Blog · Shadow AI
Your staff are already using AI. Do you know which tools?
The NCSC says most UK employees have used AI tools their employer never approved. Why a ban won't work, and what UK firms should do about shadow AI instead.
Nearly three-quarters of employees say they have used AI tools their employer hasn't approved. The National Cyber Security Centre (NCSC) quoted that figure, 71%, in September when it warned about "shadow AI" [1].
Last week it went further. Its annual review judged that shadow AI will highly likely leave organisations open to attack by criminals and state-backed groups [2].
The risk is simple. Someone pastes a client contract, a payroll file or a board paper into a free chatbot to save half an hour. That information now sits outside the firm's control, and may be kept or used to train the service [1].
It's no longer only a tool someone signs up for. AI is built into the search engine, the email client and the phone. Microsoft has put Copilot into Outlook and across Microsoft 365, and mobile devices now lean on AI for all sorts of everyday tasks. Much of it arrives without anyone in the firm choosing it.
I've seen engineers and developers use AI that management knew nothing about, simply because nobody had ever thought to block it.
I've seen emails carrying hidden instructions in plain text. The recipient couldn't see them, but an AI assistant could read them and act on them without the user knowing.
I've seen an AI agent create new containers and firewall rules to get round the guard rails it was given.
That's the real shadow IT problem. You don't know what you don't know. And if you don't know what you're looking for, how are you going to find it?
Most firms need to start catching up.
The instinct is to ban it. The NCSC doesn't recommend that, and neither do I.
People turn to their own tools when the approved ones don't do the job. Ban them and the use goes quiet, not away. The NCSC's advice is to reduce the risk rather than assume you can remove it, and to find out why people are using these tools in the first place [1].
What to do this week
- Ask, don't hunt. Run a short, no-blame survey: which AI tools do people use for work, and for what? You'll learn more in a week than a ban would tell you in a year.
- Find out what's switched on. Check which AI features your Microsoft 365, Google and phone settings have turned on, and who can use them. Before buying anything new, see what your plans already include.
- Write one page of rules. What can go into an AI tool, what never can (client data, personal data, anything under a non-disclosure agreement), and who to ask.
- Treat what an assistant reads as untrusted. An email or document can hide instructions that the AI may follow. Limit what each tool can reach, and what it can do without a person agreeing.
- Keep a human sign-off on changes. No AI agent should create servers, containers or firewall rules on its own.
Read more
Sources
- [1]National Cyber Security Centre, The hidden risks of shadow AI, 7 September 2026. https://www.ncsc.gov.uk/blogs/the-hidden-risks-of-shadow-ai
- [2]National Cyber Security Centre, Annual Review 2026: the impact of AI on cyber threat, 3 November 2026. https://www.ncsc.gov.uk/collection/annual-review-2026/understanding-threat/ai-impact-cyber-threat