Is it safe to put company data into ChatGPT?

Putting company data into ChatGPT is safe on a business plan with clear usage rules, and risky on personal accounts. ChatGPT Business, Enterprise and the API do not train on your data by default, and admins control retention. Personal accounts can train on chats unless each user turns that off. The question "is ChatGPT safe for business" comes down to the plan and the rules, not the model.

Even on a business plan, three kinds of data need more than default settings. They are protected health information, payment card numbers and anything a client contract says you cannot share. A small team can set up the right plan, rules and offboarding in one afternoon.

On this page

• Does ChatGPT train on your company data?
• ChatGPT plans compared on data privacy
• How long OpenAI keeps your chats
• Is ChatGPT safe for business use on a personal account?
• Shadow AI is a bigger risk than the model
• Other ways company data leaks through ChatGPT
• Can you see what employees type into ChatGPT?
• What company data can go into ChatGPT
• HIPAA, client contracts and other rules
• What to do after someone pastes the wrong data
• How to set up ChatGPT safely for a small team
• A one-page AI use policy you can adapt
• What safe ChatGPT costs a 12-person team
• When a private AI tool beats ChatGPT

Does ChatGPT train on your company data?

It depends on the plan, not on what you type. OpenAI's enterprise privacy page, updated January 8, 2026, says business data is not used to train its models by default. That covers ChatGPT Business, Enterprise, Healthcare, Edu and the API. OpenAI uses that data for training only if your organization opts in.

Personal accounts work the other way. Free, Go, Plus and Pro plans list model training as "opt-out available," which means it is on until someone turns it off. OpenAI's data controls FAQ says each user turns it off under Settings, then Data Controls. The switch is called "Improve the model for everyone." Chats then stay in that user's history but are not used for training.

The weak point is that the switch belongs to the employee. A company with 15 people on personal accounts has 15 separate settings and no way to check any of them. On a Business workspace, the no-training default applies to every seat the moment it is created.

OpenAI's promise also carries legal weight in the US. In January 2024, the Federal Trade Commission warned AI companies about broken data promises. Training on customer data after promising not to, or quietly changing terms to allow it, may violate the law. That does not make a promise risk-free, but it gives business terms teeth.

ChatGPT plans compared on data privacy

The table below uses OpenAI's business pricing page and privacy pages as of October 2026. Business prices are per user per month for a Standard seat.

Business also offers a Premium seat at $100 a month billed annually or $125 monthly, and you can mix seat types. OpenAI's business data page says business data is encrypted with AES-256 at rest and TLS 1.2 or higher in transit. It also lists alignment with SOC 2 Type 2 criteria.

For most US small businesses the decision is Business versus personal accounts. Enterprise earns its custom price when you need SCIM provisioning, audit logs fed to security tools, or custom retention rules.

How long OpenAI keeps your chats

Retention is where "safe" gets specific. On ChatGPT Business, admins can control how long data is kept, and deleted or unsaved conversations are removed from OpenAI's systems within 30 days. Enterprise admins set the retention window. API inputs and outputs can be kept for up to 30 days to detect abuse, and qualifying customers can request zero data retention for eligible endpoints.

Personal accounts keep chats in history until the user deletes them. Temporary Chat skips the history, and OpenAI says those chats are deleted within 30 days.

Every one of those timelines carries the same exception: OpenAI keeps data longer when the law requires it. 2025 showed what that means. In the New York Times copyright case, a court order forced OpenAI to preserve consumer ChatGPT and standard API data, including chats users had deleted. OpenAI's response to the data demands says the order covered Free, Plus, Pro and Team accounts and API customers without a zero retention agreement.

ChatGPT Enterprise, Edu and zero retention API customers were excluded. OpenAI says its obligations under that order ended on September 26, 2025, and standard 30-day deletion resumed. Deletion on consumer plans depends on legal events you do not control, while contract tiers drew a firmer line.

Is ChatGPT safe for business use on a personal account?

For anything beyond public information, no. A consumer account fails on control rather than encryption. The company cannot enforce the training setting, cannot see who uses the tool, and cannot remove chat history when an employee leaves. The data walks out with the login.

This is already common. The KPMG and University of Melbourne Trust in AI study was published in April 2025 with 1,019 US respondents. It found 46 percent of US workers have uploaded sensitive company information and intellectual property to public AI platforms. Half of the US workforce uses AI tools without knowing whether it is allowed, and 53 percent avoid disclosing their AI use.

Shared links add another gap. OpenAI's shared links FAQ says anyone who has a shared link can view the linked conversation and pass it on. One forwarded link to a chat that contains a client's financials is a disclosure, whatever the account settings say.

Shadow AI is a bigger risk than the model

Breach data points at unmanaged use, not at OpenAI's servers. IBM's 2025 Cost of a Data Breach report studied 600 organizations breached between March 2024 and February 2025. One in five organizations had a breach linked to shadow AI. Organizations with high levels of shadow AI paid an average of $670,000 more per breach.

The same report found 13 percent of organizations reported breaches of AI models or applications, and 97 percent of those lacked proper AI access controls. Of breached organizations, 63 percent had no AI governance policy or were still writing one. The average US breach cost hit a record $10.22 million.

Banning ChatGPT rarely fixes this. Staff who find it useful switch to their phones. Giving them an approved business workspace with written rules moves the use into a place you can see and manage.

Other ways company data leaks through ChatGPT

Training is the risk people ask about. These are the ones that cause incidents:

• A reused password on a consumer login exposes every chat in its history. Business seats with SSO tie access to your company login.
• When ChatGPT can read your drive, inbox or CRM, a single prompt can pull far more data than anyone pasted in. Limit connectors to the folders a role needs.
• API keys, passwords and database connection strings end up in prompts during debugging. Rotate any secret that has been pasted.
• A model can state an incorrect clause, figure or legal rule with confidence. Anything that goes to a client needs a human check.
• On consumer accounts, access survives the exit interview. On Business, removing the seat removes the access.

Connected tools deserve the closest look in 2026, because assistants now reach into business systems through standards such as the Model Context Protocol. Each connection is a permission grant, so review it the way you would review a new employee's access.

Can you see what employees type into ChatGPT?

On Enterprise, yes, through its compliance logs. OpenAI's business pricing page lists a Compliance API Logs Platform for Enterprise and not for Business. On Business, the admin console manages seats, SSO and settings, but it does not give you that log feed.

The gap matters for firms with record-keeping duties, which may need to produce AI conversations for an audit or a lawsuit. For a 12-person firm without those duties, Business plus a written policy covers the risk. For a regulated firm, price Enterprise before you roll AI out to the whole team, and ask OpenAI sales which fields the logs capture before you sign.

Either way, logs only show what happened. The policy and the plan decide what can happen.

What company data can go into ChatGPT

Sort data by what happens if it leaks, then match it to a plan. The table reflects OpenAI's published defaults as of October 2026.

Two rows trip up small businesses most. Many client NDAs bar sharing with third parties, and an AI vendor is a third party. Credentials are the row people break most, and debugging is where it happens, which is why the leak list above says to rotate them.

HIPAA, client contracts and other rules

This section is general information, not legal advice. Check your obligations with an attorney who knows your industry and state.

Health data has the clearest rule. Under HHS guidance on business associates, a vendor that creates, receives, maintains or transmits protected health information for a covered entity is a business associate. The covered entity needs a written business associate agreement first. OpenAI's business data page says it offers a BAA to ChatGPT for Healthcare and API healthcare customers. A clinic on a standard Business plan without a BAA should keep patient details out.

Client contracts come next. Read the confidentiality clause of your three largest client agreements before you write an AI policy. Some allow subcontractors and processors with equal protections, some require prior written consent, and some ban offshore processing outright.

Your own privacy notice matters too. If it lists the categories of vendors that receive customer data, an AI provider should appear on that list before customer records go into prompts.

What to do after someone pastes the wrong data

Mistakes will happen, so decide the response before the first one. A same-day routine keeps a bad paste from growing into a reportable incident:

1. Delete the conversation. OpenAI says deleting a chat also deletes any shared link created from it.
2. Rotate any password, API key or token that appeared in the chat, even on a business seat.
3. Write down what was pasted, when, and on which account and plan.
4. Check the client contract or NDA for notice duties, and call your attorney if customer personal information was involved.
5. Use the case, without naming the person, in the next team briefing.

On a Business workspace, a deleted chat leaves OpenAI's systems on the 30-day schedule described above. On a consumer account with training switched on, assume the content may already have been used for training.

How to set up ChatGPT safely for a small team

A 10-person company can finish this in one afternoon:

1. Buy ChatGPT Business seats for everyone who uses AI for work, and ask staff to stop using personal accounts for company tasks.
2. Turn on SSO so access follows your company login, and require multi-factor authentication on that login.
3. Set a retention period in the admin console that matches your document retention policy.
4. Review which apps and connectors are enabled, and switch off any that reach drives or inboxes the team does not need.
5. Publish a one-page AI use policy and walk the team through three real examples of allowed and banned prompts.
6. Add AI seat removal to your offboarding checklist, next to email and CRM access.
7. Once a quarter, review active seats, connected apps and any incidents, and update the policy.

Step 1 does most of the work. The other six take the remaining risk from a vague worry to a short list someone owns.

A one-page AI use policy you can adapt

Long policies go unread. These eight lines cover what a small business needs:

• Use only the company ChatGPT Business workspace for work tasks.
• Never paste passwords, API keys, payment card numbers or patient health details.
• Remove client names from documents covered by an NDA unless the account lead approves.
• Do not share chat links outside the company.
• Check every figure, quote and legal statement before it goes to a client.
• Label AI-drafted client deliverables internally so a reviewer knows to check them.
• Ask before connecting ChatGPT to any new app, drive or inbox.
• Report a mistaken paste to the owner the same day, so secrets can be rotated.

Put the policy where people work, such as the top of the shared drive, and repeat the first two lines at every new-hire onboarding.

What safe ChatGPT costs a 12-person team

Say a 12-person accounting firm in Denver moves everyone from personal accounts to ChatGPT Business Standard seats. On annual billing, that is 12 seats at $20 a month, or $240 a month and $2,880 a year. On monthly billing it is 12 at $25, or $300 a month and $3,600 a year.

Setup takes one afternoon of an office manager's time plus a 30-minute team briefing. Compare that with the $670,000 IBM attributes to high shadow AI use in a breach. A firm this size would see a smaller bill than IBM's averages. One exposed client tax file can still cost more in notification, legal review and lost trust than ten years of seats.

Budget for Premium seats only for staff who need the higher limits, since a mix of seat types is allowed.

When a private AI tool beats ChatGPT

ChatGPT Business covers general drafting, research and analysis. A custom tool built on the API makes sense in three cases. The AI must answer from your own documents, you need zero data retention, or the data cannot leave a specific region.

The usual pattern is retrieval: the tool searches your approved documents and sends only the relevant passages to the model. This guide to retrieval-augmented generation explains how that works. If you are still choosing a model provider, the comparison of ChatGPT, Claude and Gemini for business covers the trade-offs. Teams that paste source code into AI tools should also read the risks in vibe coding.

Count how many people at your company used a personal AI account for work last month, then price Business seats for exactly that number. That figure is your cost to move company data into a workspace you control.

The AI development team at Techno Mantra builds private assistants and retrieval tools on OpenAI, Claude or Gemini APIs, and reviews existing setups. You own the code and accounts, and an NDA is available on request. Send a short description of your data and tools through the contact page or to info@technomantraa.com. Every inquiry gets a reply within one business day.