Skip to content

Baseline Configuration

Guide to configuring agent scope baselines for the Scope Drift Monitor.

Overview

A scope baseline defines what data sources and connectors an AI agent is allowed to access. The Scope Drift Monitor compares actual agent access against this baseline to detect drift.

Agent Access Event → Compare to Baseline → Match? → No Action
                                        → No Match? → Create Violation

Creating Baselines

Option 1: Auto-Generate from Audit History

Use the New-AgentBaseline.ps1 script to analyze historical access and generate a baseline automatically.

.\scripts\New-AgentBaseline.ps1 `
    -AgentId "12345678-1234-1234-1234-123456789012" `
    -Environment "https://contoso.crm.dynamics.com" `
    -EnvironmentId "87654321-4321-4321-4321-210987654321" `
    -OwnerId "user-guid" `
    -Days 7

What it does:

  1. Queries Office 365 Management API for CopilotInteraction events
  2. Filters events by agent ID
  3. Extracts accessed connectors, sites, tables, and APIs
  4. Creates an fsi_agentscope record with populated allowed lists
  5. Sets status to Active immediately

Best for:

  • Existing agents with established access patterns
  • Quick baseline creation
  • Agents in Zone 1 or Zone 2

Option 2: Manual Configuration

Create baselines manually for precise control over allowed resources.

Steps:

  1. Navigate to your Dataverse environment
  2. Go to Tables > Agent Scope (fsi_agentscope)
  3. Click + New record
  4. Fill in required fields (see Scope Definition Fields)
  5. Set Status to Active when ready

Best for:

  • New agents before deployment
  • Zone 3 enterprise agents requiring strict control
  • Agents with known, limited scope

Scope Definition Fields

Required Fields

Field Description Example
fsi_name Display name "Customer Service Agent"
fsi_agentid Copilot Studio agent GUID 12345678-1234-...
fsi_environmentid Power Platform environment ID 87654321-4321-...
fsi_zone Governance zone (10001-10003) 10003 (Enterprise Managed)
fsi_owner Agent owner (user lookup) John Smith
fsi_purpose Declared agent purpose "Answer customer inquiries"
fsi_status Baseline status 10002 (Active)

Scope Arrays (JSON Format)

Field Description Example
fsi_allowedconnectors Allowed connector names ["SharePoint", "Dataverse"]
fsi_allowedsites Allowed SharePoint site URLs ["https://contoso.sharepoint.com/sites/KB"]
fsi_allowedtables Allowed Dataverse table names ["contact", "case"]
fsi_allowedapis Allowed external API URLs ["https://api.contoso.com/v1"]

Optional Fields

Field Description Example
fsi_dataowner Data steward (user lookup) Jane Doe
fsi_lastvalidated Last validation timestamp 2026-02-01T00:00:00Z
fsi_nextreview Next scheduled review date 2026-05-01

Scope Status Values

Value Label Meaning
10001 Draft Baseline being configured, not monitored
10002 Active Baseline active, violations detected
10003 Under Review Baseline under review (pauses monitoring)
10004 Suspended Baseline suspended (not monitored)
10005 Archived Baseline archived (no monitoring)

Monitoring behavior:

  • Active: Full monitoring, violations created
  • Draft/Under Review: No violations created
  • Suspended: Not monitored (fsi_status eq 10002 filter excludes suspended scopes)
  • Archived: No monitoring

Governance Zones

Baselines are classified by governance zone, affecting detection behavior.

Zone Label Detection Frequency Approval Rigor
10001 Personal Productivity Every 15 minutes Security team (via fsi_SDM_SecurityTeamEmail)
10002 Team Collaboration Every 15 minutes Security team (via fsi_SDM_SecurityTeamEmail)
10003 Enterprise Managed Every 15 minutes Security team (via fsi_SDM_SecurityTeamEmail)

Note: Zone-differentiated approval rigor (self-service for Zone 1, team lead for Zone 2, dual Security + Data Owner for Zone 3) is planned but not yet implemented. All zones currently route expansion requests to the security team. The schema includes fsi_dataownerapproval and fsi_dataownerapprovedby fields for future use.

Zone selection criteria:

Criteria Zone 1 Zone 2 Zone 3
User scope Individual Team/Department Organization-wide
Data sensitivity Low Medium High
Regulatory impact None Some Critical
Approval required Security team Security team Security team

Scope Array Format

All scope arrays must be valid JSON arrays of strings.

Valid Examples

// Single item
["SharePoint"]

// Multiple items
["SharePoint", "Dataverse", "Outlook"]

// Empty (monitor all access)
[]

// SharePoint sites
["https://contoso.sharepoint.com/sites/KB", "https://contoso.sharepoint.com/sites/HR"]

// Dataverse tables
["contact", "account", "case", "knowledgearticle"]

// External APIs
["https://api.contoso.com/v1", "https://data.contoso.com/graphql"]

Invalid Examples

// Not an array
"SharePoint"

// Null value
null

// Missing quotes
[SharePoint, Dataverse]

// Empty string (use [] instead)
""

Empty Baselines

An empty baseline (all arrays set to []) has special meaning:

Behavior Description
All access flagged Any data access creates a violation
Use case New agents, pre-deployment validation
Expected outcome High violation volume initially

To allow all access (no monitoring):

Set baseline status to Draft or Archived instead of using empty arrays with Active status.

Updating Baselines

Manual Update

  1. Open the agent scope record in Dataverse
  2. Modify the allowed arrays
  3. Save the record

Note: Changes take effect immediately for subsequent detection runs.

Via Expansion Workflow

When a scope expansion request is approved:

  1. SDM-ExpansionProcessor adds the resource to the appropriate array
  2. No manual intervention required
  3. Audit trail maintained in expansion request record

Programmatic Update

# Example: Add a SharePoint site to baseline
$scope = Get-DataverseRecord -EntityName "fsi_agentscope" -RecordId $scopeId
$currentSites = $scope.fsi_allowedsites | ConvertFrom-Json
$currentSites += "https://contoso.sharepoint.com/sites/NewSite"
$updatedSites = $currentSites | ConvertTo-Json -Compress

Update-DataverseRecord -EntityName "fsi_agentscope" -RecordId $scopeId -Data @{
    fsi_allowedsites = $updatedSites
    fsi_lastvalidated = (Get-Date).ToUniversalTime().ToString("yyyy-MM-ddTHH:mm:ssZ")
}

Best Practices

Initial Setup

  1. Start with audit-generated baseline - Captures actual access patterns
  2. Review before activating - Remove unexpected or one-time access
  3. Set appropriate zone - Affects detection frequency and approval rigor
  4. Document purpose clearly - Helps reviewers understand expected access

Ongoing Maintenance

  1. Review baselines quarterly - Set fsi_nextreview dates
  2. Archive unused baselines - Agents no longer in use
  3. Monitor violation trends - High volume may indicate scope creep
  4. Use expansion workflow - Maintains audit trail for changes

Zone-Specific Guidance

Zone 1 (Personal):

  • Broader allowed scope acceptable
  • Security team approval for expansion (self-service planned for future)
  • Focus on connector restrictions

Zone 2 (Team):

  • Team-specific SharePoint sites
  • Shared Dataverse tables
  • Security team approval for expansion (team lead approval planned for future)

Zone 3 (Enterprise):

  • Minimal required access only
  • Security team approval required
  • Regular compliance reviews

Scope Drift Monitor v1.2.0