Using AI and Microsoft Copilot Safely. A Practical Guide for Small Businesses
AI tools can save real time, but they also make it easier to expose data you never meant to share. Sort the basics first.
Your staff are probably already using AI tools, whether you have approved them or not. Drafting emails, summarising documents and tidying spreadsheets are all things these tools do well. The question for a business owner is not whether to allow AI, but how to make sure confidential information does not end up somewhere it should not.
The first issue is which tool is being used. A free, public chatbot generally works on terms that are different from those of a business product, and you should read the provider's current terms on how prompts are stored and whether they can be used to improve the service. As a rule, nobody should paste client personal data, passwords, contracts or financial records into a tool that the business has not assessed. A short written policy listing approved tools and what must never be entered covers most of the risk.
Microsoft 365 Copilot is a different case, because it works inside your own tenant and draws on the content your staff can already access. Microsoft states that Copilot respects existing permissions and does not use your tenant data to train its foundation models. That is reassuring, but it exposes a hidden problem: Copilot is very good at finding things. If a salary spreadsheet sits in a SharePoint site that everyone can technically open, Copilot may surface it to someone who should not see it. This is often called oversharing.
So before rolling out Copilot, tidy your permissions. Review who has access to SharePoint sites and Teams, remove the broad access groups such as everyone, and use sensitivity labels and Microsoft Purview tools to protect the most sensitive material. Start with a small pilot group, see what they can find, and fix what turns up.
UK GDPR still applies when you use AI. If personal data is involved, you need a lawful basis, you should be open with people about how their data is used, and for higher risk uses you should consider a data protection impact assessment. The ICO publishes guidance on AI and data protection that is worth reading. Also remember that AI can be confidently wrong, so a human should check anything that goes to a client or into a decision.
If you want to use AI properly, Dragonfly IT can help you check your Microsoft 365 permissions, write a simple policy and decide whether Copilot is worth the cost for your team.
