What Are the Risks When Employees Paste Company Data into ChatGPT, and How Should a Company Manage It?
The risk has two layers. Once data is pasted, it sits on the provider's servers (even if it is not used for training, it is retained for some time). And when employees use personal accounts, the company cannot control whether their chats are used for training. The answer is not a blanket ban but three steps: name the approved tools and company accounts, classify data and state which level may go where, and give confidential data a tool that does not send it outside the company.
1. Where does pasted data go?
When an employee presses send, the text goes to the provider's servers. Whether it is then stored or used for training depends on the plan. Official documentation as of 2026-10-19:
| Usage | Used for model training? | Retention |
|---|---|---|
| ChatGPT personal accounts (Free, Go, Plus, Pro) | The user turns "Improve the model for everyone" on or off in Data controls; when off, new chats are not used for training | Chats stay in history; Temporary Chats may be retained up to 30 days for safety |
| ChatGPT Business, Enterprise and other managed workspaces | Not used for training by default, per OpenAI | Follows the workspace's retention settings |
| OpenAI API | Since March 1, 2023, API data is not used for training unless you opt in | Abuse monitoring logs retained up to 30 days by default |
| Anthropic (Claude) commercial products and API | Not used for training by default | API inputs and outputs deleted within 30 days by default, with exceptions |
The point: "Is it used for training?" and "Is it retained?" are separate questions. Even on a no-training plan, the data has left the company and stayed on someone else's servers for a while. For truly confidential material, the test is not whether training is off but whether the data may leave the company at all.
2. Legal points in Taiwan
If the pasted text contains personal data of customers or staff, Taiwan's Personal Data Protection Act may come into play:
- Article 5: collection, processing, and use of personal data must not exceed what is necessary for the specific purpose.
- Article 20: non-government entities must use personal data within the necessary scope of the specific purpose of collection.
- Article 4: a party commissioned by a government or non-government entity to collect, process, or use personal data is treated as the commissioning party within the scope of the Act. Under Article 8 of the Act's Enforcement Rules, a party that commissions others to collect, process, or use personal data must properly supervise the commissioned party (for example, by agreeing on the scope of processing, security measures, and the return or deletion of data when the engagement ends, and by checking compliance regularly). In other words, if handing personal data to an outside service counts as commissioning under the Act, the company has a duty to supervise, and its responsibility does not pass to the provider just because the data was handed over. Whether pasting data into a particular AI service counts as commissioning depends on the facts, such as the service terms, the account type, and how the data is used.
For government agencies, the Executive Yuan presented the "Reference Guidelines for the Use of Generative AI by the Executive Yuan and Its Agencies" at its cabinet meeting on August 31, 2023. One aim is for agencies to maintain the confidentiality and professionalism of official business, and ministries are asked to set rules or internal controls for generative AI as their work requires. Private companies are not bound by it, but the direction is useful: decide which data must not go into generative AI before deciding how to use it.
This is a general explanation of the law, not legal advice for any specific situation; for an actual case, consult a lawyer.
3. Three steps a company can take now
1. Name the tools and accounts. List approved AI tools and use company-managed accounts or workspaces. Personal accounts leave training settings, chat history, and post-resignation data outside the company's control.
2. Classify data and post the table internally. A sample to adapt:
| Level | Examples | Where it may go |
|---|---|---|
| Public | Website copy, published catalogs | Any approved AI tool |
| Internal | General SOPs, internal notices | Company workspace or API only, with no-training settings confirmed |
| Confidential | Quotes, contracts, formulas, drawings, financials | Not sent to outside AI services; use a local model on company hardware |
| Personal data | Customer names and phone numbers, staff records, medical records | Generally not pasted; de-identify first and confirm the use fits the collection purpose |
3. Give confidential data a usable alternative. A rule that only says "don't" pushes people to workarounds. A tool that keeps data inside the company, such as a local model on company hardware, makes the rule hold.
Finally, explain the why. Most leaks are not malicious; employees simply do not know where the text goes.
How Think BIG handles "can this data leave the company?"
Think BIG (Think BIG Make it Real Co., Ltd.) is an enterprise AI Agent deployment provider headquartered in Taiwan. Starting from protecting company data, we build AI Agents inside the client's own environment (company hardware, or a cloud server under the company's account). There are two deployment options: with the on-premises version, models and documents are processed on company hardware; with the cloud-assisted version, employees' questions and the document passages the system retrieves are sent to the model provider for processing.
- Map the data flow before building: which data enters the AI, where it is stored, and whether anything leaves the company, listed item by item for the client to confirm.
- Decide which data may leave first: before building, we use the company's data classification to confirm which data must stay in the on-premises version and which may be processed with the cloud-assisted version; the cloud-assisted version is subject to the model provider's terms.
- Pre-delivery security review: before each system is delivered, its code and configuration go through a security review; if it does not pass, it is not delivered. No one can promise perfect security; what we do is review, one gate at a time.
See Enterprise and Trust.
The actual scope of Think BIG's services follows what both parties confirm.
FAQ
- Q1: If employees use ChatGPT for company data, will it be used for training?
- It depends on the plan. On ChatGPT personal accounts, the user decides whether "Improve the model for everyone" is on. ChatGPT Business, Enterprise and other managed workspaces are not used for training by default, per OpenAI. OpenAI API data has not been used for training by default since March 1, 2023.
- Q2: If training is off, is the data safe?
- Not entirely. No training is not the same as no retention: OpenAI API abuse monitoring logs are kept up to 30 days by default, and ChatGPT Temporary Chats may be kept up to 30 days. For truly confidential data, ask whether it may leave the company at all.
- Q3: Should a company ban ChatGPT outright?
- A ban alone tends not to hold. It works better to name approved tools and accounts, classify data into public, internal, confidential, and personal data with clear rules for each, and give confidential data an alternative that stays inside the company.
- Q4: Does pasting customer data into AI violate Taiwan's Personal Data Protection Act?
- It can, depending on the data and the purpose. Article 5 says collection, processing, and use of personal data must not exceed what the specific purpose requires, and Article 20 says non-government entities must use personal data within the necessary scope of the specific purpose of collection. If pasting data into AI serves a purpose beyond the one it was collected for, that may be a problem. Confirm the use fits the collection purpose, and de-identify where possible.
- OpenAI Help Center, Data controls in ChatGPT ("Improve the model for everyone", Temporary Chat up to 30 days, managed workspaces not used for training by default): help.openai.com
- OpenAI API, Your data (API data not used for training since 2023-03-01; abuse monitoring logs up to 30 days): platform.openai.com
- Anthropic Privacy Center, Is my data used for model training? (commercial products not used for training by default): privacy.claude.com
- Anthropic Privacy Center, How long do you store my organization's data? (API deletion within 30 days and exceptions): privacy.claude.com
- Personal Data Protection Act (Articles 4, 5, 20), Laws & Regulations Database of the ROC (Taiwan): law.moj.gov.tw
- Enforcement Rules of the Personal Data Protection Act (Article 8, supervision of commissioned parties): law.moj.gov.tw
- Executive Yuan, Reference Guidelines for the Use of Generative AI by the Executive Yuan and Its Agencies (cabinet meeting, 2023-08-31): www.ey.gov.tw
This article summarizes public official documents, laws, and security standards to help companies evaluate AI adoption. It is not legal advice. For Taiwan law, the latest text on the Laws & Regulations Database prevails; external documents, prices, and specifications follow their original sources. The actual scope of Think BIG’s services depends on what both parties confirm.