Defence plan
How to defend against AI powered cyber attacks
There is no single fix. The NCSC points to fundamental security practices and careful design, so work through people, systems and AI tools in that order.
Last reviewed against NCSC (UK), Anthropic and UK Government (DSIT and Home Office). Sources
Principle
Basics first, then AI specifics
The NCSC says fundamental cyber security practices in the integration and configuration of AI and connected systems will be key [1]. Its senior operations lead asked organisations to ensure up-to-date defences are in place [3]. Nothing here requires new products. It requires finishing the ordinary work.
The order
Where to put effort
-
1
Add a second-channel check to payments, bank detail changes and access resets. -
2
Make it easy to report a suspicious message, with no blame for raising a doubt. -
3
Shorten the time between a security fix being available and being applied. -
4
List every AI tool in use and cut its reach to what it needs. -
5
Log AI inputs, outputs and tool calls, and review failures. -
6
Compare your AI projects with the government's published AI security code. -
7
Give a named director accountability and revisit it each quarter.
Judgement
Where leaders go wrong
Treating awareness as the whole answer
The Arup employee was won over by what appeared on screen: apparent colleagues who spoke and moved normally [8]. A rule that says "confirm through a number you already hold" works whether or not anyone spots the fake, so process has to carry the weight.
Buying a product to stop prompt injection
The NCSC advises caution about any product claiming it can stop prompt injection [2]. Reduce what the tool can do instead.
Waiting for a perfect inventory
Patching speed matters because the NCSC expects the gap between disclosure and exploitation to shrink further [1]. Start with the systems exposed to the internet.
If you do one thing this week
Choose the riskiest payment or access process and write down the second check. Test it by asking someone to try to get around it.
Related reading: prompt injection defence and using AI tools safely at work.
What to do
Seven steps, in order
-
Step 1
Add a second check to payments and access resets[8]
Require confirmation through a separate channel you already hold before bank details change, money moves or access is reset. Apply it to video calls and voice notes as well as email.
Who: Finance, HR and service desk teams
-
Step 2
Make reporting a suspicious message easy[6]
Give staff one obvious route to report doubts and make clear nobody is blamed for asking. Phishing was the most disruptive attack type for businesses that had a breach or attack.
Who: Security leads and managers
-
Step 3
Keep patching and configuration basics current[1]
The NCSC says fundamental practices in integrating and configuring AI and connected systems will be key. Keep AI systems and their dependencies up to date, and shorten the time you take to apply security fixes.
Who: IT and operations teams
-
Step 4
Reduce what an AI tool can do when it reads outside content[2]
Lower the privileges of an LLM to match the least trusted content it processes. An LLM reading external email should not have access to privileged tools.
Who: Developers and system owners
-
Step 5
Log AI tool inputs, outputs and tool calls[2]
Record what goes into and comes out of each LLM and which tools and APIs it calls. Failed calls can show someone probing the system.
Who: Security operations
-
Step 6
Compare your AI projects with the government's AI security code[10]
The government's published code on AI cyber security offers baseline principles for those who develop and deploy AI systems. Check each AI project you run or build against it.
Who: Security and product leads
-
Step 7
Put AI risk on the board agenda[4]
The NCSC suggests leaders ask about accountability, supply chain security, incident response and skills gaps for every AI system. Record who owns each answer.
Who: Directors and executives
Keep reading
Read next
Timeline
A dated, sourced timeline of AI related cyber attack cases and UK guidance, from the Arup deepfake fraud in 2024 to the 2026 breaches survey.
Read the guide
Questions
Sourced answers on AI powered cyber attacks: what they are, how deepfake fraud works, prompt injection, UK figures and what organisations should do.
Read the guide
AI phishing
How AI strengthens phishing and impersonation, what the UK survey shows about phishing, and the process checks that still work.
Read the guideHow CyPro can help
Need a hand with any of this?
Patching is one job. Keeping a vulnerable system safe until it is patched, and proving it, is several more. CyPro's consultants, testers and SOC analysts can take any of them off your plate.
Vulnerability scanning
Find every internet-facing system you run, including the forgotten ones, and see which are still on a vulnerable build.
See the scanning service
Hardening and segmentation
Our consultants put the controls in place that limit the damage when something is missed: firewall rules, network isolation and tighter admin access.
More on cypro.co.uk
24/7 security monitoring
Our SOC analysts watch your systems day and night for the signs of an attack in progress: unexpected processes, new outbound connections, logins that do not fit.
More on cypro.co.uk
Penetration testing
A focused test of your systems and the controls around them, so you know they hold before an attacker finds out for you.
More on cypro.co.uk
Incident response
If you may already be compromised, our responders contain it, preserve the evidence, find out what was touched and get you running again safely.
More on cypro.co.uk
Fractional CISO for the board
A senior security lead who advises the board on what to fix first, what to accept and what to tell customers, and owns the decision record.
More on cypro.co.uk
Know your exposure
Not sure you are protected?
Many of the risks here sit in how an organisation uses AI: what tools staff run, what they can reach and who owns the risk. AI governance work sets that out so controls match the tools in use.