Copilot Semantic Index and Oversharing Fix

Deploying Microsoft 365 Copilot is one of the most significant digital workplace upgrades available to UK organisations, promising immediate acceleration in drafting documents, analysing financial data, and managing projects. However, the moment Copilot licences are assigned, many organisations suffer immediate, embarrassing internal data leaks. Executive compensation spreadsheets, pending restructuring plans, confidential disciplinary hearings, and acquisition targets suddenly surface in conversational answers provided to regular staff members.

Copilot does not bypass Microsoft 365 permissions or invent new security vulnerabilities. Instead, it acts as an unforgiving, comprehensive mirror of your organisation's existing SharePoint and OneDrive permissions. In doing so, it immediately destroys "security through obscurity". Understanding how the Copilot Semantic Index queries your data and implementing strict permission boundaries is critical to preventing catastrophic oversharing before enterprise licences are activated.

Quick answer

Copilot uses the Semantic Index for Microsoft 365 to map relationships between users, documents, emails, and chats, faithfully returning any content the querying user has permission to read. To eliminate oversharing before rolling out licences, audit your SharePoint Online environment to remove broad group claims—specifically "Everyone except external users" (EEEU) and public Microsoft 365 Groups—from sensitive HR, payroll, and leadership document libraries. If deep permission remediation across hundreds of legacy sites cannot be completed prior to your scheduled pilot, immediately activate Restricted SharePoint Search via PowerShell to confine tenant-wide Copilot queries to an explicitly approved list of up to 100 enterprise sites.

Why this keeps happening

For more than a decade, organisations have accumulated SharePoint sites, Microsoft Teams channels, and OneDrive folders without maintaining rigorous access governance. In the pre-AI era, poor permission hygiene rarely triggered immediate alarm because files remained buried deep inside convoluted folder hierarchies. Staff simply did not know the files existed, what they were named, or where they were stored.

Four primary factors turn dormant permission sprawl into live data exposure when Copilot arrives:

1. The Myth of Security Through Obscurity: When an IT administrator or department head mistakenly granted access to "Everyone except external users" on an executive document library five years ago, nobody noticed because the link was never shared. The Semantic Index indexes all accessible text; when an employee asks Copilot "What are our bonus allocations this year?", Copilot locates and summarises the forgotten spreadsheet in seconds.
2. Default Public Teams and SharePoint Sites: When users create a new Microsoft Team, the default setting is often "Public", allowing anyone in the organisation to browse or join the site. Confidential strategy documents uploaded into general channels of public teams inherit open read permissions across the entire workforce.
3. Broken Permission Inheritance: Folder-level and file-level sharing links ("People in your organisation with the link") break clean site inheritance, scattering ad-hoc permissions across thousands of individual documents that cannot be easily tracked from the SharePoint Admin Center.
4. Natural Language Semantic Discovery: Unlike traditional keyword search, which required exact phrase matching, the Semantic Index uses vector embeddings. A user searching for "staff pay raises" will retrieve documents titled Remuneration_Strategy_2027.xlsx even if the word "pay" never appears in the title or text.

Fix path

Eliminating Copilot oversharing requires auditing tenant-wide permission grants and deploying immediate containment controls while long-term data cleanup proceeds.

Auditing broad group sharing

The first line of defence is identifying and eliminating tenant-wide read access on sensitive sites.

Administrators must locate any document library or site where the following security principals hold read or edit rights:

• c:0(.s|true (The internal claim for Everyone except external users)
• Everyone
• Public Microsoft 365 Groups containing all staff members

To discover public sites and overshared libraries using the PnP PowerShell module:

# Connect to SharePoint Online Admin tenant
Connect-PnPOnline -Url "https://yourtenant-admin.sharepoint.com" -Interactive

# Query all SharePoint sites and flag public Microsoft 365 Groups
$sites = Get-PnPTenantSite -Filter "Template -eq 'GROUP#0'"
foreach ($site in $sites) {
    $group = Get-PnPMicrosoft365Group -Identity $site.GroupId
    if ($group.Visibility -eq "Public") {
        Write-Warning "Public Team Site Detected: $($site.Url) - Convert to Private immediately!"
    }
}

For high-priority sites holding financial, personnel, or commercial data:

1. Convert Public Teams to Private: In the Teams or SharePoint Admin Center, switch site visibility from Public to Private so that only explicitly added members can access documents.
2. Remove "Everyone except external users": Inspect site permissions, remove EEEU from the Members and Visitors groups, and replace it with defined, role-based Microsoft Entra security groups (e.g., SG-Finance-Team).
3. Turn Off Open Sharing Links: In SharePoint Admin Center, set tenant and site sharing policies to restrict new sharing links to "Specific people" rather than "People in your organisation".

Implementing Restricted SharePoint Search

Auditing hundreds of historical SharePoint sites takes weeks or months of coordinated effort across department heads. If your business has already procured Copilot licences and needs to launch safely without delaying deployment, Microsoft provides a dedicated containment mechanism: Restricted SharePoint Search (RSS).

When Restricted SharePoint Search is enabled, Copilot and Microsoft Search tenant-wide queries are restricted to:

• A curated list of up to 100 organisation-wide SharePoint sites that your IT team has explicitly reviewed and approved for broad discovery (such as corporate intranets, public policies, and approved reference materials).
• The user's personal OneDrive files and content they have explicitly authored or interacted with.
• Sites where the user is already an explicit member or regular visitor.

To enable Restricted SharePoint Search using the SharePoint Online Management Shell:

# Connect as SharePoint Administrator
Connect-SPOService -Url "https://yourtenant-admin.sharepoint.com"

# Enable Restricted SharePoint Search mode across the tenant
Set-SPOTenant -RestrictedSearchMode $true

# Confirm the feature status
Get-SPOTenant | Select-Object RestrictedSearchMode

# Define your approved organization-wide sites (up to 100 sites)
$AllowedSites = @(
    "https://yourtenant.sharepoint.com/sites/Intranet",
    "https://yourtenant.sharepoint.com/sites/CompanyPolicies",
    "https://yourtenant.sharepoint.com/sites/BrandGuidelines"
)

# Apply the approved list to the tenant search boundary
Set-SPOTenantRestrictedSearchList -Sites $AllowedSites

# Verify active sites in the restricted search list
Get-SPOTenantRestrictedSearchList

By enforcing Restricted SharePoint Search, Copilot will no longer crawl or surface information from uncurated, legacy, or misconfigured departmental sites during broad search prompts. This provides an immediate, robust safety perimeter while your teams systematically clean up background site permissions.

Practical SME workflow

For a UK business deploying Microsoft 365 Copilot across 20 to 250 users, follow this sequenced rollout framework:

1. Triage Crown Jewel Repositories: Map where HR records, payroll sheets, board packs, legal disclosures, and merger documentation are stored. Confirm that every one of these sites is private, has unique permissions, and contains zero broad claims.
2. Activate Restricted SharePoint Search: Turn on Restricted SharePoint Search before issuing licences to pilot users. Populate the allowed list only with verified, general-interest sites.
3. Deploy Microsoft Purview Sensitivity Labels: Apply sensitivity labels with mandatory encryption (such as "Confidential - Board Only" or "Strictly Internal - HR") to your highest-risk document classes. Files encrypted with Purview labels cannot be indexed or summarised by Copilot unless the querying user possesses explicit rights management decryption credentials.
4. Conduct "Red Team" Pilot Prompts: Before broad launch, have a trusted pilot user execute deliberately targeted prompts designed to test boundary leakage, such as "Summarise recent salary adjustments" or "Show me draft redundancy memos". Confirm that Copilot returns zero citations.
5. Progressive Site Expansion: As departmental managers review their site permissions and certify their data hygiene, add approved departmental repositories to the Restricted SharePoint Search list.

Risk and control

Deploying generative search over enterprise document repositories requires proactive permission and encryption controls.

What good evidence looks like

When validating data boundaries for Cyber Essentials Plus, ISO 27001 scopes, or internal compliance audits, compile the following evidence:

• Restricted Search Configuration Proof: PowerShell command outputs proving that RestrictedSearchMode is active and displaying the curated array of allowed site URLs.
• Tenant Sharing Baseline Report: Exported SharePoint Admin Center reports verifying that default sharing links are set to "Specific people" rather than company-wide links.
• Crown Jewel Site Permission Audit: Detailed permission matrices for HR, Finance, and Leadership sites demonstrating that only designated personnel hold read access.
• Purview Label Adoption Telemetry: Compliance dashboard reports demonstrating that sensitivity labels are actively applied to confidential executive files.

Keep the change reversible

Restricted SharePoint Search is designed to be fully reversible, ensuring organisations can adjust boundaries without data loss or administrative lockouts:

• Dynamic Site Adjustments: Sites can be added to or removed from the restricted list at any time using Add-SPOTenantRestrictedSearchList or Remove-SPOTenantRestrictedSearchList. Changes propagate across the search index within 24 hours.
• Instant Full-Index Restoration: If permission remediation is complete across all tenant sites and you wish to restore native full-tenant semantic indexing, execute:

  Set-SPOTenant -RestrictedSearchMode $false
  

  Copilot will automatically resume indexing all user-accessible content across the entire tenant without requiring licence reassignments or configuration rebuilds.

The common mistake

The most common failure pattern is blaming Copilot for enterprise security breaches and attempting to solve the problem by revoking Copilot licences or demanding refunds. Copilot does not create the exposure; it merely reveals the dangerous permission sprawl that had already been present in your Microsoft 365 tenant for years. Removing Copilot leaves the underlying security vulnerability completely intact—any curious employee could still locate those same payroll spreadsheets via manual SharePoint URL queries. Use Copilot as the catalyst to fix your underlying data architecture once and for all.

Related field note

To evaluate whether your Microsoft 365 environment, identity policies, and data boundaries are ready for an enterprise AI rollout, book our AI rollout diagnostic.

Source basis

• Microsoft Learn: Data, Privacy, and Security for Microsoft 365 Copilot
• Microsoft Learn: Manage Restricted SharePoint Search