If your patching depends on someone remembering to do it at the end of the month, that's now too slow. Automatic updates, a clear list of what you run and a named person who checks it are the basics.

4. The digital divide is getting wider (almost certain)

This is the judgement that matters most for SMEs. Organisations with mature security programmes, resources and robust AI governance are better placed to detect threats, prioritise vulnerabilities and automate their defences. Organisations with legacy technology and limited resources may struggle to keep up, and the NCSC judges they are likely to become more attractive targets for both state and criminal actors.

What it means for you. Attackers go where the effort is lowest. If AI lets them scan and approach thousands of businesses at once, the ones that get breached are the ones with old systems, no multi-factor authentication and nobody watching. You don't have to match a bank's security budget. You do have to stop being the softest target in your supply chain.

5. China is best placed to use AI against the UK (highly likely)

The NCSC judges that China's investment in AI, and its ambition to be a leading science and technology superpower, make it well placed to benefit from AI-enhanced cyber operations. Its frontier AI programmes would likely improve vulnerability identification, targeting and access operations against critical national infrastructure (CNI) and other strategic targets.

What it means for you. Most SMEs aren't the direct target of state actors. But if you supply energy, water, transport, telecoms, healthcare, defence or government, you're part of the route in. Expect clients in those sectors to ask harder questions about your security, and your AI use, in tenders and supplier reviews.

6. Shadow AI is creating blind spots (highly likely)

When staff move sensitive or proprietary data onto personal devices to use AI tools that haven't been approved, the organisation loses sight of how that data is handled and who can access it. The NCSC adds a second, less obvious point: when work is handed over to shadow AI, the business also loses sight of how capable its own people actually are at security. The NCSC judges that both effects would likely increase exposure to advanced persistent threat groups and cyber criminals.

What it means for you. This one is almost entirely within your control. Banning AI doesn't work. Staff just use it on their phones instead. What works is giving people approved tools that are good enough to use, a short policy that says what data can and can't go into them, and a register of what's in use. It's also the first step of any ISO 42001 preparation.

7. AI-enabled ransomware against UK infrastructure (highly likely)

Criminal use of AI is likely to increase the scale and efficiency of ransomware, social engineering and vulnerability exploitation. The NCSC expects open-source AI tools built for security researchers to be repurposed to automate criminal workflows, which lowers the cost of running an attack. Attack volumes against less resilient targets are expected to rise.

What it means for you. Cheaper attacks mean more attacks, and the economics favour going after smaller businesses in bulk. Ransomware recovery depends on backups that are offline or immutable, tested, and not reachable with the same admin password as everything else.

What a UK SME should do now

None of the NCSC's judgements call for exotic new defences. They call for the basics done well and done quickly, plus control over your own AI use.

1. Get the basics certified. Cyber Essentials, the NCSC-backed scheme, covers patching, access control, firewalls, malware protection and secure configuration. Those are exactly the controls that AI-scaled attacks test first. If you supply the public sector or CNI, aim for Cyber Essentials Plus.
2. Shorten your patch cycle. Turn on automatic updates wherever you can. For anything that can't update itself, set a target time to patch critical fixes and track it.
3. Turn on multi-factor authentication everywhere. Email, finance systems, remote access and admin accounts first. Better-written phishing is far less useful against an account that needs a second factor.
4. Find out what AI your staff are using. Build a simple register of every AI tool and feature in use, including AI built into software you already pay for, and what data each one sees.
5. Write a short AI use policy. One or two pages: which tools are approved, what data must never go in, and who approves new tools.
6. Keep humans in charge of your own AI. If you run AI agents that can send emails, change records or make payments, make sure a person reviews the actions that matter and can switch the agent off.
7. Test your backups. A backup you've never restored from is a hope, not a plan.

If you want to see how your business stands, the AI readiness scorecard includes questions on governance and security. Our AI readiness assessment maps the AI already in use across your workflows and the data it touches, which is the visibility the NCSC says organisations are losing to shadow AI. For how this fits with formal standards, see ISO 42001 vs ISO 27001, GDPR and Cyber Essentials Plus.

Frequently asked

Does the NCSC say AI is making attacks more dangerous?

It says AI is making existing attacks faster, cheaper and more widespread, not that it has created new kinds of attack. That's still a serious change. More volume and better-crafted phishing means more businesses will be hit, especially those with weak basics.

Is my small business really a target?

Probably not individually, which is the problem. AI lets criminals target thousands of businesses at once and pick off whichever ones are easiest. The NCSC judges that organisations with weaker security and legacy systems are likely to become more attractive targets.

What is shadow AI?

It's staff using AI tools the business hasn't approved, often free accounts on personal devices. The risk is that company or client data ends up in places you can't see or control. The fix is approved tools, a clear policy and a register of what's in use, not a ban.

Should we stop using AI because of this?

No. The NCSC's point is that organisations with good AI governance are better placed, not worse. Using AI with clear ownership, approved tools and human oversight is part of how you stay on the right side of the divide.

Is Cyber Essentials enough?

It's the right starting point and it covers the controls AI-scaled attacks hit first. It doesn't cover how your business governs AI. For that, a simple AI policy and register is the first step, and ISO 42001 is the formal standard if clients ask for it.