Is it okay to put work data into ChatGPT? Information leak risks and safety settings for each AI service that you should know about for business use [2026 edition]

``Let's paste this Excel data into ChatGPT and have it analyzed.'' ``I'd like to have AI summarize the minutes, but the content is confidential...'' I think there are quite a lot of people who use AI in their work with such ambivalence.

As of March 2026, generative AI such as ChatGPT, Claude, and Gemini are reliable tools that greatly increase work productivity. However, the worry that ``the data you input may be used for AI learning'' is a real worry that many business people feel.

In fact, in 2023, Samsung employees entered confidential semiconductor source code and internal meeting recordings into ChatGPT, resulting in three incidents of leakage of confidential information, which became big news.

In this article, we will provide an easy-to-understand explanation of the three major AI services, ChatGPT, Claude, and Gemini, by plan. Settings can be completed in less than 5 minutes, so if you use AI at work, please check it out.

Why is "learning from input data" dangerous in the first place?

Generative AI such as ChatGPT may use the set of sentences (prompts) entered by the user and the AI's answers for future model improvement (learning). Roughly speaking, this means that the content you input may be captured in the AI's head.

This becomes a problem in the following cases:

Company confidential information: source code, blueprints, unpublished product information, etc.
Customer personal information: name, address, phone number, transaction history, etc.
Non-public information: sales data, personnel evaluations, meeting minutes, etc.

Although the data used for learning is not usually directly displayed in other users' answers, the truth is that it is "not zero risk". As in the case of Samsung, content that is casually copied and pasted within the company is sent to an external server and stored as learning data - this is the true nature of "information leakage via AI."

In other words, configuring settings that are not used for learning is a major prerequisite for business use.

ChatGPT learning opt-out settings (by plan)

According to OpenAI's official policy, as of March 2026, data handling differs greatly depending on the plan.

Free/Plus/Pro plan (personal use)

By default, the input data is used for learning. To opt out (reject learning), change your settings by following the steps below.

Log in to ChatGPT
Click your profile icon in the top right → Settings
Select Data Control
Toggle OffImprove the model for everyone

This is all you need. The content you enter after setting will not be used for learning. However, the information entered before the settings has already been sent, so it is not included.

Business/Enterprise plan (corporate use)

For corporate ChatGPT Business ($30/person/month) and Enterprise plans, input data is not used for training by default. There is no need to set it individually, and the data processing agreement (DPA) at the time of contract clearly states that it will not be used for learning.

If you want to implement ChatGPT in earnest as a company, it is safe to choose the Business or Enterprise plan.

API usage

If you are using OpenAI's API to integrate it into your system, data sent to the API will not be used for training by default. If you have incorporated ChatGPT functionality into your in-house tools via API, this is a good thing.

Claude/Gemini learning opt-out settings

More and more people are using not only ChatGPT but also Claude and Gemini for work. Let's check each setting method.

Claude (Anthropic)

According to Anthropic's official policy, input data may be used for training by default in Claude's free version and Pro plan.

Team/Enterprise plan: Not used for learning by default. Administrators can control it all at once through organization settings
API usage: Not used for learning by default (storage for safety monitoring only for 30 days)
Free/Pro plan: Learning usage can be turned off from the "Privacy" section of the settings screen

Gemini (Google)

Gemini is handled differently for personal Google accounts and corporate Google Workspace.

Personal account: Gemini app activity is turned on by default and input data may be used for learning
Google Workspace (Gemini Business/Enterprise):Input data is guaranteed not to be used for learning

If you use Gemini for business with a personal account, follow the steps below to turn off the activity.

Visit myactivity.google.com
Click “Gemini App Activity”
Toggle off
Select "Pause" in the dialog that appears

Is setting alone not enough? 5 rules to follow when using for business

Even if you set up an opt-out, "No learning = absolutely safe" Since data is sent to a cloud server, there is no risk of zero risk. You can further reduce your risk by following these five rules.

Rule 1: Mask sensitive information before entering it

Anonymize proper nouns and specific numbers, such as replacing the customer name with "Company A" and the amount with "X million yen" before passing it to the AI. This alone will greatly reduce your risk.

Rule 2: Avoid copying and pasting the entire source code

This is the lesson from the Samsung incident. If you want to know the cause of the error, extract only the relevant part and rewrite project-specific variable names and file paths before entering.

Rule 3: Never enter personal information

As a general rule, do not enter personal information such as name, address, telephone number, email address, My Number, etc., regardless of whether you have opted out or not. From the perspective of the Personal Information Protection Act, you should carefully consider sending information to third-party servers.

Rule 4: Consider introducing a corporate plan

Opt-out settings for individual plans are based on self-reporting. There is no guarantee that all employees will set the settings correctly. With a corporate plan, administrators can turn off learning usage all at once, eliminating the risk of forgetting settings.

Rule 5: Create internal guidelines

Let's clearly state within the company how much input is acceptable to the AI. It would be helpful to have a classification like the one below, even if it's just a rough one.

OK: General questions, summary of public information, correction of email text
Conditional OK: Internal data (after masking), part of code
NG: Customer personal information, non-public management data, contract documents

Comparing safety levels by plan

We have compiled a list of data learning policies for each plan for the three main services. Information as of March 2026.

Service	Plan	Used for learning?	How to opt out
ChatGPT	Free/Plus/Pro	Default YES	Off in settings screen
ChatGPT	Business/Enterprise	NO	Not required (off from the beginning)
ChatGPT	API	NO	Not required
Claude	Free/Pro	Default YES	Off in settings screen
Claude	Team/Enterprise	NO	Not required (admin controlled)
Claude	API	NO	Not required
Gemini	Personal account	Default YES	Turn off activity
Gemini	Workspace	NO	Not required (guaranteed by contract)

In short, If you use a corporate plan or API, it will not be used for learning from the beginning. For individual plans, manual opt-out settings are required.

FAQ

Does the performance of ChatGPT deteriorate if I set the opt-out setting?

No, it won't fall. Opting out is just a setting to ``not use your input data for future learning'' and does not affect the current model performance or available functions. You can still save and view your chat history.

Is it okay to use ChatGPT's free plan for business purposes?

You can prevent learning use by setting the opt-out setting, but the free plan has restrictions in terms of use and functional limitations. If you plan to use it seriously for business, we recommend purchasing a corporate plan (Business or higher), which includes audit logs and administrator functions.

Can I delete confidential data that I have already entered?

Although it is possible to delete chat history, it is not possible to confirm whether data already sent to OpenAI's server has been used for learning. OpenAI Help Center explains that data will be retained for up to 30 days for safety monitoring purposes even after history deletion. It is most important to pay attention to future inputs.

Which plan should freelancers and sole proprietors choose?

When handling client data, we recommend using a paid plan (ChatGPT Plus or Claude Pro) with at least an opt-out setting. If you have the budget, consider using APIs or corporate plans.