If your staff use AI chatbots for work, the account they sign in with matters more than the tool. Banning them rarely works. A short policy, an approved tool and a few settings protect your data far better.
Generative AI is genuinely useful for drafting emails, summarising documents and tidying up spreadsheets. That's exactly why staff reach for it, often without asking. Microsoft calls this shadow AI: people using AI tools without the knowledge, approval or governance of their organisation, and it lists data leaks, regulatory non-compliance and reputational harm as the risks.
For a small business, the risk rarely looks dramatic. It looks like this:
In each case the business has no record of what was shared, no agreement with the AI provider covering that data, and no way to retrieve or delete it later. Your client trusted you with their information, not a chatbot your employee signed up for on their personal email.
The tool matters less than the account behind it. The same AI service can come with very different protections depending on whether someone signs in personally or through your organisation.
| Personal or free AI account | Copilot Chat with a Microsoft 365 work account | |
|---|---|---|
| Whose terms apply | The provider's consumer terms, agreed by the individual | Microsoft's Data Protection Addendum and Product Terms, with Microsoft as data processor |
| Used to train foundation models | Depends on the provider and the user's settings | No |
| Business can audit what was shared | No | Yes: prompts and responses are logged for audit and eDiscovery |
| Your access controls and labels apply | No | Yes: Copilot respects your permissions and sensitivity labels |
| When the employee leaves | Their chat history leaves with them | It stays within your Microsoft 365 tenant |
| Extra cost | Free or the employee's own subscription | No additional cost for users signed in with a work account |
Summarised from Microsoft's enterprise data protection documentation at time of writing. Exact controls vary by subscription plan.
Microsoft shows when enterprise data protection is active with a green shield at the top of Copilot Chat. That's a simple thing to teach: no green shield, no client data.
A policy works better when the right tool is also the easiest one. If you're on Microsoft 365, these steps make that happen:
If you're weighing up the paid Copilot licence as well, our Microsoft Copilot explainer covers what it adds over Copilot Chat, and our guide to getting your team ready for Copilot covers adoption.
Most small businesses don't need a 20-page AI governance framework. They need one page that everyone reads. Here's a starting point you can adapt:
1. Approved tools. For work, use only [approved tool, e.g. Copilot Chat] signed in with your work account. Don't use personal AI accounts for company work.
2. Never enter. Client or employee personal data (names with NRIC, contact details, salaries, health information), passwords or keys, unreleased financials, or anything marked Confidential, unless the tool is approved for that data.
3. Check the output. AI can be confidently wrong. You're responsible for anything you send, publish or decide, whether or not AI helped.
4. Be open about it. Tell your manager if AI drafted something important, such as client advice or a contract clause.
5. Ask first. Want to try a new AI tool or plug-in? Ask [IT contact] before signing up or connecting it to company data.
6. Report mistakes. If you've put sensitive data into the wrong tool, tell [IT contact] the same day. We'd rather know early.
Review the approved tool list every six months; AI products change quickly. And pair the policy with a short session showing people what good and bad use looks like. Our security awareness training guide covers how to fit that into the working week.
Under the PDPA, your organisation remains responsible for personal data in its possession or under its control, and its Protection Obligation requires reasonable security arrangements to prevent unauthorised disclosure. An employee pasting client records into a personal AI account is a disclosure your business can't track or undo. An approved tool, a clear rule about personal data and a record of who uses what are the kind of reasonable arrangements our PDPA IT checklist recommends.
AI tools are moving faster than most IT policies. The businesses that handle this well aren't the ones that ban AI; they're the ones that pick a safe default, write down three or four rules, and make sure everyone knows them. If you already pay for Microsoft 365, the safer tool is likely already in your subscription.
At Cloudeli, our AI and automation team helps SMEs set up Copilot Chat properly, decide which data can go where, and move on to automation that actually saves time. Our Cloudeli Complete package (US$36.60 per user per month on an annual plan) also includes security awareness training for your team.
This article is general guidance, not legal advice.
It depends on what they type. Anything entered into a personal or free AI account is handled under that service's consumer terms, not an agreement with your business, so you can't control retention or audit it. Keep client personal data, confidential documents and credentials out of personal AI accounts.
When staff sign in to Copilot Chat with their Microsoft 365 work account, Microsoft applies enterprise data protection: prompts and responses are covered by Microsoft's Data Protection Addendum and Product Terms, aren't used to train foundation models, and are logged for audit and eDiscovery. A green shield in the interface shows it's on.
Copilot Chat with enterprise data protection is available at no additional cost to users signed in with a Microsoft Entra work account. The paid Microsoft Copilot licence adds deeper access to your organisation's emails, files and chats.
A blanket ban is hard to enforce and tends to push usage onto personal phones where you have no visibility. A better approach is to approve a work tool with proper data protection, set clear rules on what data can go into any AI tool, and block or restrict the rest where you can.
At minimum: which AI tools are approved, which types of data must never be entered, the rule that people check AI output before using it, who is accountable for the final work, and how to report a mistake. One page is enough for most SMEs.
Book a free consultation. We'll check which AI tools your team can reach today, set up Copilot Chat with enterprise data protection, and help you put a short AI policy in place.
