Your staff are probably already using AI tools, whether or not you have chosen any. The question for a small business is how to use them without leaking information or opening a new way in for attackers. The good news is that most of the answer is ordinary security applied to a new kind of tool.
Key takeaways
- Treat AI tools like any other business software: business accounts, multi-factor sign-in, an owner.
- The biggest everyday risk is what staff paste in. Agree simple rules and write them down.
- Check each supplier's terms for how your data is stored and used. Do not assume.
- Tools that can act on your behalf need tighter limits than tools that only write text.
- Cyber Essentials covers the basics that protect AI tools too.
Start with accounts and settings
Most AI security problems in small firms start with accounts nobody manages. A member of staff signs up with a personal email on a free plan, uses it for client work, then leaves.
Simple fixes:
- Use business accounts, not personal ones. Many AI products have plans for teams or businesses with an admin console. Compare what each plan offers on security and data handling before you choose; the free tier is often different.
- Turn on multi-factor sign-in (a code or prompt on your phone as well as a password) for every AI account.
- Give someone ownership. One named person keeps a list of which AI tools the business uses, who has access and who pays.
- Review the admin settings. Look at who can share chats, install add-ons and connect other systems.
- Remove access when people leave, the same day you close their email.
Be careful with connectors
Many AI tools can now connect to your email, calendar, cloud storage or customer system. That is useful, and it means the tool can see everything those systems hold. Connect only what a task needs, and check who in your team can add new connections.
Stop data leaking through what staff paste in
Every time someone pastes a client email, a contract or a spreadsheet into an AI tool, that information goes to the supplier. Whether it is stored, for how long and whether it is used to improve the supplier's models depends on the product, the plan and the settings.
Do not take anyone's word for it, including ours. Read the supplier's terms and privacy information for the plan you are on, and check the settings that control training and retention. If anything is unclear, ask the supplier in writing.
Then agree some house rules. A one-page note is enough:
- Which AI tools staff may use for work, and which they may not
- What must never be pasted in (for example passwords, client files, bank details)
- Who to ask when unsure
- That a person checks AI output before it goes to a customer
If you want to use AI with customer or staff records, speak to a data protection adviser first. That is a question about your obligations, and it needs someone qualified to answer it.
Prompt injection in brief
Prompt injection is when text an AI tool reads, such as an email or a web page, contains hidden instructions that the tool then follows. The UK's National Cyber Security Centre (NCSC) says it is a risk that "cannot be fully mitigated with a product or appliance".1 The practical defence is limiting what the tool can reach and do. Our prompt injection explainer has examples and questions to ask.
Go carefully with AI agents
An AI agent is a tool that takes actions for you: sending emails, booking meetings, updating records, even paying bills. The risk rises with every permission it has. The NCSC's advice on prompt injection is to assume the AI can be manipulated and to restrict what it can do with controls that are not themselves AI, such as permissions and approval steps.1 It has also published guidance titled "Thinking carefully before adopting agentic AI".2
For a small business that means:
- start with low-risk tasks where a mistake is easy to undo
- keep a person approving anything involving money, customers or deletions
- make sure you can see what the agent did, and switch it off quickly
If you cannot tell what an agent is doing, it is not ready to run on its own.
Cyber Essentials covers the basics
Cyber Essentials is the government-backed baseline security scheme run with the NCSC. It covers five technical controls: firewalls, secure configuration, security update management, user access control and malware protection.3 The current requirements, version 3.3, took effect on 27 April 2026.4
None of those is specific to AI, and all of them protect your AI tools: user access control is exactly the "who has an account" question above. The NCSC says small and medium-sized businesses preparing for Cyber Essentials can book a free 30-minute consultation with an NCSC-assured Cyber Advisor.3 It also says UK organisations with turnover under £20 million that certify their whole organisation are entitled to cyber liability insurance; check the current terms on the NCSC site.3
If you build or commission AI systems, the government's AI Cyber Security Code of Practice, published 31 January 2025, sets out baseline security principles, with an implementation guide.5
When to get professional help
You can do most of the above yourself. Get help when:
- you want an AI tool to connect to your core systems or act on its own
- you are handling sensitive client information under contract
- a supplier's answers about security do not make sense to you
- something has already gone wrong
Our list of security questions to ask an AI supplier helps you test anyone you are thinking of hiring, and you can browse consultants in our directory.
Sources
-
National Cyber Security Centre, "Prompt injection is not SQL injection (it may be worse)", published 8 December 2025, https://www.ncsc.gov.uk/blog-post/prompt-injection-is-not-sql-injection, accessed 2026-11-14. ↩ ↩2
-
National Cyber Security Centre, "Thinking carefully before adopting agentic AI", undated PDF, https://www.ncsc.gov.uk/sites/default/files/2026-06/Thinking-carefully-before-adopting-agentic-AI.pdf, accessed 2026-11-14. ↩
-
National Cyber Security Centre, "Cyber Essentials: overview", undated, https://www.ncsc.gov.uk/cyberessentials/overview, accessed 2026-11-14. ↩ ↩2 ↩3
-
National Cyber Security Centre, "Cyber Essentials: help and resources", undated, https://www.ncsc.gov.uk/cyberessentials/resources, accessed 2026-11-14. ↩
-
Department for Science, Innovation and Technology, "AI Cyber Security Code of Practice", published 31 January 2025, https://www.gov.uk/government/publications/ai-cyber-security-code-of-practice, accessed 2026-11-14. ↩