- Free ChatGPT users are opted into training data use by default — their conversations can improve OpenAI's models
- ChatGPT Plus users can opt out; ChatGPT Team and Enterprise exclude conversations from training by default
- The OpenAI API has stricter defaults: inputs and outputs are not used for training without explicit opt-in
- Anything typed into chatgpt.com — including client data, source code, or internal reports — may be retained and reviewed by OpenAI staff
- For organizations, employee use of the consumer website is a GDPR and EU AI Act exposure — not the API
Yes — by default. When employees use chatgpt.com on a free account, their conversations can be used to train OpenAI's models and reviewed by staff. This default can be changed per user, but most employees haven't changed it. Business plans (Team, Enterprise) and the API have stricter defaults, but most employees aren't using those.
What data does ChatGPT collect?
When you use ChatGPT, OpenAI collects several categories of information. According to their privacy policy, this includes:
- Conversation content — every message you send and every response ChatGPT returns
- Account information — email address, name, and payment details if applicable
- Usage data — which features you use, how often, browser and device information
- Feedback signals — thumbs up/down ratings, regeneration requests, which responses you copy
The conversation content is the sensitive part. It’s not metadata or anonymized usage statistics — it’s the literal text you type, including any proprietary information, client details, or confidential business context you include in your prompts.
Does ChatGPT train on your conversations?
It depends on which product you’re using and whether you’ve changed your settings.
For free accounts, the answer is yes by default. OpenAI uses conversations to train and improve its models unless the user has explicitly opted out. The opt-out is available in Settings → Data Controls → “Improve the model for everyone” — but this setting is off by default for free users, meaning training is on.
This matters because training data use means your conversations may be reviewed by OpenAI trainers, incorporated into future model versions, or used to evaluate model quality. OpenAI says they take steps to remove personal information before using data for training, but this process is not guaranteed or verifiable from the outside.
Samsung’s ChatGPT leak (2023): Engineers pasted proprietary semiconductor code and internal meeting notes into ChatGPT to debug and summarize. The data was ingested by OpenAI’s systems. Samsung subsequently banned ChatGPT for internal use. The incident was a landmark case in corporate AI data exposure — and the tool they used was the same free website millions of employees still use today.
How privacy defaults differ by plan
OpenAI offers meaningfully different privacy guarantees depending on which product you’re using. Understanding the difference is important for any organization where employees use ChatGPT.
| Product | Used for training? | Conversation retention | DPA available? |
|---|---|---|---|
| ChatGPT Free | Yes, by default | Stored, 30-day safety hold after deletion | No |
| ChatGPT Plus | Yes, opt-out available | Stored, 30-day safety hold after deletion | No |
| ChatGPT Team | No, by default | Stored, admin controls available | Yes |
| ChatGPT Enterprise | No, by default | Zero data retention option available | Yes |
| OpenAI API | No, by default | 30 days by default, configurable | Yes |
The gap between ChatGPT Free/Plus and the business products is significant. The consumer website has no Data Processing Agreement, no organizational controls, and no admin oversight. That’s the version most employees access when they open a browser tab and go to chatgpt.com.
What this means when employees use ChatGPT for work
The practical risk for organizations is not that OpenAI is malicious — it’s that employees on personal or free accounts are using a consumer product with consumer-grade privacy defaults to handle business data.
Three scenarios create the most exposure:
- Client and customer data. Pasting client names, emails, project details, or personal information into ChatGPT may constitute a GDPR data breach — sending personal data to a third party without a legal basis or DPA.
- Confidential business information. Product roadmaps, financial projections, M&A discussions, or HR matters typed into a free account can be retained and reviewed by OpenAI staff.
- Source code and IP. Developers using ChatGPT to debug proprietary code on the consumer site are sharing that code with OpenAI under terms that allow retention and review.
The issue isn’t that employees are doing something obviously wrong — ChatGPT is extremely useful. The issue is that they’re using a consumer tool with consumer terms for tasks that require business-grade privacy protections.
EU AI Act obligation: Under the EU AI Act, organizations are considered deployers of AI systems used within their operations — including tools employees adopt individually. If employees use ChatGPT without organizational oversight, this constitutes unmanaged AI deployment, which may trigger transparency and documentation requirements from August 2026.
The API vs. the website: why they’re not the same
A common misunderstanding: because OpenAI’s API doesn’t train on data by default, some assume the same is true for chatgpt.com. It isn’t.
The OpenAI API is a developer product used to build applications. It has different terms: inputs and outputs are not used for training by default, customers can sign a Data Processing Agreement, and data retention can be configured. When a company builds an internal tool on the API, they have much stronger privacy guarantees.
Chatgpt.com is a consumer product. It has different — and looser — defaults. When an employee opens a browser and visits chatgpt.com, they are not using the API. They are using a consumer product with consumer defaults, regardless of what their organization has negotiated with OpenAI.
This distinction matters enormously for risk management. An IT team saying “we have an API agreement with OpenAI” does not protect against the risk of employees using the consumer website independently.
What organizations should do
There are three concrete actions organizations can take to manage ChatGPT data risk:
- Know who is using what. Before you can manage ChatGPT usage, you need to know it’s happening. Browser-level monitoring tools like VetoShield show which AI tools employees use — including whether they’re accessing chatgpt.com — without reading prompt content.
- Define an acceptable use policy. Clearly specify which tiers of ChatGPT are approved (e.g., only Team/Enterprise accounts through the organization’s billing), which data categories may not be entered into any AI tool, and what employees should do instead.
- Move approved use to business plans. If ChatGPT is valuable to your team — and it likely is — the answer isn’t to ban it. It’s to provision a Team or Enterprise account, sign the DPA, and ensure employees use the approved channel rather than personal accounts.
The goal is to keep getting the productivity benefit while eliminating the uncontrolled data exposure that comes from free-tier, unsanctioned use.