Risks
The AI security risks that matter, and what the evidence says
The main risks are manipulated tools, exposed data, tampered training data, over-trusted output and tools with too much access. The NCSC or OWASP documents each.
Last reviewed against NCSC, Department for Science, Innovation and Technology and OWASP GenAI Security Project. Sources
The core risks
Five to understand
Prompt injection
An attacker crafts input that steers a model away from what its owner intended. In the NCSC's words, prompt injection attacks are among the weaknesses reported most often in LLMs [2]. OWASP puts it first in its 2025 list [8]. See the prompt injection controls.
Sensitive information disclosure
OWASP lists it second, as LLM02 [8]. A model, or the tools around it, can reveal confidential data in what it outputs.
Data poisoning
Unreliable output
The wider picture
What the threat assessment says
The NCSC's assessment looking to 2027 sets out six judgements. Two stand out for a UK organisation. The first is that intrusion work will almost certainly get more efficient in places as attackers adopt AI. The second is that a divide will almost certainly open between systems that stay abreast of these threats and a large share that do not [4]. It also judges that expanding AI use, particularly in critical national infrastructure, almost certainly enlarges the attack surface [4].
Read the judgements as planning input
Where defences lag or stand still, the assessment allows that critical systems may become easier for advanced actors to compromise by 2027 [4].
Readiness
How prepared organisations are
The government survey puts a figure on the gap. Of businesses using or adopting AI, about a quarter (24%) had practices for managing AI-related cyber risk, against 27% of the charities in the same group [10]. Across all businesses, 43% reported a breach or attack in the 12 months before the survey [10].
The survey counts only incidents that organisations identified and chose to report, so the true rate is likely higher [10].
Next, turn the risks into action with the practical plan.
Key facts
AI security controls in brief
- Top LLM risk
- Prompt injection, LLM01 in OWASP's 2025 list[8]
- NCSC view
- Prompt injection is a residual risk to manage, not a flaw to patch[1]
- UK government code
- Voluntary, 13 principles, published 31 January 2025[6]
- Businesses using, adopting or considering AI
- 31%[10]
- Of those using or adopting AI, with AI security practices
- 24% of businesses[10]
- Agentic AI
- NCSC: know what access agentic tools have to systems and data[3]
Keep reading
Read next
Plan
A practical AI security plan for UK organisations: inventory, least privilege, human approval, logging and ownership, drawn from NCSC advice.
Read the guide
Controls
AI security controls UK organisations can apply now: inventory, least privilege, approval, logging and ownership, from NCSC and OWASP guidance.
Read the guide
Code of practice
The UK government's voluntary AI cyber security code of practice: who it covers, the 13 principles and how a UK organisation can use it as a checklist.
Read the guideHow CyPro can help
Need a hand planning for this?
Understanding the change is one job. Working out what it means for your systems, your suppliers and your roadmap is several more. CyPro's consultants can take any of them on.
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
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
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
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
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
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
Plan ahead
Want a plan for your organisation?
AI security starts with knowing which tools are in use, who owns them and what they can reach. CyPro can help set the governance, accountability and controls around that.