This website uses cookies

Read our Privacy policy and Terms of use for more information.

Imagine a member of staff asking AI to tidy a customer email. They paste the whole conversation, including contact details and order information.

The task sounds routine. Sharing the information needs a separate decision.

This is a made-up example, but the question is practical: which tool, for which task, using which information, and who checks the result?

By the end of this issue, you will have a rule staff can follow, three examples to make it clear and a one-page Rule Sheet to define your first approved task.

1. Approve the task, not just the tool

Buying an AI tool does not approve every use of it. The business needs to decide which account, task and information are permitted.

The UK government's AI Management Essentials consultation response, updated in February 2026, sets out plans for more proportionate foundational guidance aimed at small and medium-sized enterprises (SMEs). It is not a certification or a legal requirement. Source: government response.

ILLUSTRATIVE DEMONSTRATION

A team is allowed to use its approved business account to rewrite public website text. A colleague wants to upload a confidential supplier contract.

The first approval does not cover the second task. Paying for the account does not settle whether that contract may be uploaded.

Blanket approval, “We pay for it, so use it”, leaves that boundary unclear. A blanket ban also gives staff no route for approved low-risk work.

2. Check the information before pasting

The National Cyber Security Centre (NCSC) advises against entering sensitive information into public large language models, the technology behind many AI assistants. It also says the service's terms and privacy arrangements need to be understood. Source: NCSC guidance.

This does not mean personal or confidential information can never be used with AI. Properly controlled business systems may support approved processing. That decision needs organisational approval and appropriate data-protection and security assessment, rather than an employee improvising.

The Information Commissioner's Office (ICO) explains the data-protection responsibilities involved in using AI. Its guidance is currently under review following the Data (Use and Access) Act. Source: ICO guidance.

ILLUSTRATIVE DEMONSTRATION

Task: “Help me write a clearer reply to this customer.”

Pause over: the real customer's email, address and order details.

For practising an approved task: use a completely made-up enquiry, with no real customer record.

Removing a name alone does not make a document anonymous. Other details may still identify someone.

Do not use identifiable customer, employee or child information, health, safeguarding, payroll or banking details in an informal experiment with a general-purpose external AI tool. Start with public, made-up or effectively anonymised information, or explicitly approved aggregates.

3. Name the human who checks

Public input does not make AI output automatically accurate. Staff need a named reviewer and a clear person to ask when approval is uncertain.

ILLUSTRATIVE DEMONSTRATION

Made-up public website text: “Our office is open Monday to Friday, 9am to 5pm. Contact us to arrange a visit.”

Prompt: “Using only this text, draft two customer questions and answers. Do not add facts. Flag anything you are unsure about.”

If the draft adds Saturday opening or promises an immediate appointment, the reviewer removes those unsupported claims before it is used.

For the first week, test one approved low-risk task. Record minutes saved after checking and corrections. If rework outweighs the saving, stop and review the task.

The takeaway

Give staff the rule from the SME AI Staff Safety Rule Sheet:

❝

Use only an approved business AI account for an approved task. Start with public, made-up or effectively anonymised information, or aggregates the business has approved. If the information, tool or task is not clearly approved, stop and ask. Keep a human responsible for the result.

This is a starter rule, not a complete AI policy or permission to process personal data.

One action

Download the SME AI Staff Safety Rule Sheet and use it to define your first approved AI task.

Download the Rule Sheet:

SME_AI_Staff_Safety_Rule_Sheet_v1.3.pdf

SME AI Staff Safety Rule Sheet

48.72 KB • PDF File

Barbara
AI Made Human

P.S. Keep the Rule Sheet where staff can find it before they open an AI tool.