How to Safely Upload Company Data to AI (Checklist)

Before you upload company data to AI, check both the material and the account receiving it. The same document can have different retention, access and training settings in different workspaces.

This checklist uses OpenAI as a worked example. It helps identify questions to resolve before uploading; it does not replace your organization’s data policy or an applicable agreement.

Sort the data before you sort the tool

Ask one question about the input: if this ended up somewhere you did not choose, would anyone need to be told?

That is a useful first screen, not a complete classification rule. Published information may be suitable for an approved assistant; an unpublished blog draft can still contain an embargoed launch or confidential strategy. Customer records, employment information and vulnerability details need closer review. Follow your organization’s classification and sharing rules even when the material contains no names.

Keep sensitive data out until its use is approved

My default for a general-purpose assistant is to exclude:

• Credentials: passwords, API keys, tokens and secrets inside connection strings. Use placeholders; connect services through an approved credential mechanism.
• Government identifiers and payment details: omit these unless an approved workflow requires them and its protections have been reviewed.
• Health records and other regulated information: use only an approved service and workflow with the necessary agreements and controls. A provider offering a HIPAA business associate agreement does not make every product or account eligible.
• Client or third-party confidential material: check whether the agreement permits this provider and this use. Permission to hold information is not automatically permission to upload it elsewhere.

Removing an identifier or switching off training does not by itself authorize disclosure.

Account first, prompt second

Account choice matters. OpenAI says business data is excluded from model training by default unless the customer opts in. Consumer accounts offer a training control, but that is separate from retention and access.

OpenAI’s enterprise privacy page says ChatGPT Business workspace admins can access, export and delete workspace conversations. For Enterprise and Edu, the Compliance Platform can expose conversation content to authorized integrations; only a workspace owner can grant its Conversation messages permission. The published policy does not establish the exact browsing interface or controls enabled in your employer’s workspace. Ask your administrator what is collected and who can access it.

If you want the plan-by-plan comparison of what each vendor does with your text, ChatGPT vs Gemini vs Claude covers it properly. The point here is narrower: decide the account before you decide the prompt.

How to strip company data before you paste it

For the large middle category, sensitive but useful, the fix is boring and effective: send the shape of the problem, not the record.

For a generic reply draft, try describing the issue without the real customer name. “Draft a response to a customer who was charged twice and has emailed three times” can be enough without the account number. Replace identifiers with placeholders. Keep the structure, drop the specifics.

I do this with numbers rather than names. Real figures get rounded or shifted before they go in, and the answers stay useful, because what I am asking for is a direction rather than a computed result. Whatever comes back gets checked against the real figures before I act on it, which is my job either way. So the precision I would have been handing over was never doing any work.

Data minimization helps even without specialist tooling, but it does not guarantee anonymity or permission to share. Context can identify a person, and a trade secret can survive removal of names.

Technical controls can complement that habit. Microsoft documents Endpoint DLP policies that warn on or block sensitive pastes into third-party AI sites on onboarded Windows devices. Coverage depends on supported browsers, configuration and licensing; an unconfigured browser does not gain that protection merely because the organization uses Microsoft products.

Upload company data to AI when you have no budget for tooling

A small team can start with three administrative steps. Some approved products or technical controls may still require a budget:

1. Write down what may not be pasted. Not a policy document, a short specific list, closer to the one above than to a compliance framework. “Do not paste client PII, credentials, or unsigned contracts into any AI tool” gives someone a decision they can make in a second.
2. Name one approved tool and make it the easy one. An approved option can reduce incentives to use personal accounts. A ban still needs enforcement and a clear route for permitted work. Shadow AI at work covers what that costs when it goes wrong.
3. Turn off training on the accounts you already use. In ChatGPT the setting is called “Improve the model for everyone” and sits under Settings, then Data controls. When available, it excludes new conversations from training but does not delete them. Voluntarily submitted feedback can make its associated conversation eligible for training, so do not submit sensitive conversations as feedback.

My captured Data controls screen. Training preferences and retention are separate controls.

What to do if you already pasted company data into AI

First identify the material, account, time and any connected services involved. Follow your organization’s incident-reporting process promptly. Revoke or rotate exposed credentials through a trusted channel; preserve enough evidence to understand what happened.

Coordinate cleanup with the responsible person if an investigation or retention obligation applies. Deleting a chat removes it from the account view immediately and schedules deletion from OpenAI systems within 30 days, with exceptions for already de-identified content and security or legal obligations. Deletion does not recover information already shared onward.

My captured bulk-delete dialog. Preserve relevant evidence before clearing chats.

Check files and memory separately. A file saved in Library can remain after its chat is deleted. Review Library, project or GPT files and shared copies as applicable. Saved memories can also survive chat deletion; review the available Memory controls and the underlying sources. These are separate cleanup steps, not a guarantee that every retained copy disappears immediately.

Review training settings for future use. Assess the incident according to the actual sensitivity, recipients, account controls and obligations. Neither a single paste nor a training opt-out tells you whether an exposure is minor.

FAQ

Sources

• OpenAI: How we’re responding to The New York Times’ data demands (including the October 22, 2025 update)
• OpenAI: Enterprise privacy, retention and admin visibility by product
• OpenAI Help Center: Chat and file retention policies in ChatGPT
• OpenAI Help Center: Compliance Platform for Enterprise and Edu, including the Conversation messages permission

• Microsoft Learn: using Purview to warn on or block sensitive pastes into third-party AI apps

• OpenAI: Memory controls and deletion

• OpenAI: Data controls and feedback

Product details checked against OpenAI and Microsoft documentation on October 8, 2026. Other providers and workspace configurations differ.