How to Create Internal Rules for Generative AI | Usage Guideline Template and Practical Examples
Key takeaway: When creating internal rules for generative AI, first define the permitted environments, use cases, input information, reviewers, and points of contact so that employees can decide within one minute whether a use is allowed. Start with tasks that use public information, rather than covering all business activities from the outset, and revise the rules based on questions and incidents.
This article explains how to create internal rules for generative AI using an input information decision table and a usage guideline template. The template is for internal review. Check it against your company's information management policies, customer contracts, and the terms of the products you use, and obtain approval from the person responsible before applying it.
What Should You Decide First When Creating Internal Rules for Generative AI?
Start by summarizing the following four items on one page. When deciding the scope of use based on risk, consult the Digital Agency's page listing the latest Guidelines for the Procurement and Use of Generative AI (DS-920). Also review IPA's security guidance for AI users for information input and output verification. Reading these resources alone does not complete your company's approval process or contract review.
Digital Agency's page listing the latest Guidelines for the Procurement and Use of Generative AI (DS-920)https://www.digital.go.jp/resources/standard_guidelines
IPA's security guidance for AI usershttps://www.ipa.go.jp/digital/ai/security/ai_security_tips.html
| Item to define | Example entry | What can happen if it remains unclear |
|---|---|---|
| Permitted environment | Only company-approved organizational accounts | Business information becomes scattered across personal accounts |
| Covered tasks | Summarizing public materials and drafting emails | Employees may leave even final decisions to AI |
| Input categories | Start with public and fictional information | Individuals are left to decide whether to include customer information |
| Final review | The person sending an external document checks it against the original | Incorrect figures or terms are shared externally |
Avoid a rule that says, "Anything can be entered if you use a paid version." Use for model training, storage, sharing with third parties, and administrative controls each need to be checked separately.
How Do You Decide Which Information Can Be Entered?
Before entering information, use the following table to check when you should hold off. For personal information, refer to the Personal Information Protection Commission's advisory on the use of generative AI services, and have your company's person responsible for personal information protection check the purpose of use and the service's data handling terms.
Personal Information Protection Commission's advisory on the use of generative AI serviceshttps://www.ppc.go.jp/news/careful_information/230602_AI_utilize_alert/
| Example information or task | Initial decision | What the person responsible should check |
|---|---|---|
| Summarizing a published company brochure | A trial is permitted in an approved environment | Purpose of use, reproduction terms, and factual verification of the output |
| Drafting an email from fictional sales meeting notes | A trial is permitted in an approved environment | Whether the content could allow someone to identify a real customer |
| Unpublished materials containing customer names and contract amounts | Hold off before entering | Contractual restrictions, administrator approval, and product settings |
| Personal information such as names and contact details | Hold off before entering | Purpose of use, laws and internal policies, and how the service handles the information |
| Passwords and private keys | Do not enter | If entered by mistake, report it to the administrator immediately |
Explicitly state in the rules: If you are unsure, do not enter the information; ask for guidance. Even with details redacted, a combination of project information and amounts may identify the other party. For more detailed examples, see Information You Should Not Enter into Generative AI.
In What Order Should You Create Internal Guidelines?
| Step | Task | Completion criteria |
|---|---|---|
| 1. Narrow the scope | Define the departments covered, approved products and plans, and permitted tasks | Employees can explain when use is allowed |
| 2. Classify information | Set input conditions for public, internal, personal, and confidential information | There is a point of contact for materials whose status is unclear |
| 3. Assign reviewers | Decide who checks output facts, approves external sending, and approves exception requests | The person making the final decision is explicitly identified |
| 4. Run a small trial | Test with public information or fictional data and record inquiries | Users' questions and operational mistakes can be collected |
| 5. Revise | Regularly check for changes in product settings, contracts, and business activities | The version, approval date, and next review date are recorded |
What Internal Rules Template Can You Use for Review?
The following template includes fields to fill in. Do not distribute it as an official company policy until the fields have been completed and the person responsible has approved it.
- Purpose and scope: These rules apply when [covered departments and employment types] use generative AI for [covered tasks].
- Approved environments: Permitted products, plans, and accounts are recorded in [management register], and [person responsible] decides whether to approve additions.
- Permitted inputs: Use public information and materials for which permission to use has been verified as the default. Customer information, personal information, and unpublished information must go through [review procedure]. Do not enter passwords or private keys.
- Output verification: Check figures, proper names, citation sources, and conditions against the original materials. The relevant department makes decisions on matters such as contracts and accounting.
- External sending: Send external materials only after review by [approver]. Automated sending and updates to production data require separate approval.
- Storage and sharing: Record the storage locations and retention periods for input materials, generated content, and activity logs in [management register].
- Incident response: If you discover an incorrect input or a message sent in error, stop using the service and report the date and time, the environment involved, and the type of information to [contact]. Do not decide on your own that no harm has occurred.
- Review: At [review timing], the administrator checks usage and changes to product terms and communicates revisions.
Next step: Download the editable internal rules template and complete the fields with the relevant departments. Before distributing it as an official generative AI usage policy, record the responsible person's approval date and the next review date.
What Should You Include in an Exception Request?
For example, if you want to summarize a confidential proposal, include the owner of the materials, contractual restrictions, the environment you will use, the purpose of use, and the retention period in your request. Keep the decision on whether the materials may be used separate from the procedure for issuing an account. Even after names have been removed, combinations of project names and transaction amounts may allow identification, so it is also important not to grant blanket permission based solely on anonymization.
How Do You Establish the Rules After Distributing Them Internally?
At an awareness session, ask participants to decide whether information can be entered using three examples: a published company brochure, an unpublished customer contract, and fictional sales meeting notes. Aim for participants to be able to explain whom to consult when unsure, rather than simply memorizing the correct answers.
Add guidance on frequently asked questions the following month. Tracking the plan, account type, and verification date alongside the product name can reduce oversights caused by changes in terms.
When sharing AI-generated text or images externally, check their relationship to third-party copyrighted works as well as comparing them with the original materials. You can refer to the Agency for Cultural Affairs' "AI and Copyright" to inform your assessment. Do not leave source verification or approval for external publication solely to AI.
Agency for Cultural Affairs' "AI and Copyright"https://www.bunka.go.jp/seisaku/chosakuken/aiandcopyright.html
Frequently Asked Questions
Should We Start with a Complete Ban?
There is no single answer for every company. One option is to clearly define a scope that fits your company's circumstances and provide tasks that employees can try using public information or fictional data. Show examples of permitted use alongside prohibited activities.
Does This Template Alone Address All Legal and Contractual Requirements?
No. The necessary measures vary by information type, industry, and customer contract. Combine reviews by legal and information management staff with checks of the actual settings.
Is There a Standard Template for Internal Generative AI Rules?
A single template cannot be applied unchanged to every company. Fill in the approved environments, permitted input information, output reviewers, and incident contacts based on your company's circumstances, then run a trial and revise the rules.
Before Consulting ZETTAi
Bringing your current internal policies, the types of information you handle, and the operations that require approval makes it easier to identify where your usage rules need review. You can discuss how to connect communication of the rules with practical exercises through ZETTAi's corporate AI training. The scope of individual support will be confirmed during your consultation.
Related Articles
- What Information Should You Not Enter into Generative AI? Criteria and Responses for Business Data
- Why Is Generative AI Failing to Take Hold Internally? Improvement Steps Based on Usage
References
The following are publicly available resources from public agencies, also cited in the article. Resource titles and link destinations are listed separately.
- Scope of use and risk management: Digital Agency, page listing the latest "Guidelines for the Procurement and Use of Generative AI (DS-920)" https://www.digital.go.jp/resources/standard_guidelines
- Handling personal information: Personal Information Protection Commission, "Advisory on the Use of Generative AI Services" https://www.ppc.go.jp/news/careful_information/230602_AI_utilize_alert/
- Security measures for users: IPA, "Security Tips for AI Users" https://www.ipa.go.jp/digital/ai/security/ai_security_tips.html
- Copyright checks: Agency for Cultural Affairs, "AI and Copyright" https://www.bunka.go.jp/seisaku/chosakuken/aiandcopyright.html
References checked on: November 14, 2026. Update the template to match the services your company uses, its information categories, and its approval structure.