AI agency in Dubai · a free AI skill every morning at 6:00 Gulf time
AskWisely.ai
Guide5 min read

Client Data Safety With AI Tools: Rules for Your Team

A practical set of rules for using AI assistants without exposing client information: what never to paste, which accounts to use, and who checks what.

To keep client data safe when your team uses AI tools, allow only approved business accounts, sort information into three levels with a clear rule for each, strip out names and identifiers before pasting, and keep a person responsible for every output. In a small company the usual risk is an ordinary habit, such as pasting a full contract into a personal account, and a one page policy that people actually follow addresses it.

This guide is general information, not legal advice. Data protection obligations depend on your sector, your contracts and where you operate, so confirm your position with a qualified legal adviser.

Where the risk really is

When someone types into an AI assistant, that text leaves your company and is processed on the provider's systems. Whether that is acceptable depends on what the text contains, which account was used, and what the provider's terms say about storing it and using it to improve their models.

The common failure points are simple:

  • Personal accounts. Staff use free or personal plans whose terms the company has never read and whose history it cannot see.
  • Pasting everything. A whole email thread or contract is pasted in when two paragraphs would do.
  • Identifiers left in. Names, phone numbers, Emirates ID numbers, passport details and bank information travel with the text.

Sort your data into three levels

A simple traffic light scheme works better than a long classification policy.

Green: free to use

Public information and generic work. Marketing copy, published prices, product descriptions, general questions, text with no client or staff details.

Amber: use with care

Internal and client related information that is not highly sensitive. Use it only in approved business accounts, and remove names and identifiers first.

Red: never paste

Identity documents, passport and visa copies, Emirates ID numbers, bank and card details, salaries and HR files, medical information, passwords and access keys, and anything a client has given you under a confidentiality agreement that does not allow outside processing.

Eight rules for the team

1. Use approved tools only. Publish a short list of the assistants the company has chosen. Anything else needs approval first.

2. Use business accounts. Company managed plans let an administrator add and remove users and see settings in one place. When someone leaves, their access ends that day.

3. Check the data settings. For each approved tool, read how conversations are stored, how long they are kept, and whether they are used for training. Choose the options that limit this, and write down what you chose.

4. Paste the minimum. Give the model only what the task needs. A summary of the issue often works as well as the full file.

5. Remove identifiers. Replace names with Client A and Supplier B, and remove phone numbers, addresses and account numbers.

6. Limit connections. If you connect an assistant to email, files or a CRM, grant the narrowest access that works and review it every few months.

7. Review before sending. A named person reads every output that goes to a client and checks facts, figures and names against the source.

8. Report mistakes quickly. If something sensitive is pasted by accident, the person tells the owner of the policy straight away, without fear of blame, so the conversation can be deleted and any needed steps taken.

Client contracts and local rules

Check what you have already promised. Many client contracts and non-disclosure agreements restrict sharing information with third parties or moving it outside a country, and an AI provider can count as a third party. If a contract is unclear, ask the client or keep that client's data at the red level.

The UAE has a federal personal data protection law, free zones such as DIFC and ADGM have their own data protection regulations, and other GCC countries have their own laws too. Which of these apply to you, and how, is a question for a qualified legal adviser.

Skill files help with safety

A skill file holds instructions, never client records, so it can be shared freely inside the team. It is also a good place to write safety rules once: tell the model to use placeholders for names, to refuse to guess missing facts, and to flag anything that looks like sensitive data in the input. The ready made files in the Vault are written this way, and the skill file builder in Labs includes a rules section for it.

A one page policy you can adopt this week

Put the approved tools, the three lists and the eight rules on a single page with examples from your own business, name who reviews client facing output, walk the team through it, and review it every six months. The AI readiness scorecard in Labs includes a data section that shows where your current practice has gaps. If you want help setting this up, contact us.

Questions people ask

Is it safe to put client information into ChatGPT, Claude or Gemini?

It depends on the information, the account and your obligations. Low sensitivity information with identifiers removed, in an approved business account with suitable data settings, is a reasonable working practice for many companies. Highly sensitive data should stay out. This is not legal advice, so confirm against your contracts and the laws that apply to you.

Do AI tools train on what my team types?

It varies by provider, by plan and by settings, and the terms change over time. Read the current data terms for the exact plan you use, choose the settings that limit retention and training, and record your choice.

What should we do if someone pastes sensitive data by mistake?

Act quickly and calmly. Delete the conversation where the tool allows it, tell the person responsible for the policy, record what was shared, and take advice on whether the client or a regulator needs to be informed.

Share this guide
Newsletter

The daily skill, in your inbox

One email each morning with the new drop. Unsubscribe in one click.

One email each morning. Unsubscribe in one click. We never sell or share your address.