THE NCSC SAYS AI IS ABOUT TO FLOOD THE INTERNET WITH VULNERABILITIES. HERE IS WHAT UK BUSINESSES NEED TO DO.

WHAT THE NCSC ACTUALLY SAID

In May 2026, the NCSC published two closely linked pieces of guidance that together paint a clear picture of how AI is changing the UK's threat landscape.

The first was a blog post titled Preparing for a 'Vulnerability Patch Wave', warning that AI tools are finding software vulnerabilities significantly faster than the security industry - and most organisations - can respond. The pace of CVE disclosures is accelerating. The window between a vulnerability being publicly known and it being actively exploited in the wild is getting shorter. And the volume of vulnerabilities requiring urgent attention at any given time is increasing.

The second was a 10-question checklist for organisations considering using AI tools to discover vulnerabilities in their own systems. Alongside it came a companion warning about agentic AI - AI systems that can take autonomous actions across multiple systems - and the specific risks of granting those systems excessive permissions.

Taken together, the NCSC's message is clear: AI is raising the floor on what attackers can do. And the organisations most at risk are those that have not yet got the basics right.

WHAT IS THE AI VULNERABILITY PATCH WAVE?

The "patch wave" refers to the growing volume of disclosed vulnerabilities that organisations are expected to assess, prioritise, and remediate - and the shrinking time they have to do it in.

Historically, finding a new software vulnerability required deep technical knowledge, significant time investment, and specialist tooling. AI has changed this. Large language models, code-analysis models, and automated fuzzing tools can now:

• Scan large codebases and identify logic flaws in hours rather than weeks
• Generate working proof-of-concept exploit code for known vulnerability classes
• Chain together multiple low-severity weaknesses into higher-impact attack paths
• Scan the internet for exposed vulnerable services at scale

This does not only mean defenders can find vulnerabilities faster. It means attackers can too - and increasingly at low cost, with low technical skill requirements. A threat actor who could previously only exploit a well-publicised, pre-packaged vulnerability can now use AI-assisted tooling to adapt exploits to specific targets.

The result is that the 14-day patching window - already a requirement under Cyber Essentials - is becoming the minimum viable response time, not a conservative safety margin.

WHAT THE NCSC'S 10-QUESTION CHECKLIST COVERS

The NCSC's checklist is aimed at organisations considering adopting AI tools for their own security purposes - to find vulnerabilities in their own systems before attackers do. The 10 questions cover:

• Data handling: What data does the AI tool access, retain, or transmit? Where is it processed?
• Model provenance: Who built the model, and has it been independently assessed?
• Access permissions: What systems and credentials does the tool require to function?
• Vulnerability disclosure: What happens when the tool finds something? Who is notified, and how?
• Output accuracy: What is the false positive rate? How are findings validated before action is taken?
• Scope controls: Can the tool be constrained to authorised targets only?

The underlying concern is straightforward: an AI security tool with broad network access, cloud credentials, and the ability to execute code is itself a high-value target. If it is compromised, misconfigured, or simply behaving unexpectedly, it can cause significant damage quickly - potentially more quickly than a human operator can intervene.

WHY THE FIVE CYBER ESSENTIALS CONTROLS ARE THE RIGHT STARTING POINT

The NCSC's guidance is direct: get the basics right before investing in AI-powered security tools. And the five Cyber Essentials controls map almost exactly onto the defences that limit AI-assisted attacks.

1. Patch Management

If AI is finding vulnerabilities faster and attackers are weaponising them more quickly, the single most important defensive action is timely patching. Cyber Essentials requires critical and high-severity patches to be applied within 14 days of release. Under the new Danzell question set, failing to meet this requirement on either operating systems or applications is now an automatic certification failure - because the NCSC and IASME recognise how critical it is.

For SMBs, the practical implication is that "we patch when we get round to it" is no longer an acceptable approach. Patch management needs to be a scheduled, evidenced process - not a reactive one.

2. Firewalls and Internet Gateways

AI-assisted scanning tools can identify exposed services across the internet rapidly. Boundary firewalls that limit what is exposed - blocking unnecessary inbound connections, restricting outbound traffic to known destinations - reduce the attack surface that AI-powered reconnaissance can find and exploit.

3. Secure Configuration

Default credentials, unnecessary open ports, unused services, and misconfigured cloud permissions are among the most common vulnerabilities that AI scanning tools are used to find and exploit. Secure configuration - changing defaults, removing what you do not need, and hardening what you do - directly addresses this.

4. User Access Control

Even if a vulnerability is found and exploited, limiting what any single compromised account can access restricts the damage an attacker can do. Least-privilege access, removing local administrator rights from standard users, and using separate accounts for privileged tasks all limit the blast radius of a successful attack.

5. Malware Protection

AI-generated exploit code and novel malware variants are increasingly hard to detect using signature-based tools alone. Behaviour-based malware protection, application allow-listing, and execution controls limit what malicious code can do even when it gets past initial defences.

IS THIS RELEVANT TO SMALL BUSINESSES?

Yes - and the NCSC's guidance makes no distinction by organisation size. AI-powered vulnerability scanning is available to attackers at increasingly low cost and with increasingly low technical skill requirements. Small businesses are not less interesting to attackers than large ones; they are often more interesting, precisely because they are less likely to have defensive controls in place.

The pattern the NCSC is warning about is not targeted attacks on specific high-value organisations. It is broad, automated scanning that finds exposed, unpatched services across the internet - and then exploits them. Any organisation running software, using cloud services, or connecting to the internet is in scope for this kind of attack. Which means every SMB.

The good news is that the response is not complicated or expensive. It is the five controls that Cyber Essentials has always covered. The NCSC's warning is essentially validation that those controls - patch management, firewalls, secure configuration, access control, and malware protection - remain the right foundation in an AI-accelerated threat environment.

WHAT ABOUT AI SECURITY TOOLS FOR SMBs?

The NCSC's 10-question checklist is aimed at organisations considering AI tools for their own security use - vulnerability scanning, threat detection, and similar. For most UK SMBs, the honest answer is that this is not where to start.

AI security tools work best when the basics are already in place. If your patching process is manual and unreliable, if your cloud services are not MFA-protected, and if your network has no meaningful boundary controls, adding an AI vulnerability scanner will find a long list of problems without giving you the organisational capability to fix them systematically. You will spend money on tools when the more valuable investment is in the controls and processes those tools are designed to supplement.

Get Cyber Essentials certified first. That is the NCSC-backed baseline. Once those controls are in place, you are in a much stronger position to evaluate whether additional tooling - AI-powered or otherwise - adds value for your specific environment.

THE PRACTICAL CHECKLIST FOR UK SMBs

In response to the NCSC's May 2026 guidance, here is what UK SMBs should focus on:

• Review your patching process - is it scheduled, documented, and evidenced? Can you demonstrate that critical updates are applied within 14 days? If not, this is the highest priority fix.
• Audit your cloud services - every SaaS tool that holds your data is a potential attack vector. Ensure MFA is enforced (not just enabled) on all of them.
• Check what is exposed to the internet - review your firewall rules. Are there services accessible from the internet that do not need to be? Open RDP ports, unprotected admin interfaces, and unnecessary cloud storage access are common findings.
• Review user access levels - who has administrator rights? Does everyone need them? Remove privileges that are not required for the role.
• Check your malware protection - is it active, up to date, and covering all your devices including personal devices used for work?
• Consider Cyber Essentials certification - this is not just a badge. It is a structured way to evidence that these controls are in place and are working, assessed against the UK Government's own security standard.

SUMMARY

The NCSC's May 2026 guidance is not a theoretical warning about a future threat. AI tools for vulnerability discovery are widely available now, and the pace at which vulnerabilities are being found and exploited is accelerating. The patch wave is already here.

The NCSC's response - and ours - is not to tell businesses to panic, or to invest immediately in AI security tooling. It is to get the basics right. Patch on time. Enforce MFA. Limit what is exposed to the internet. Control who has access to what. Protect against malware. These five things, done well and evidenced through Cyber Essentials certification, remain the most effective defence available to UK SMBs - whether the threat is AI-assisted or not.