← LEARN

What Not to ShareWith AI.

AI can be useful for drafting, analysing and organising work, but the information you provide matters. Before entering workplace or personal data into an AI tool, decide whether that information should be shared at all.

01 / THE CORE PRINCIPLE

Do not share information simply because AI can process it.

A useful AI prompt does not need every detail you know. In many situations, the safest and most effective approach is to provide only the minimum information required for the task.

This is especially important when working with information that belongs to customers, employees, suppliers, employers or other people.

CORE PRINCIPLE

If a piece of information is not necessary for the AI task, leave it out.

02 / SENSITIVE INFORMATION

Some information deserves extra protection.

01

Passwords & authentication details

Do not paste passwords, one-time codes, API keys, recovery codes, private tokens or login credentials into an AI tool.

02

Financial information

Avoid sharing bank account numbers, card details, payment credentials, confidential pricing information or sensitive financial records unless you are using an approved system specifically designed for that data.

03

Personal identity information

Be careful with passport numbers, national identification numbers, employee IDs, private addresses, personal phone numbers and other data that could identify someone.

04

Confidential workplace information

Internal reports, customer records, contracts, pricing, unreleased plans, operational incidents and company documents may be confidential even when they do not look highly sensitive.

05

Customer & employee data

Names, contact details, performance records, payroll information, complaints, HR information and customer histories may be protected by company policy or privacy requirements.

06

Restricted technical information

Source code, security configurations, internal architecture, proprietary formulas, private credentials and vulnerability information should only be shared through approved systems and processes.

IMPORTANT

Something does not have to be secret to be sensitive. A normal workplace document may still contain personal, commercial or operational information that should not be entered into an unapproved AI system.

03 / BEFORE SHARING ANYTHING

Ask six questions before putting real information into AI.

01
OWNERSHIP

Do I have the right to share this information?

02
NECESSITY

Does AI actually need this detail to complete the task?

03
SENSITIVITY

Could this information harm someone or the organisation if exposed?

04
POLICY

Does my organisation allow this information to be used with this AI tool?

05
MINIMISATION

Can I remove names, numbers or confidential details first?

06
ALTERNATIVE

Can I describe the situation without sharing the original data?

SIMPLE DECISION RULE

If you would hesitate to send the information to an unknown external person, do not automatically paste it into an AI tool.

04 / SAFER INPUTS

Most AI tasks can be completed with less sensitive information.

Redaction and anonymisation can often preserve the useful context while removing information that AI does not need.

Replace real names
INSTEAD OF

Customer: Mohamed Ali

USE

Customer A

Remove account identifiers
INSTEAD OF

Account 4892-7721-1184

USE

Customer account

Generalise commercial details
INSTEAD OF

Supplier rate: QAR 7.85 per unit

USE

Current supplier rate

Use sample values
INSTEAD OF

Actual payroll figures

USE

Example figures for demonstration

Describe instead of uploading
INSTEAD OF

Full confidential incident report

USE

A summary of the relevant non-sensitive facts

Limit the scope
INSTEAD OF

Complete customer database

USE

Only the non-identifying fields required for the task

DATA MINIMISATION

Reduce the prompt to the smallest amount of real information necessary to achieve the result.

05 / THINK IN RISK LEVELS

Not every AI input carries the same level of risk.

LOWER RISK

Public or generic information

Public product descriptions, generic examples, common knowledge and fictional practice data are normally easier to use safely.

MEDIUM RISK

Internal operational information

Internal processes, non-public reports and workplace details may require approval, redaction or use of an organisation-approved AI system.

HIGHER RISK

Personal, confidential or security-sensitive data

Credentials, personal records, financial details, restricted business information and security data should not be entered into unapproved AI services.

Higher-risk information deserves stronger controls. This may mean not using AI at all, using only an organisation-approved AI environment, or removing sensitive details before processing.

06 / WORKPLACE EXAMPLE

You need AI to help write an inventory investigation summary.

SITUATIONStock discrepancy investigation

A warehouse team identifies a stock difference. The internal investigation file contains employee names, customer details, document numbers and other operational information.

POOR APPROACH

Paste the complete investigation file.

This may expose personal and internal operational information that AI does not need in order to help structure the report.

BETTER APPROACH

Extract only the necessary facts.

Provide the verified system quantity, physical quantity, variance, checks already performed and investigation status without including unnecessary identifying details.

MINIMISED INPUT
  • System quantity: 1,780 units
  • Verified physical quantity: 1,753 units
  • Variance: 27 units shortage
  • Receiving records reviewed
  • Outbound transactions reviewed
  • Adjustment history reviewed
  • Root cause not yet confirmed
UNSAFE

Write a complaint response to customer John Perera, passport number N1234567, phone +974 XXXXXXXX, regarding account 984277 and the payment dispute described in the attached internal customer file.

SAFER

Write a professional response to a customer regarding a disputed payment. The customer says the amount charged does not match the agreed amount. Acknowledge the concern, explain that the transaction is being reviewed, and ask the customer to wait for confirmation. Do not invent the investigation result.

07 / SAFE INPUT WORKFLOW

Build privacy checks into the workflow before AI is used.

01
PAUSE

Identify the information

Before pasting anything, recognise whether the material contains personal, confidential, restricted or commercially sensitive information.

02
REDUCE

Remove unnecessary details

Keep only the information AI genuinely needs to perform the task.

03
ANONYMIZE

Replace identifying information

Use labels, placeholders or fictional examples where the real identity is not necessary.

04
CHECK

Confirm the tool is approved

Follow your organisation's rules about which AI systems may be used for workplace information.

05
REVIEW

Inspect before submitting

Read the final prompt once more and remove anything that should not leave the approved environment.

WORKPLACE RULE

If your employer has an AI policy, data-handling policy or list of approved tools, follow that policy before using workplace information with AI.

08 / PRACTICAL TAKEAWAY

Useful prompts do not require unnecessary exposure.

Responsible AI use begins before the prompt is submitted. Identify sensitive information, remove what is unnecessary, anonymise what can be anonymised and use approved systems for information that requires stronger protection.

01Identify

Recognise sensitive data.

→
02Reduce

Remove unnecessary details.

→
03Check

Confirm the tool is appropriate.

→
04Use

Submit only what is needed.

BUILD SAFER PROMPTS

Describe the task without exposing unnecessary data.

Use placeholders, anonymised facts and clear constraints when building prompts for workplace tasks.

Open Prompt Builder
CONTINUE LEARNING

AI & Workplace Information

Learn how to think about company data, internal documents and workplace information when using AI tools.

CONTINUE →