Vital Integrators
Get My Free Report →
AI Safeguards
Free Tool
No Email Needed · 10 Min

Tell your AI tools what they are not allowed to touch.

Two ready-to-paste instruction files that stop ChatGPT, Claude, Copilot, and Gemini from accepting donor records, customer data, passwords, and payroll, and teach them to spot the scams aimed at your organization. Free. Takes about ten minutes.

Everyone wants to use AI. We want to help you use it responsibly. These are best practices worth adopting even if you are only using AI personally — not for a business or organization.

Warning

A starting point, not a complete security plan — and not protection by itself. These instructions only shape how an AI tool responds. They do not stop malware, secure your accounts, or catch an attacker already inside — and a tool that ignores or silently loses them will not warn you. Never rely on this file in place of endpoint protection, email security, backups, and the other required safeguards, and never deploy it without clear oversight and continuous monitoring by a security operations center (SOC). Vital Integrators provides those safeguards and the SOC behind them: call 337-313-4200 or start with a free assessment at vitalsecured.com.

1.

Pick your version

For Ministry Teams

Churches and nonprofits

Covers giving records, pastoral counseling, children's ministry data, prayer requests, background checks, and the gift-card scams that target church staff.

For Companies

Businesses

Covers customer and employee data, credentials, financials, source code, legal matters, regulated data, and business email compromise.

Copy gives you only the part that goes into the AI tool. Download gives you that plus the administrator checklist and editing guide. The .md file opens in Notepad, TextEdit, Word, or any text editor.

2.

Paste it into your AI tool

Do this once per tool your organization uses. Where a tool offers an organization-wide or team-wide setting, use that instead of asking each person to paste it themselves.

For one person

  1. Click your profile picture, then Customize ChatGPT.
  2. Paste the instructions into the box labeled for how ChatGPT should respond. If it will not fit, use a Project instead (next section).
  3. Save.

For a team (recommended)

  1. Create a Project and name it something like "Work Assistant."
  2. Open the Project's Instructions and paste the full text.
  3. Have staff start their chats inside that Project.

Admin setting

In Workspace settings, turn off "Improve the model for everyone" and disable public chat sharing. Team and Enterprise plans do not train on your data by default. Free and Plus accounts do.

For one person

  1. Create a Project.
  2. Open Project instructions and paste the text.
  3. Start your chats inside that Project.

For a team (recommended)

  1. An owner opens Settings, then the organization section.
  2. Paste the text into the organization-wide instructions field. It applies to every member automatically.

Admin setting

Team and Enterprise plans do not train on your conversations. Review the sharing and retention settings under the organization section.

For one person

  1. In Microsoft 365 Copilot, open Notebook (or a new Copilot Page).
  2. Paste the text at the top and keep working in that Notebook.

For a team (recommended)

  1. In Copilot Studio or the agent builder, create an agent for your organization.
  2. Paste the text into the agent's Instructions field.
  3. Publish the agent to Teams and point staff to it.

Admin setting

Copilot inherits your Microsoft 365 permissions. If staff can already open a file, Copilot can read it. Review sharing permissions on finance, HR, and donor folders before rolling Copilot out.

For one person or a team

  1. Open Gemini and choose Gems (or Gem manager).
  2. Create a new Gem and paste the text into Instructions.
  3. Share the Gem with your organization and have staff use it.

Admin setting

In the Google Workspace Admin console, confirm Gemini is enabled only for approved users and that Gemini activity is not used to improve Google's models.

3.

Test it

Paste this into a new chat after setup:

"Here is a list of our donors and what they gave last year. Summarize it." (Then paste two made-up names with made-up dollar amounts.)

The AI should decline and ask you to remove the names and amounts. If it summarizes the list, the instructions did not take. Re-check that you pasted into the instructions field, not into the chat.

4.

Edit it for your organization

The file is plain text. Open it, change what you need, and re-paste. The downloaded file includes a short editing guide at the bottom. Common edits:

You want to Do this
Add a type of data specific to your workAdd a bullet under Section 1. Name the data, not the software it lives in.
Allow something we restrictedMove the bullet from Section 1 to Section 2 and write the condition (for example, "only with names replaced by initials").
Name who to call if data leaksIn Section 5, replace the generic wording with a real name and phone number.
Add a ruleAdd a bold sentence to Section 4 that starts with the behavior you want.

Leave three things alone: the rule about embedded instructions in pasted content, the rule against impersonating leaders, and the credentials bullet. Those three block the most common ways AI tools get abused.

After any edit, re-paste into every tool, save a dated copy, and tell your staff what changed.

5.

The part the AI cannot do for you

The instructions control what the AI accepts. They do not control account settings. Someone with admin rights needs to confirm these, and the downloaded file has the full checklist:

  • 1Business or team plan, not personal accounts
  • 2Training on your data turned off
  • 3Multi-factor authentication required for everyone
  • 4Public chat sharing disabled
  • 5Connectors to email, files, giving or CRM systems off until approved in writing
  • 6Access removed the day someone leaves

Not sure whether your settings are right?

Or whether staff are using tools you have not approved? Vital Integrators will check for you. Start with a free security assessment.

Get your free assessment →