Broker guide
Is ChatGPT Safe for Client Data in 2026?
ChatGPT privacy depends on your plan and settings. See what it keeps, whether it trains on your chats and what client data to keep out.
- Published
- Updated
It depends: ChatGPT’s safety for client data changes with your plan, settings and the information you submit. Australian mortgage brokers must keep identifying client information out of personal accounts, including paid Plus and Pro accounts. Turning training off reduces one use of your data but doesn’t stop storage or all authorised access.
Is ChatGPT Safe for Client Information?
ChatGPT can help with a de-identified question, but a personal account isn’t an appropriate place for a client’s payslip or loan file. A business workspace changes the contractual protections and training defaults. Your brokerage still needs to approve the service and the information staff can submit.
Separate these risks before deciding what to enter:
- Storage means OpenAI keeps the conversation or uploaded file on its systems.
- Training means OpenAI uses content to improve its models.
- Human review means authorised people can access content for specific purposes.
OpenAI says limited authorised staff and service providers can access consumer content for security, support, legal matters or model improvement. Opting out removes the model-improvement purpose, while the other purposes remain. Its consumer data guidance describes these access restrictions.
A chat being absent from public search doesn’t make it confidential in the same way as an approved client-record system. Shared links, connected services and the wrong workspace can also change who receives information.
What ChatGPT Keeps and for How Long
ChatGPT keeps regular and archived chats until you delete them or an applicable workspace retention policy removes them. Archiving hides a conversation without starting deletion.
Under OpenAI’s chat and file retention rules, deletion removes a chat from your account view immediately. Permanent deletion is scheduled within 30 days. Exceptions cover content already de-identified and separated from your account, plus longer retention required for security or legal obligations.
Files can exist separately in Library or Space, so deleting the conversation doesn’t necessarily delete an uploaded document. Files added to a custom GPT or project have their own retention lifecycle. Remove the relevant saved file or container too.
Memory Is Separate From Chat History
ChatGPT memory can use details from earlier conversations to personalise later answers. Turning memory off doesn’t delete those conversations, and deleting a chat doesn’t necessarily delete a separate saved memory.
OpenAI’s memory controls let you remove remembered information and its sources. For a client detail, remove the memory and the chats or saved files containing it. OpenAI can retain deleted-memory logs for up to 30 days for safety and debugging.
Temporary Chats Still Have Retention
A temporary chat stays outside history and isn’t used for model improvement while it remains temporary. It doesn’t create or update memories. OpenAI can keep a copy for up to 30 days for safety.
Personalised temporary chats can use existing memories and custom instructions. Select Unpersonalized before starting when you don’t want that context used. Saving a temporary chat converts it into a regular chat governed by your account’s settings.
Australian Storage Doesn’t Mean Australian Processing
OpenAI says consumer content is stored in the United States and elsewhere through its systems and service providers. Using ChatGPT in Australia doesn’t select an Australian data centre.
Eligible new Enterprise and Edu customers can select Australia for supported content stored at rest. OpenAI’s residency documentation separates that storage from inference, the processing that generates answers. Australian storage doesn’t establish that every processing step stays in Australia.
ChatGPT Business has a separate, gradual storage-region rollout. Its available regions appear during checkout. For a non-US region, OpenAI still stores a temporary copy of every prompt and response in the United States for safety and abuse monitoring.
Does ChatGPT Train on Your Chats?
Yes, personal ChatGPT workspaces allow model-improvement sharing by default, but you can turn it off. OpenAI’s data controls instructions give this web procedure:
- Open your account menu and select Settings.
- Select Data controls.
- Turn off Improve the model for everyone and select Done.
- Reopen Data controls and confirm the setting remains off.
For signed-in users, the setting applies across web and mobile. Your chats can remain in history after you turn training off. Voice has a separate data-sharing control, so don’t assume the text setting covers every recording.
Business and Enterprise workspace content is excluded from training by default under OpenAI’s business commitments. That default doesn’t make the workspace a service with no data retention.
Does OpenAI sell your data?
No, OpenAI’s privacy policy says, “We don’t ‘sell’ Personal Data”. The policy qualifies this using US state privacy-law definitions. Its consumer guidance also says ChatGPT conversations aren’t shared with advertisers.
OpenAI still discloses information to service providers and in other circumstances described in the policy. A promise not to sell data doesn’t mean nobody else processes it.
Personal Plans vs Business Plans
Personal and business plans differ in data use and organisational control, even when both require payment. Plus and Pro remain personal plans.
| Control | Personal Free, Plus and Pro | ChatGPT Business | ChatGPT Enterprise |
|---|---|---|---|
| Model training | Sharing on by default; personal opt-out available | Excluded by default | Excluded by default |
| Chat retention | Saved until deletion, with deletion exceptions | Saved chats follow applicable retention rules | Organisation-configured retention available |
| Organisation access | Individual account settings | Workspace membership and administrative controls | Broader contracted access and feature controls |
| Sign-in control | Personal account sign-in | Single sign-on available | Enterprise authentication and access controls |
| Processing terms | Consumer terms and privacy policy | Business terms and privacy commitments | Business agreement and privacy commitments |
| Australian residency | Consumer hosting in the US and other countries | Separate region rollout; US safety copies remain | Australian storage available to eligible customers; processing scope is separate |
OpenAI’s Enterprise privacy commitments describe training defaults, encryption and qualifying retention controls. Its Business sign-in documentation confirms single sign-on support. A Data Processing Addendum sets contractual processing responsibilities where it applies to your agreement.
Business chats aren’t automatically visible to other members. However, workspace sharing and administrative access differ from a personal account. Usage analytics doesn’t itself let admins read every chat, while owners can export member data where that option is available.
Use the ChatGPT Business guide to assess the workspace choice. Your brokerage must still approve its use and protect client information. No plan name alone establishes compliance with the Privacy Act.
What a Broker Must Never Enter
Keep identifying client records out of personal ChatGPT accounts. Even in an approved business workspace, submit only information your brokerage authorises for that purpose.
Keep these out of an ordinary prompt:
- A client’s name beside income, debts, assets or other financial details.
- Tax file numbers, account numbers, passwords and verification codes.
- Passports, driver licences and other identity documents.
- Bank statements, payslips and tax documents containing identifying details.
- Information a client supplied in confidence without authorisation for this use.
Replacing a name with initials isn’t enough when an address, employer or unusual circumstances still identify the client. Remove identifying combinations, document metadata and unnecessary exact figures. Use broad categories where the answer doesn’t depend on precision.
For this hypothetical example, both prompts describe invented people:
Incorrect: “Draft a lender note for Mia Carter at 18 Example Street, employed by Example Plumbing, earning $104,280 with account 12345678.”
Safer: “Draft a generic lender-note template for an employed applicant. Leave income, employer, address and account details blank for completion in our approved system.”
The safer prompt asks for structure without sending a client record. Enter the real details only in your approved system. De-identification reduces exposure, but it doesn’t override your brokerage’s restrictions on sharing confidential information.
ChatGPT Data Breaches and Security
ChatGPT has had a documented cross-user exposure, and OpenAI has also reported a supplier incident affecting some ChatGPT users. The incidents exposed different information.
In its 2023 incident report, OpenAI described a software bug exposing other users’ chat titles and potentially a new conversation’s first message.
Payment-related details could also be visible for 1.2% of Plus subscribers active during a nine-hour window. These included names, email and payment addresses, card type, expiry and the last four digits. Full card numbers weren’t exposed.
OpenAI’s 2025 supplier incident report concerned stolen analytics data. It affected some API users and a limited number of ChatGPT users who submitted support tickets or were logged into the API platform. Potential exposure included names, email addresses, approximate locations and device or account metadata.
OpenAI said the supplier incident didn’t expose chat content, passwords, payment details or government IDs. It wasn’t a breach of OpenAI’s own systems. Neither incident makes a training opt-out a substitute for controlling what you submit.
If client information is exposed, activate your brokerage’s incident response immediately. Stop further sharing, preserve the evidence securely and tell the person responsible for privacy and security. Record the affected information, account, workspace and known recipients, then seek containment and deletion assistance from OpenAI.
The Office of the Australian Information Commissioner (OAIC) sets out containment, assessment, notification and review. Where its Notifiable Data Breaches scheme applies, assess whether notification is required. Don’t wait for a vendor investigation to begin your response.
Your mortgage broker information security process needs to cover accidental AI uploads as well as compromised accounts.
Delete Your ChatGPT History
Deleting ChatGPT history removes saved conversations from your view and starts the vendor’s deletion process. It doesn’t instantly erase every retained copy, remove separate memories or recall information already shared with another service.
Follow the delete ChatGPT account guide for chat deletion and the separate account-deletion procedure. Also review saved memories, Library or Space files, project files and shared links.
Archiving a chat leaves it stored.
If a client record was uploaded accidentally, preserve the incident details securely before removing the content. Delete the relevant copies, request further assistance through OpenAI’s privacy process and follow your brokerage’s incident procedure. A disappearing conversation confirms removal from your view, not immediate completion of backend deletion.
