Security & Data Governance

Can I Put Customer Data Into ChatGPT? A Malaysian PDPA Answer

Whether Malaysian companies may enter customer data into ChatGPT or similar AI tools under the PDPA, which plans carry less risk, and what to do instead.

  • By
  • Updated
  • 3 min read
  • Facts checked
Short answer

Only if you have a lawful basis, your privacy notice covers it, and the tool is a business plan with a vendor contract. Do not use personal accounts or free plans for customer data. For sensitive data such as health information, get explicit consent first or remove identifiers before anything is entered.

Part of the guide: PDPA and AI in Malaysia: What Companies Must Know (2026 Guide)

On this page
  1. What counts as customer data under the PDPA?
  2. What are the four checks before you paste?
  3. Which plans carry more risk?
  4. How can staff use AI without exposing customer data?
  5. What if it already happened?
  6. What should you do next?
  7. Sources
  8. Frequently asked questions

The short answer is in the box above. The longer answer depends on four checks you can run in ten minutes.

Part of the guide: PDPA and AI in Malaysia. Checked 22 October 2026.

What counts as customer data under the PDPA?

Anything that relates to a living person who can be identified from it: name, IC or passport number, phone, email, address, photo, order history tied to a name, or an account ID that can be linked back to a person. Health, biometric, religious and political data are sensitive personal data and need explicit consent. If you could not publish the text on your website without worry, treat it as personal data.

What are the four checks before you paste?

  1. Lawful basis. Did the customer consent, or is the processing necessary for a contract with them? Using the data for a purpose they were never told about breaks the Notice and Choice and Disclosure principles.
  2. Privacy notice. Does your notice mention disclosure to service providers, including AI tools, for the stated purpose?
  3. Tool and plan. Is it a company-managed business plan with a vendor contract, or someone's personal login? Anthropic says its commercial products, including the API, are not used for training by default, and Microsoft says prompts and responses in Microsoft 365 Copilot are not used to train its foundation models. Consumer plans follow different terms, and we have not verified OpenAI's current terms here, so read them for the plan you would buy.
  4. Where it goes. Many AI services run outside Malaysia. Section 129 of the PDPA requires one of its conditions before personal data is transferred abroad. See Sending data to US AI tools.

If any answer is "I don't know", remove the personal data first.

Which plans carry more risk?

Plan type Typical position Risk for customer data
Free or personal account Vendor terms for consumers; settings differ; no contract with you High. Avoid.
Paid individual plan Better settings, still a personal contract Medium-high. Not a company control.
Business or team plan Company admin, no training by default, vendor terms Medium. Acceptable with notice, basis and transfer check.
API or private deployment Contract, logs, access control, possible regional hosting Lower. Best for systems that touch customers.

Vendor terms change, so read the current pages and keep a dated copy.

How can staff use AI without exposing customer data?

Use a three-step habit: strip, then ask, then check. Strip the identifiers (swap "Ahmad bin Ali, 880101-14-xxxx" for "Customer A"). Ask the tool for the draft or analysis. Check the output before it goes back to the customer. For repeat work, an integration that removes identifiers automatically is safer than asking people to remember.

Hypothetical example for a KL online retailer: complaint emails are summarised by an approved business-plan tool after a script replaces names and order numbers with codes. The mapping stays in the company's own system.

What if it already happened?

Stop further use, find out what was entered and where, and ask whether it is a "personal data breach" as defined in the amended Act. A breach likely to cause significant harm, or affecting more than 1,000 people, must be reported to the Commissioner within 72 hours under the 2025 breach notification guideline. Talk to your lawyer or DPO the same day. Write down what you find; the record helps if the Commissioner asks.

What should you do next?

Put the rule in writing in a short policy. Start with the AI policy template. If you want staff trained on safe habits, see our courses, or book a free call.

Staff learn to sort requests into allowed, allowed with care and never in the safe-use module of Generative AI at Work.

Sources

(checked November 2026)

General information, not legal advice.

Frequently asked questions

Is it illegal to paste customer names into ChatGPT?

Not automatically. The PDPA regulates how personal data is handled, not which tool you use. It becomes a problem when there is no lawful basis, the customer was not told, the tool is unapproved, or the data goes abroad without meeting a section 129 condition.

Does turning off chat history make it safe?

It reduces what the vendor keeps and may limit training use on consumer plans, but it does not create a lawful basis, a customer notice or a transfer basis. Treat it as one control, not a solution.

What can staff safely put into AI tools?

Public information, your own drafts, and data with identifiers removed. Replace names, IC numbers, phone numbers and account details with placeholders before pasting, and keep a rule that sensitive data never goes into tools that are not approved.