Is ChatGPT safe? What OpenAI keeps from your chats
Is ChatGPT safe for store data? Check ChatGPT privacy settings, what it keeps and trains on, and which plans protect customer details.
- Published
- Last updated
Yes, ChatGPT can be safe for general store work when you keep customer details and confidential material out of a personal account. Free, Plus and Pro accounts save ordinary chats and enable model training by default. Turning training off changes that use of your data, but it doesn’t delete the conversation.
For a customer complaint, give ChatGPT an anonymous description of the problem. Add the customer’s details yourself after you’ve checked the reply. A business workspace changes the data terms, but your store still needs rules about what staff can upload.
Is ChatGPT safe for store data?
ChatGPT is suitable for tasks such as rewriting public product descriptions or drafting a reply from an anonymous scenario. An identifiable customer complaint needs a different approach because submitting it gives an external service access to that information.
Your chats aren’t automatically public. However, OpenAI’s systems process them, and authorised people can access content for specific purposes. OpenAI lists security investigations, support and legal matters among those purposes. Model improvement is another purpose when you haven’t opted out. Its consumer data guidance describes those access limits.
That makes privacy and security different questions. Access controls can protect stored chats while the service still retains and processes them. A safe staff rule is to use public or anonymised material in personal accounts and reserve approved business data for an assessed workspace.
What ChatGPT saves and trains on
ChatGPT can retain conversations, remembered information and uploaded files in different places. Changing one setting doesn’t necessarily remove the others.
- Ordinary and archived chats remain saved until you delete them or a workspace retention policy removes them. Archiving only changes where you find a chat.
- Memory can use details from earlier conversations. Deleting a chat doesn’t necessarily delete a separate saved memory. OpenAI’s memory controls explain how to remove remembered information and its sources.
- Uploaded files can be saved separately in Library or Space. Deleting the conversation doesn’t remove a separately saved file. Files in a custom GPT or project have their own retention rules.
To stop ordinary personal-account conversations being used for model improvement, open Settings, then Data controls, and turn off “Improve the model for everyone”. OpenAI’s data controls instructions cover the web and mobile apps. Your saved history remains available after the change.
Temporary chats
A temporary chat stays out of history and isn’t used for model improvement while it remains temporary. It doesn’t create or update memories. OpenAI may keep a copy for up to 30 days for safety purposes.
Temporary chats can use existing personalisation if you enable it. Saving a temporary chat turns it into a regular conversation with your normal account settings. Temporary mode therefore doesn’t mean that OpenAI never receives or stores the content.
OpenAI says it doesn’t sell your data or share ChatGPT conversations with advertisers. That statement is separate from its permissions to process content and use eligible conversations for training.
Deleted chats and accounts
A deleted chat disappears from your account view immediately and is scheduled for permanent deletion within 30 days. Exceptions apply to data already de-identified and separated from your account, and to security or legal retention. OpenAI’s chat and file retention rules explain the separate treatment of saved files.
Account deletion also normally removes personal data within 30 days, with exceptions. OpenAI’s privacy policy allows longer retention for legal duties and some financial records, among other reasons. To remove stored conversations without closing the account, follow the guide to delete ChatGPT history.
Is ChatGPT confidential?
No, a personal ChatGPT account isn’t a confidentiality agreement with your supplier or customer. A private-looking chat window doesn’t give you permission to disclose information covered by a non-disclosure agreement (NDA).
For example, a supplier’s unpublished wholesale prices can remain confidential even when every customer name has been removed. Keep that file out of a personal account. Turning training off doesn’t change the supplier’s restrictions on disclosure.
Business and Enterprise workspaces have different contractual protections and no training on business content by default. OpenAI’s enterprise privacy commitments also describe limited provider access and business data controls. These terms can support an approved business use, but they don’t automatically satisfy your existing NDA. The agreement must permit that processing and any required supplier approval must be in place.
Customer details to keep out of ChatGPT
Keep identifiable customer information out of consumer accounts, including paid Plus and Pro accounts. The Office of the Australian Information Commissioner (OAIC) recommends against entering personal information into publicly available generative AI tools.
Staff must remove the following information before asking for writing help:
- Shopper names, email addresses and phone numbers.
- Delivery or billing addresses.
- Order numbers and identifiable order histories.
- Payment details, identity documents and account credentials.
- Personal details in attachments, screenshots or the quoted email thread.
Here’s a hypothetical request that keeps the drafting task useful:
Write a polite reply to a customer whose parcel arrived after the delivery estimate. Acknowledge the delay. Explain that our team will review the delivery charge. Don’t promise a refund or invent a delivery date.
The prompt gives the problem and the permitted response without sending the original complaint. Remove unusual details that could identify the shopper too. PII redaction helps remove personal identifiers, but a staff member must read the final text before sending it to an assistant.
ChatGPT data breaches and leaks
OpenAI has disclosed incidents affecting chat information and account-related data. Each incident’s scope matters when assessing your store’s exposure.
| Incident | What OpenAI reported | What it means for a store |
|---|---|---|
| ChatGPT service bug in 2023 | Some users could see other users’ chat titles and possibly a new conversation’s first message. Payment-related information could also become visible for a subset of active Plus subscribers. | Even a title or first message can disclose a customer or supplier issue. |
| Analytics-provider incident in 2025 | An attacker obtained account and analytics data from an external provider. Some API users and a limited number of ChatGPT users were affected. OpenAI said chat content and payment details weren’t exposed. | Exposed names and email addresses can support convincing phishing messages. |
In its 2023 incident report, OpenAI said the payment exposure could affect 1.2% of Plus subscribers active during a nine-hour window. Possible fields included names, billing addresses and partial card details. Full card numbers weren’t exposed, and OpenAI patched the bug.
OpenAI’s 2025 incident report distinguishes an external analytics breach from a breach of its own systems. Affected information could include account names, email addresses and approximate location. The report also excludes passwords and API keys from the exposed data.
These reports don’t establish that your own chats leaked. They do support limiting sensitive input and protecting staff accounts against phishing, even when model training is disabled.
ChatGPT plans compared on data use
Free, Plus and Pro use personal-account data settings. Business and Enterprise add organisational controls and different default training terms. A higher personal subscription doesn’t turn the account into a business workspace.
| Plan | Model training | Retention | Administrative control |
|---|---|---|---|
| Free, Plus and Pro | Enabled by default for ordinary personal chats, with an opt-out setting | Saved chats remain until deleted | Individual account controls |
| Business | Business content excluded by default | Workspace retention controls apply | Workspace membership, roles and access controls |
| Enterprise | Business content excluded by default | Administrators set retention policies | Additional permissions and compliance controls |
OpenAI’s business data documentation describes encryption and access management across its business products. Business administrators can access workspace conversations under OpenAI’s stated terms. A work account therefore isn’t private from your employer in the same way as a personal notebook.
The ChatGPT Business guide covers the wider workspace features. For data handling, decide which material the workspace may receive and set its retention policy before staff upload customer records.
ChatGPT and the OpenAI API
A tool using OpenAI’s application programming interface (API) follows different data rules from a personal ChatGPT account. This includes business applications such as shopping agents. OpenAI doesn’t train on API inputs or outputs by default unless the organisation opts in.
Standard API abuse-monitoring logs can contain content and are generally kept for up to 30 days, with stated exceptions. Features that save conversation state or files can retain data separately. The API data controls identify the rules for each endpoint and feature.
Zero Data Retention excludes customer content from abuse-monitoring logs and changes storage behaviour on supported endpoints. It requires OpenAI approval and additional requirements. Some features remain ineligible or have storage exceptions, so the label doesn’t promise that every feature stores nothing.
The application provider can also keep its own conversation history. Before approving a shopping agent or staff assistant, establish what the provider retains alongside the model service. Record who can access each copy and how deletion works across both systems.
Give your store its own shopping agent
Customers ask in their own words. Your agent finds, compares and checks out from your catalogue, right on your site.