1. Your prompt is information you are sharing

When you type into an online AI service or upload a file, you’re providing that information to a service provider. Exactly how the information is processed, retained or used depends on that product, plan and settings. Read the provider’s privacy information rather than assuming every service works the same way.

Before sharing a prompt, ask: Would I be comfortable sending this to an external organisation? If not, leave it out, anonymise it or use an approved alternative.

2. What should you avoid uploading?

• Passwords, access codes, bank details or identity documents.
• Customer lists, employee details, payroll records, business plans or confidential contracts unless your organisation specifically permits an appropriate service and workflow.
• Other people’s health, family, financial or contact details without a valid reason and proper safeguards.
• Private messages, photographs or copyrighted material where you lack permission to share them.

At work, approved tools, data agreements and company policy matter. A manager’s good intentions are not a substitute for permission. In the UK, personal information must be handled in line with applicable data-protection law; the ICO has specific AI guidance.

3. Make a safe example instead

Suppose you want help replying to a customer complaint. Rather than pasting the customer’s name, address, order number and private history, describe a fictionalised scenario:

This is useful and avoids unnecessarily sharing someone else’s data. Remember that removing a name alone may not fully anonymise a record if the remaining details identify the person.

4. Watch for common risks

Scams and impersonation: AI can help create convincing fake messages, voices or pictures. Verify surprising requests through an independent contact method. Bias: an answer can reflect stereotypes or leave out perspectives; inspect it carefully, especially when it affects people. Copyright: don’t assume an AI-created answer gives you ownership of every source it may have drawn from, or permission to publish third-party material.

5. Keep a human in the loop

Let AI propose, summarise and draft. Let an authorised person review decisions affecting money, employment, safety or rights. Never let the convenience of a plausible answer replace your responsibility.

Try it yourself

Turn this unsafe request into a safe one: “Here are my colleague’s full medical notes, home address and contact details. Write a performance review.”

Write your own replacement prompt. Check it against these three questions: Have I removed unnecessary personal details? Do I have permission to use the tool? Will a person review the result?

Can you explain it back?

Answer each question before opening the explanation. You aren’t being graded — this is how you make the knowledge stick.