Is it safe to put business data into ChatGPT? A plain-English checklist

It depends on the plan, the settings and the data. As at November 2026, OpenAI says it may use chats on personal Free, Plus and Pro accounts to train its models unless you switch off "Improve the model for everyone", while Business, Enterprise and Edu accounts are not used for training by default.

Australia's privacy regulator, the OAIC, recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available AI tools, and says privacy obligations apply to personal information you put in and get back. The safer pattern is a business account in the company's name, training off, a short list of data that never goes in, and a person checking every output before it is used.

Which ChatGPT plans use your chats to train models?

ChatGPT's personal plans are Free, Go, Plus and Pro. Business, Enterprise and Edu are for organisations, with an admin who runs the workspace. OpenAI's ChatGPT Business rename FAQ says Team was renamed Business on 29 August 2025.

On Free, Plus and Pro, OpenAI says in How your data is used to improve model performance that it may use your chats to train its models. The off switch is in Settings, under Data controls: turn off "Improve the model for everyone". One catch: OpenAI says a thumbs up or thumbs down on a response can still send that whole conversation to training, even with the setting off. OpenAI's article doesn't mention Go, so check Data controls on any Go account too. And the switch belongs to whoever owns the login, so on personal accounts each employee decides.

Business, Enterprise and Edu work the other way. OpenAI's Enterprise privacy page says it does not train its models on business data from these plans by default.

Where does ChatGPT keep your data, and for how long?

Deleting a chat doesn't remove it at once. OpenAI's article on deleting and archiving chats says deleted chats are scheduled for permanent deletion within 30 days, unless OpenAI must keep them longer for security or legal obligations. Its article on temporary chat says temporary chats aren't used to train its models and don't create or update memories, but OpenAI may keep a copy for up to 30 days for safety purposes. Saved memories are managed under Settings, "Personalization", according to its article on Memory.

On a Business plan, OpenAI's Enterprise privacy page says workspace admins can view, access, export and delete members' conversations. Tell staff before they start.

Deletion also has legal exceptions. In 2025, a court order in The New York Times' lawsuit against OpenAI required OpenAI to keep consumer ChatGPT and API content indefinitely. OpenAI says in How we're responding to The New York Times' data demands that those obligations ended on 26 September 2025, but it still holds some April to September 2025 user data under legal hold. It says Enterprise and Edu were not covered. Team, now Business, was. In January 2026, a US district judge upheld an order for OpenAI to hand 20 million de-identified ChatGPT logs to the plaintiffs, the National Law Review reported.

OpenAI's article on data residency and inference residency lists Australia as a region where new Enterprise and Edu customers can choose to have their content stored at rest. Processing is separate. Inference residency, which keeps the model's processing of that content in the region, is offered only for Europe, the United States and the United Arab Emirates, so Australian storage doesn't decide where your prompts are processed. OpenAI's pricing page ticks data residency, with Australia among ten regions, for Enterprise but not for Business.

What does Australian privacy law expect?

The OAIC's Guidance on privacy and the use of commercially available AI products, first published in October 2024, says: "As a matter of best practice, the OAIC recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools". The OAIC's page What is personal information? lists health or genetic information, racial or ethnic origin and criminal record among the sensitive categories.

Three Australian Privacy Principles (APPs) and the breach scheme are the ones to know:

• APP 6 sets out when a business may use or disclose personal information it holds.
• APP 8 sets out the steps a business must take before personal information is disclosed overseas. The OAIC's APP 8 guidelines say the business can be accountable for what the overseas recipient does with it.
• APP 11 requires reasonable steps to protect personal information from misuse, interference, loss and unauthorised access.
• Under the Notifiable Data Breaches scheme, a breach likely to cause serious harm must be reported to the people affected and to the OAIC. If you suspect one, you must take all reasonable steps to finish an assessment within 30 calendar days.

Whether putting information into an AI tool counts as a "use" or a "disclosure" changes which rules apply. The APP 8 guidelines say giving personal information to an overseas cloud provider only to store it may be a use, if conditions such as a binding contract and effective control are met. Whether a particular AI tool meets them is a question for your lawyer.

Most businesses with annual turnover of $3 million or less are exempt from the Privacy Act, though the OAIC's small business page lists 12 kinds covered anyway, including health service providers. Tax file numbers are the catch for small employers. The Privacy (Tax File Number) Rule 2015 applies to any person or organisation holding tax file number information, and the breach scheme covers TFN holders when TFN information is involved in a breach.

A separate rule on automated decisions starts on 10 December 2026.

What should never go into a personal ChatGPT account?

ASD's small business guide recommends an internal AI use policy that clearly defines what data can't be uploaded. A starting list for personal accounts:

• Tax file numbers.
• Bank account and card numbers.
• Health information, including medical certificates.
• Passwords and login codes.
• Client documents covered by confidentiality terms or legal privilege.
• Pay records and disciplinary records.
• Driver licences, passports and Medicare cards.

The guide describes a case from early 2025: a contractor for an Australian organisation uploaded names, contact details and health records of people in a government program into an AI system. The guide says it led to a serious data spill and is considered a notifiable data breach. It recommends removing, anonymising or changing personal details so they can't identify anyone.

On a business account, decide each item with advice and write it down.

Free and Go accounts can also show ads: OpenAI's ChatGPT release notes say it began rolling them out to those users in Australia on 16 April 2026. Its Ads in ChatGPT article says personalised ads can draw on past chats and memory, and that advertisers don't have access to your chats.

Is it safe to connect ChatGPT to email, files and calendar?

It can be, once you know what each app can reach and do. Apps link ChatGPT to tools such as Gmail, Google Drive and calendars. OpenAI's Google app data controls FAQ says ChatGPT can access the content available to the Google account you connect, so connect a login that sees only what the task needs. Some apps can send or change things as well as read them. The same FAQ lists Google's gmail.modify permission for the Gmail app, which Google's Gmail API scopes page describes as "Read, compose, and send emails from your Gmail account".

OpenAI's article on admin controls for apps says apps are enabled by default in Business. In Enterprise and Edu, new workspaces start with a selected set of apps on, and new apps are generally off until an admin turns them on. Where an app offers action controls, admins can allow its read actions and leave its write actions off.

Start read-only, with one team and one process. For accounting software, What AI can do with Xero and MYOB right now covers what each connector can and can't do.

Five questions to ask before any AI tool sees client data

These work for ChatGPT, Claude, Gemini and Copilot alike. Write the answers down.

1. Is it a business account in your company's name, or a personal login?
2. Does it train on what your team puts in, and is that switched off?
3. Where is your data stored and processed, for how long, and who can read it?
4. What can it connect to, and can it only read, or also send, share and delete?
5. What must never go in, and who checks what comes out before it is used?

The questions line up with Artificial intelligence for small business, published on 14 January 2026 by ASD's Australian Cyber Security Centre with New Zealand's NCSC and COSBOA. Its checklist asks whether you know what business information can be safely shared, where the tool stores data, and whether your data will be used to train AI models. It also says to "have a human involved in the decision-making process especially in high-stakes or sensitive operations".

Copilot, ChatGPT or Claude: which suits an Australian small business? compares the three on price, training and storage.

Where a process consultant would start

Start with the process, then pick the tool. List the steps in your business that touch tax file numbers, bank details or health information, such as new starter paperwork, supplier bank detail changes and sick leave backed by medical certificates. Those steps stay out of AI tools until each has a written rule.

Draw the same line for every tool. AI tools prepare: they draft, match, sort and flag. People send, approve and pay. A tool can sort a supplier inbox and flag an email asking to change bank details. A person checks that request and approves any change.

AI consultant, in-house hire or ongoing support compares who could do this work, and Why most Australian businesses aren't getting value from AI yet looks at what holds businesses back.