The Short Answer for Brokers
It depends on your ChatGPT plan, settings and the information you enter: keep identifiable client documents out of a personal account. Use fictional examples or properly de-identified material for everyday drafting and explanations. ChatGPT Business or Enterprise provides stronger business data protections, but your brokerage must approve the processing arrangement before staff upload client information.
What ChatGPT Stores and Who Can See It
ChatGPT stores saved conversations and can process the files you upload. Paying for a personal subscription does not automatically remove training use or give you business data terms.
Plans and Data Handling
| Account or mode | Model training | Retention and access |
|---|---|---|
| Personal Free, Plus or Pro | Model-improvement sharing is enabled by default in personal workspaces. You can turn it off. | Saved chats remain until deleted. Authorised personnel and service providers can access content for specified purposes. |
| Temporary chat | Not used for model improvement while it remains temporary. | Stays outside history unless saved. OpenAI may keep a copy for up to 30 days for safety. |
| ChatGPT Business | Business content is excluded from training by default. | Workspace retention controls and business access rules apply. A business workspace still processes and stores data. |
| ChatGPT Enterprise | Business content is excluded from training by default. | Administrators control retention and can use compliance tools to access conversation records. |
OpenAI’s consumer data guidance limits human access to authorised people with a need to access the content. Reasons include security investigations, support, legal matters and model improvement unless you opt out. A training opt-out leaves the other permitted reasons in place.
OpenAI says it does not sell your data or share ChatGPT conversations with advertisers. Its privacy policy nevertheless permits disclosures for operating the service and complying with legal requirements. The policy also describes limited marketing-partner disclosures. Selling information, using content for training and allowing a service provider to process it are different activities.
Under OpenAI’s chat and file retention rules, deleting a saved chat removes it from view immediately. Permanent deletion is scheduled within 30 days. Exceptions apply where data is already de-identified and disassociated from your account, or longer retention is required for security or legal obligations.
Files saved separately in Library or Space can remain after the conversation is deleted. Saved memories are another separate source of information. A disappearing sidebar entry cannot establish that every copy of a client’s information has been removed.
Is ChatGPT Confidential or Private?
A consumer ChatGPT account does not establish that your brokerage can lawfully disclose a client’s information to OpenAI. Access controls and privacy settings also do not create legal professional privilege.
OpenAI’s consumer terms make you responsible for having the rights and permissions needed to submit your input. A client’s decision to give you a payslip for a loan application does not, by itself, establish permission for every later AI use.
The Office of the Australian Information Commissioner (OAIC) recommends keeping personal information out of publicly available generative AI tools. Its guidance particularly warns about sensitive information. A brokerage must assess its privacy duties and the provider’s terms before using client information.
What Reported Incidents Exposed
OpenAI’s 2023 ChatGPT incident report describes a software bug that exposed other users’ chat titles and potentially the first message of a new conversation. Some active Plus subscribers’ payment-related information was also potentially visible. It included contact and billing details and partial card information. Full card numbers were not exposed. OpenAI says it patched the bug.
A separate 2025 Mixpanel incident affected analytics data held by OpenAI’s analytics provider. It involved some API users and a limited number of ChatGPT users who submitted help tickets or logged into the developer platform. Potentially exposed information included names, email addresses and account identifiers. OpenAI says chat content, passwords and payment details were not exposed.
The incidents involved different systems and different information. Neither report establishes that every ChatGPT user’s conversations became public. They do show why a brokerage must limit what it submits and prepare for a provider incident.
Settings That Reduce the Risk
Use ChatGPT’s data controls before entering work material, and keep identifiable client data out of consumer chats even after changing them.
Turn Off Model Improvement
OpenAI’s data controls instructions give this web path: account menu, Settings, Data controls, Improve the model for everyone. Turn the setting off and select Done. Reopen the setting to confirm it remains off.
For ordinary personal chats, this disables the model-improvement sharing controlled by that setting. It does not delete saved chats or prevent every form of authorised access. Voice has a separate training-sharing control. OpenAI also identifies a separate exception for specialised Trusted Access sessions, whose training terms need their own review.
If you use ChatGPT without signing in, open Settings and turn off Improve the model for everyone there. OpenAI says the signed-out choice is saved in that browser. Guest use does not remove the need to control what you submit.
Use Temporary Chat for Suitable Work
Open a new chat and select Temporary. OpenAI’s Temporary chat instructions describe a choice between Personalized and Unpersonalized before the first message. Choose Unpersonalized when you want the session to avoid existing memories, custom instructions and plugins.
While the chat remains temporary, it does not create or update memories and is excluded from model improvement. OpenAI may retain a copy for up to 30 days for safety. If you save it, it becomes a regular chat governed by your account settings.
Temporary chat therefore reduces history and training exposure. It still sends the material to OpenAI for processing.
Remove Each Stored Source
Delete a conversation containing client information, then check for separately saved files. Archiving keeps a chat stored.
OpenAI’s memory guidance explains that deleting a chat does not necessarily remove a saved memory. Review the memory controls as well as the original conversation. Remove shared links and check whether information was sent to an external app or service.
If a client document was uploaded accidentally, tell your brokerage’s privacy contact promptly. Deleting the chat is one containment step. The brokerage still needs to assess what was disclosed and whether other action is required.
What Never to Paste Into ChatGPT
Keep identifiable client records out of consumer ChatGPT and other public AI tools. This includes information copied into a prompt as well as attachments.
- Tax file numbers and government identity numbers.
- Passport, driver’s licence and identity-verification images.
- Full bank statements, account numbers and payment-card details.
- Unredacted payslips, tax returns and loan applications.
- Health information and details of hardship or family circumstances that identify the client.
- Passwords, one-time codes and access credentials.
Start with a fictional case when the task is to explain a concept or draft a general email. A made-up salary and loan amount can support a repayment explanation without exposing a real borrower.
If actual figures are necessary, extract only the figures required for the task into a separate document. Remove names and addresses, then remove account references and other identifying details. Check combinations of facts that could still identify someone, such as an unusual employer and a distinctive transaction.
For a PDF, use permanent redaction that removes the underlying content. A black rectangle over visible text can leave the text recoverable. Check the exported copy by searching and trying to copy the removed information before considering an upload.
Keep the identifiable original in the brokerage’s approved system. Redaction reduces what you disclose, but it does not replace approval for the remaining information or a check that the output is accurate.
When a Business Plan is the Better Choice
Choose a managed business workspace when staff need AI for ongoing brokerage work and the business needs control over access and data handling. A personal Plus or Pro subscription remains a personal account for these purposes.
OpenAI’s business privacy commitments exclude Business and Enterprise content from training by default. OpenAI describes administrator-controlled retention for both. Enterprise also provides conversation audit access through compliance tooling. These protections require deliberate workspace configuration.
The ChatGPT Business privacy guide says members do not automatically see one another’s private chats. Sharing a conversation is a separate action. Organisational access and retention policies still matter, so staff must understand the workspace’s approved uses.
Before approving client documents, have the brokerage’s privacy contact assess the contract and retention settings. Set rules for staff access, external apps and shared links. Keep only the information required for the approved task.
OpenAI’s Zero Data Retention arrangement applies to eligible API use, subject to qualification and endpoint rules. It is not an automatic feature of a ChatGPT subscription or Temporary chat.
For a payslip already open on your screen, the next action is to decide whether the task can use fictional or de-identified information. If identifiable details are necessary, use only the processing arrangement your brokerage has approved for that information.
