ChatGPT is safe enough for most business work when your team uses a business workspace, keeps sensitive information out of it and has a person check what it produces. The bigger risk for a small business usually isn’t a hacker reading chats. It’s staff pasting customer details into personal accounts the business can’t see, control or close when someone leaves.
Below is what OpenAI’s own pages say happens to that information, checked on October 7, 2026. These policies change, so check again before you rely on them. When a business wants AI working on its own customer records and documents, that’s our AI integration work, and it starts with the same questions this post asks.
What happens when someone presses Enter
When someone presses Enter in ChatGPT:
- The text leaves the computer. So does any file, photo or spreadsheet attached to it.
- It travels over the internet to OpenAI on an encrypted connection, which protects it on the way there.
- OpenAI’s servers do the work. The model runs on OpenAI’s systems, not on your machine, and the answer travels back the same way.
- The conversation is stored. OpenAI’s article on how it handles data in consumer services says content is stored on its systems and its service providers’ systems “in the US and around the world.” A saved chat stays there until someone deletes it or a workspace rule removes it.
- Other people may be able to see it. On a business workspace that includes your admin. At OpenAI it includes a limited group of authorized staff and contractors, for reasons such as investigating abuse.
Deleting a chat later doesn’t un-send anything. That same OpenAI article puts it plainly: “Please do not enter sensitive information that you would not want reviewed or used.”
If some of your information should never leave the building at all, the answer is a model running on your own computer or server. We cover that route in local AI for business.
Personal accounts, business workspaces and the API
The kind of account decides most of the answer. Here’s how OpenAI’s pages compare them.
| Personal (Free, Go, Plus, Pro) | ChatGPT Business or Enterprise | API (software built on OpenAI) | |
|---|---|---|---|
| Used to train OpenAI’s models? | May be, unless the setting is off | Not by default | Not by default |
| Who controls how long chats are kept | Each person | Workspace admins | The business that built it. OpenAI says abuse-monitoring logs are usually kept up to 30 days, and some API features store data until the business deletes it |
| Who in your business can see the chats | Only the employee | Workspace admins | Depends on how the software was built |
Sources: OpenAI’s model training article and its enterprise privacy page.
On a personal account, training is on until someone turns off Improve the model for everyone under Settings > Data controls. After that, new conversations aren’t used for training. OpenAI names a catch: a thumbs up or thumbs down on an answer can send that whole conversation for training even after opting out. While the setting is on, information pulled from connected apps may be used for training too.
On a business workspace, OpenAI says it doesn’t train on inputs or outputs by default, including data reached through apps, and the business’s admin manages the account. OpenAI’s managed account article says an admin can suspend or delete a work account, and that a personal account and a work account stay separate even when someone switches between them. When work happens in a personal account, it walks out the door with the employee.
Which plan fits your team is its own question, covered in ChatGPT Plus vs Business.
How long it’s kept, and the exception OpenAI names
OpenAI’s chat retention article sets out the rules:
- Deleted chats disappear from view right away and are scheduled for permanent deletion within 30 days, unless they were already de-identified and separated from the account or OpenAI “must retain it longer for security or legal obligations.” A deleted chat can’t be restored.
- Temporary Chat stays out of history and isn’t used for training while it stays temporary, but OpenAI may keep a copy for up to 30 days for safety.
- Files follow their own clock. A file given to a custom GPT is kept until the GPT is deleted.
- On ChatGPT Business, admins control how long data is kept. The enterprise privacy page says deleted or unsaved conversations leave OpenAI’s systems within 30 days “unless longer retention is required by law, or is reasonably necessary to protect our services or any third party from harm.”
That legal exception has been used. In 2025 a court order in The New York Times’ lawsuit against OpenAI required it to keep consumer ChatGPT and API content indefinitely instead of following its normal deletion schedule. OpenAI’s post on the order listed Free, Plus, Pro and Team subscriptions and the API as affected, and said Enterprise and Edu customers were not. Its October 22, 2025 update says the obligation ended September 26, 2025, and that OpenAI still stores limited user data from April to September 2025 under a legal hold. So treat “deleted within 30 days” as the normal rule, and a court can override it. Anything pasted in could be kept longer.
Who else can see a chat
Chats aren’t public, but three groups can reach them.
- Your workspace admin. On ChatGPT Business, OpenAI says admins can open members’ conversations in the workspace and delete them. Work chats are business records, so that’s useful. Tell your staff up front.
- OpenAI. For personal accounts, authorized OpenAI staff and service providers bound by confidentiality may access content to investigate abuse or security incidents, to help with support, for legal matters, and to improve models unless the person opted out. For ChatGPT Business, OpenAI limits it to authorized employees for engineering support, abuse investigations and legal requirements, plus contractors who review for abuse and misuse.
- Anyone holding a shared link. OpenAI’s shared links article says anyone who has a link created from a personal account can view the shared conversation, and uploaded files and images can appear in it. A link made inside a Business workspace opens only for members of that workspace. Shared conversation links start with chatgpt.com/share/, and they can be deleted from the Shared links list under Data controls.
GPTs, apps and connectors
Some features send information past OpenAI to another company.
- Custom GPTs. OpenAI’s GPTs article says the people who build a GPT can’t view individual conversations with it. A GPT that uses apps or outside services may send relevant parts of what you type to that service, and “OpenAI does not audit or control how those services use or store your data.” OpenAI’s Temporary Chat article adds that data sent this way follows the recipient’s privacy policy and may be kept longer than 30 days.
- Apps and connectors. These link ChatGPT to tools like Google Drive or Slack. OpenAI’s apps article says data shared with an app follows that app’s own terms and privacy policy, and that on ChatGPT Business many apps are on by default until an admin changes it. OpenAI’s business data page also lists a Business control to switch third-party GPTs on or off.
Connecting a tool to your files is a decision for the business to make.
Fake apps, phishing and the login itself
Whoever gets into an account can read every conversation in it.
- Get the real app. ChatGPT runs at chatgpt.com. OpenAI’s download page links its official phone apps, so install from there and check that the developer listed is OpenAI.
- Watch the inbox. OpenAI’s account security article warns about emails that ask for login details and says to double-check the sender’s address and the web address.
- Turn on two-step sign-in. OpenAI’s MFA article offers an authenticator app, a passkey or a text code. It also says MFA “cannot currently be enforced at the ChatGPT workspace” level, so an admin can’t switch it on for everyone. Check each person, or use single sign-on, which OpenAI lists as a Business feature.
The owner’s checklist
- Put the team on a business workspace. Work stays in an account the business controls, training is off by default, you can shut off access when someone leaves, and your admin decides which apps and third-party GPTs are allowed.
- Turn off training on any personal use. If anyone touches work from a personal account, the training setting goes off. Better still, work stays out of personal accounts.
- Write a one-page rule on what never goes in. Customer personal details, health and payment information, passwords, NDA material and employee records. Our guide to getting started has the longer list, and AI training for employees covers making it stick.
- Turn on MFA for every account, and check it person by person.
- A person checks the output. Careful data handling doesn’t make the answers right. Someone reads every reply, number and claim before it goes out.
One more line for anyone handling patient information: OpenAI’s business data page lists a Business Associate Agreement for its ChatGPT for Healthcare product and its API healthcare customers. Sort that out before anyone pastes a thing. How we treat our own clients’ information is on our data handling page.
If you want help drawing these lines for your business, that’s what the call below is for.