Quick answer: Putting customer data into ChatGPT is only reasonably safe on a business plan like ChatGPT Business or Enterprise, where inputs aren't used for training by default. On Free, Plus and Pro, chats train models unless switched off, and Temporary Chats are kept up to 30 days. Safer default: keep personal information out.
The real risk usually isn't ChatGPT itself. It's a staff member pasting a customer complaint, a spreadsheet of contacts or a medical note into a personal account nobody in the business controls. The Australian privacy regulator agrees: the OAIC recommends organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools. If you want AI working on customer data properly, there are private AI options built around your own data controls, and they're covered further down.
This article is for Australian business owners, operations managers and IT leads who suspect staff are already using ChatGPT and want a clear rule they can enforce.
Why is customer data in ChatGPT a privacy problem in the first place?
Customer data in ChatGPT becomes a problem because pasting it in is a use or disclosure of personal information, and the Privacy Act 1988 limits what you can do with it. Under Australian Privacy Principle 6, personal information can generally only be used for the purpose it was collected for, unless the person consented or would reasonably expect the new use.
Most customers gave you their details to get a quote or a delivery, not to have their complaint processed by an overseas AI service. The OAIC guidance on privacy and commercially available AI products says consent is the safer path for AI-related secondary uses, and that your privacy policy should explain your use of AI clearly.
Smaller businesses sometimes assume none of this applies to them. That depends on turnover and on whether an exception applies, which is covered in Does the Privacy Act Apply to Small Businesses Using AI Tools?
Does it matter whether staff use free ChatGPT or a business plan?
Yes, the plan makes the biggest difference to what happens with the data. Consumer and business versions of ChatGPT have different default settings for training and different admin controls.
| ChatGPT Free, Plus, Pro (consumer) | ChatGPT Business, Enterprise, API | |
|---|---|---|
| Used to train models? | Yes by default. "Improve the model for everyone" is on until the user turns it off. | No by default, according to OpenAI's business data page. |
| Who controls the setting? | Each individual user | The organisation's admin |
| Retention | Chats kept until deleted; Temporary Chats kept up to 30 days for safety | Retention controls for qualifying organisations; zero data retention available on the API for eligible customers |
| Data residency | No choice | Australia is listed as a data-at-rest region for eligible customers |
| Encryption | Standard | AES-256 at rest, TLS 1.2+ in transit, optional customer-managed keys |
Sources: OpenAI Data Controls FAQ and OpenAI business data privacy page, checked October 2026. Vendor terms change, so recheck before relying on them.
The catch with consumer accounts is control. Even if one careful employee switches training off, you can't see or enforce that across the team.
What actually happens to customer data once someone pastes it in?
Once pasted, the data leaves your systems and is stored on OpenAI's infrastructure under that account's settings. On a consumer account with default settings, it may be used to improve future models. On any plan, it sits in chat history until someone deletes it, which means it can be exposed if that account is compromised or shared.
There is also the output to think about. If ChatGPT drafts a reply using a customer's details and the draft is wrong, you're still responsible for what gets sent.
Which customer data should staff never paste into ChatGPT?
Staff should never paste these five types of data into a consumer AI tool, and should only use them in an approved business tool when there's a clear reason:
- Health information, including NDIS, aged care or clinic notes. This is sensitive information under the Privacy Act.
- Identity and financial details: tax file numbers, driver licence numbers, bank or card details.
- Full customer lists or CRM exports, even "just to clean up the formatting".
- Complaints and disputes that name individuals, especially anything heading to a regulator or lawyer.
- Employee records: performance notes, payroll, medical certificates.
A useful test: if this text turned up in a data breach notification, would you have to tell the customer? If yes, it doesn't go into a personal AI account.
What are the safer ways to use AI with customer data?
The safer options are a business AI plan with admin controls, de-identifying data before it goes in, or a private AI setup that keeps data inside systems you control. Which one fits depends on how much customer data the work genuinely needs.
- Business plan with training off by default. Fine for drafting, summarising and internal work, as long as you update your privacy policy and set retention.
- De-identify first. Replace names, emails and account numbers with placeholders such as "Customer A". Many tasks don't need the real details at all.
- Private AI inside your own environment. For chatbots, document processing or anything touching health or financial data, a private deployment in an Australian cloud region gives you control over access, logging and retention.
Before any of these, the OAIC also expects a privacy impact assessment for higher-risk uses and human review of AI output.
Example: a 25-person Geelong physiotherapy group
Front desk staff were pasting patient emails into free ChatGPT to draft replies. The fix took two weeks: a ChatGPT Business workspace with training off, a rule that no clinical notes go in, de-identified templates for common replies, and one line added to the privacy policy about AI-assisted correspondence. Clinical summaries stayed out of general-purpose AI completely. (Illustrative scenario, not a client case study.)
When can you fix this yourself, and when do you need help?
You can fix this yourself if staff only use AI for drafting and internal work: buy a business plan, write a one-page rule, and train people in a 30-minute session. Bring in help if customer data needs to flow into AI automatically, if you handle health or financial information, or if you're building a customer-facing chatbot. Those need access controls, logging and a privacy impact assessment that a policy document alone won't give you.
A simple staff rule you can use this week
Copy this into your staff handbook or team chat:
"Only use the company ChatGPT workspace for work. Never paste customer names, contact details, health, financial or ID information into any AI tool unless your manager has approved that specific use. If unsure, replace names with 'Customer A' first. Always check AI output before it goes to a customer."
Zynex Technologies is a Melbourne-based AI security and automation consultancy that helps Australian businesses set rules like this and build private AI systems that keep customer data under their control.
Plan your next steps with Zynex Technologies
Want AI on customer data without the privacy risk?
If your team needs AI working with customer records, we can set up the controls or build a private system that keeps the data in Australia. Book a call with our Melbourne team or phone 03 9292 2022.
Book a call