How to Apply Role-Based Permissions to an AI Knowledge Hub

A company launches an internal AI assistant for employees.

A customer-service employee asks about the refund process and receives the correct answer. Then the same employee asks about payroll corrections and sees information intended only for HR. A contractor asks about project procedures and receives access to internal management guidance.

The assistant works. The permission model does not.

An AI knowledge hub should never become a shortcut around Microsoft 365 security. The system must identify who is asking, determine what that person is authorized to access and retrieve only from approved sources within those boundaries.

Start with Business Roles, Not Individual Employees

The cleanest permission model begins with groups aligned to real responsibilities.

Typical groups may include All Employees, Operations Staff, Customer Service, Managers, Human Resources, Finance, Executives and Contractors.

These groups should be created and maintained through Microsoft Entra ID or Microsoft 365 groups.

SharePoint sites and libraries can then grant access to the appropriate groups instead of assigning permissions directly to individual employees.

Design Permissions at the Source

Copilot Studio should not create a separate security system.

Permissions should be established in SharePoint before the agent is connected. The assistant should operate as a retrieval layer over the company’s existing identity and access controls.

A practical architecture may use a general employee knowledge site, department-specific libraries, restricted HR and finance sites, a manager-only knowledge area, a contractor library and separate archive locations.

The agent should retrieve only from sources the employee is authorized to access.

Avoid Excessive File-Level Permissions

SharePoint can apply permissions at the site, library, folder and document level.

That flexibility can become a maintenance problem.

A system containing hundreds of individually secured files becomes difficult to understand and almost impossible to test thoroughly.

Whenever possible, permissions should be assigned at the site or library level. Break inheritance only when there is a clear business requirement.

Connect Copilot Studio to Approved Sources

The agent should not be connected to an entire tenant or every SharePoint site.

Each source should have a defined business purpose, accountable owner, intended audience, tested permissions, approved content and a documented review process.

Copilot Studio should authenticate organizational users and retrieve only from approved knowledge sources.

Maisy uses SharePoint, Copilot Studio, Teams and Microsoft 365 identity controls to create this permission-aware retrieval layer.

Test the Same Question Across Different Roles

Permission testing should use realistic user accounts, not administrator credentials.

Create representative accounts for employees, managers, HR, finance and contractors. Ask the same questions using each identity.

Testing should also confirm that restricted citations do not reveal file names, locations or metadata. A response may appear safe while the linked source exposes information the employee should not know exists.

The secure AI knowledge-hub architecture explains why testing must cover both the answer and every supporting citation.

Maintain Access as Employees Change Roles

Permissions are not finished at launch.

Employees transfer departments, contractors complete assignments and managers gain temporary responsibilities. Group membership must be reviewed and updated as part of normal employee lifecycle processes.

Periodic access reviews should identify former employees still in groups, transferred employees retaining old access, temporary permissions that never expired and direct permissions outside the standard model.

This helps the company retain institutional knowledge without allowing sensitive information to spread beyond its intended audience.

Make the AI Layer Respect the Business

A useful internal assistant should provide the most complete answer that can be created entirely from information the employee is authorized to access—and nothing else.

Maisy helps companies retain institutional knowledge while preserving the security boundaries already established in Microsoft 365.

The AI assistant should make approved knowledge easier to reach. It should never make restricted knowledge easier to bypass.