Is Your Team Pasting Customer Data into ChatGPT?

Before an employee pastes customer data into ChatGPT, your business needs to know which account they are using, what they are sharing and whether that use is approved. A useful analysis should not depend on staff making privacy decisions on their own.

A quick analysis can move customer data outside your controls

A sales manager downloads a Salesforce report, uploads it to a personal ChatGPT account and asks which customers are likely to leave. The spreadsheet includes names, email addresses, purchases and notes from recent calls. The intention is useful. The upload still sends that information to another service.

This is an illustrative scenario, not a reported Adaptal customer incident. It shows why a routine task needs a clear rule: staff should know which AI tools and accounts are approved, what information they can use, and who can approve an exception.

Copying information out of Salesforce also means its existing sharing rules no longer control that copy. The original record may be restricted to a small team, while the exported file, chat or generated summary follows different access and storage settings.

What are the actual risks?

• Unapproved disclosure. Customer information may reach a provider your business has not assessed for that purpose.
• Extra copies. An export can remain in downloads, chat history or shared files after the analysis is finished.
• Lost control over access. Personal accounts, shared chat links and connected tools can make it harder to manage who sees the information.
• Confidentiality problems. Pricing, contracts, commercial plans and internal notes can be sensitive even when they contain no personal information.
• Incorrect conclusions. An AI-generated assessment can misrepresent a customer. Someone must check the evidence before acting on it.

Pasting information into ChatGPT does not automatically make it public or prove that another user can retrieve it. The concern is whether your business has authorised and assessed how that information will be processed, retained and accessed.

Personal ChatGPT, business workspaces and APIs are different

Do not assume that paying for an individual subscription makes it an approved business system. Check the actual workspace and its settings.

Sources: ChatGPT data controls, business workspace protections and OpenAI API data controls.

“Not used for training” is one protection. It is not permission to share every customer record.

Your organisation still needs to decide whether the task is appropriate, whether customer information is necessary and whether the selected tool meets its requirements. An approved account is a starting point, not approval for every possible use.

What Australian businesses should consider

The OAIC recommends, as best practice, that organisations avoid entering personal information—especially sensitive information—into publicly available generative AI tools. For entities covered by the Privacy Act, existing privacy obligations apply to AI use, including how personal information is used or disclosed. Read the OAIC’s guidance on commercially available AI products.

The right assessment depends on your business, the information and the proposed use. Ask your privacy or legal adviser about applicable obligations and customer commitments. This article provides practical operational guidance, not a legal assessment of your organisation.

Use less data to answer the same question

Start by asking whether the analysis can happen in an existing approved report or dashboard. A request to understand monthly sales may need totals by product and region, not a file containing every customer’s contact details.

If an approved AI workflow is useful, prepare the minimum information it needs. Our suggested approach is to remove unnecessary columns, summarise before exporting and use invented sample records when asking for a formula or analysis method.

Removing names alone may not make data anonymous. A unique job title, location, purchase or free-text note can still identify someone. Treat customer IDs as potentially identifiable when they can be linked back to a person. Check notes, attachments, hidden spreadsheet tabs and screenshots too.

For example, instead of uploading real customer records to ask how to calculate churn, use fictional rows to develop the calculation. Have a colleague check it, then apply it inside your approved business environment. The AI can help with the method without receiving the customer list.

Give employees a rule they can follow

A short policy is easier to use when it names the approved tools, the data allowed in each one and the person responsible for questions. Here is a starting point to adapt with your IT and privacy owners:

Use only company-approved AI accounts for work. Do not paste or upload identifiable customer information, confidential documents, passwords or access keys unless that tool and use have been explicitly approved. Use fictional examples or approved summaries where possible. If you share something by mistake, report it promptly.

Support the policy with practical controls:

1. Provide an approved route. Give staff a usable alternative and examples of acceptable tasks.
2. Review export access. Check who can export Salesforce reports and whether they need all the available fields.
3. Set up account controls. Use company-managed access and review sharing, retention and integrations for the selected product.
4. Test the workflow. Check what information is actually sent, what is stored and how outputs are reviewed.
5. Revisit the setup. Assign an owner to review changes when tools, permissions or business needs change.

An integration can make the approved process easier, but it also creates a data path that needs review. Limit the records and fields it can access; do not connect an entire CRM simply because the connector allows it. Our integration services can help define that scope.

If customer information has already been shared

Ask the employee to stop further uploads and notify the responsible manager or privacy/security contact promptly. Establish what was shared, when, through which account, and whether any links or third-party tools were involved. Record the necessary facts securely without spreading more copies of the data.

The incident owner should assess access, retention and available removal options, preserve necessary evidence, and determine whether legal or contractual notification obligations apply. Deleting a visible chat should not be treated as proof that every copy has disappeared. If passwords or access keys were included, involve IT immediately to revoke or replace them.

Make it easy for staff to report mistakes. The aim is to contain the issue, understand what happened and improve the workflow.

Make the approved option the easy option

Employees often use AI because they need a faster way to finish a real task. Find out which tasks they are doing, then provide a clear and workable route. That may be a better Salesforce report, an approved business AI workspace or a narrowly scoped integration.

Adaptal can help review Salesforce reporting, data access and integration requirements, and plan practical AI workflows with your business and security teams. See our AI services or read which skills to develop as AI changes your work.

Product guidance checked on 26 October 2026. Settings and terms can change; review the linked provider guidance before approving a workflow.