Control 1.5: Data Loss Prevention (DLP) and Sensitivity Labels

Overview

Control ID: 1.5 Control Name: Data Loss Prevention (DLP) and Sensitivity Labels Regulatory Reference: FINRA 4511, SEC Reg S-P, GLBA 501(b), SOX 404 Setup Time: 2-4 hours

Purpose

Prevent sensitive financial data from unauthorized exposure through AI agents by implementing DLP policies that detect and block sensitive information in agent knowledge sources, user prompts, and agent responses. This control helps protect customer PII, financial data, and trading information when accessed or processed by Microsoft 365 Copilot and Copilot Studio agents.

Description

Data Loss Prevention policies protect sensitive information from unauthorized exposure through AI agents. When combined with sensitivity labels, DLP provides comprehensive data protection across Microsoft 365 Copilot, Copilot Studio agents, and other AI applications. Integration with DSPM for AI enables oversharing detection and AI-specific policy enforcement.

US-only scope note

This control implementation guidance is U.S.-only:

Use U.S. identifiers (e.g., SSN, ITIN, U.S. bank account/routing formats, CUSIP, FINRA CRD) and organization-specific internal identifiers.
Avoid non-U.S. examples or patterns. If your organization operates outside the U.S., create a separate addendum with jurisdiction-specific DLP policies, SITs, and validation steps.

Dependency: Sensitive Information Types (Control 1.13)

DLP effectiveness depends on Sensitive Information Types (SITs). Implement and validate SITs per Control 1.13 before enforcing DLP in AI scenarios.

See DLP for M365 Copilot for AI-specific DLP capabilities.

DLP Enforcement Status Update (2025)

DLP Enforcement Now Enabled by Default

As of early 2025, data policy enforcement for Copilot Studio is enabled by default for all tenants (per Microsoft message center alert MC973179). Organizations no longer need to manually enable DLP enforcement. (Source)

This means:

New tenants automatically have DLP enforcement active for Copilot Studio
Existing tenants were migrated to enabled status in early 2025
DLP policies targeting Copilot Studio connectors are now enforced immediately

Copilot Studio Channel Connectors (6 Total)

DLP policies can control which publishing channels Copilot Studio agents can use. Microsoft supports 6 channel connectors:

#	Channel Connector	Description
1	Direct Line channels	Web chat, custom apps via Direct Line API
2	Microsoft Teams + M365	Teams chat and M365 surfaces
3	Facebook	Facebook Messenger integration
4	Omnichannel	Dynamics 365 Omnichannel for Customer Service
5	SharePoint	SharePoint site embedding
6	WhatsApp	WhatsApp Business integration

Blocking Agent Publishing

To prevent agents from being deployed (while still allowing creation), block ALL 6 channel connectors via DLP. If no channels are allowed, agents cannot be published to any audience. See Control 1.1 for the "Sterile Default Environment Strategy."

Source: DLP Example 6 - Channel Control

Key Capabilities

Capability	Description	FSI Relevance
AI-aware DLP policies	Policies targeting AI applications	Prevent sensitive data in AI
Sensitivity label enforcement	Block/warn based on labels	Classification-based protection
Oversharing assessment	Identify data exposure risks	Risk identification
DSPM integration	Unified AI data protection	Comprehensive visibility

DLP for AI Applications

AI-Specific DLP Policies

DLP policies can specifically target AI applications:

Policy Type	Description	Use Case
Detect sensitive info added to AI sites	Monitor sensitive data in AI knowledge sources	Prevent exposure via grounding data
Detect sensitive info shared in AI prompts	Monitor user prompts for sensitive content	Prevent data leakage in queries
Block sensitive info in agent responses	Prevent agents from returning sensitive data	Response filtering

Configuring AI DLP Policies

Open Microsoft Purview
Navigate to Data loss prevention → Policies
Create or modify a policy (start in Test with notifications when possible)
Select only the in-scope AI locations:
- Microsoft 365 Copilot (prompts and responses)
- Copilot Studio (agent interactions)
- Optional: SharePoint/OneDrive/Teams/Exchange only if you intend to control those workloads as well
- Optional (endpoints): Devices if your tenant uses Purview Endpoint DLP and you want endpoint enforcement for AI-related exfiltration paths
Configure conditions using SITs (see Control 1.13) and/or sensitivity labels
Choose actions appropriate to the governance tier (Audit, Warn, Block)
Save, publish, and allow time for propagation before declaring pass/fail test results

AI DLP setup steps (Copilot prompts/responses)

Use these steps to validate that AI-specific DLP is correctly targeted and producing audit-ready outputs before enforcement.

Confirm prerequisites:
- Required roles are assigned (see Prerequisites)
- Unified audit logging is enabled (see Dependencies)
- SITs are validated per Control 1.13
Create a dedicated policy for AI interactions first (avoid combining with broad workload DLP until stable).
Ensure the policy scope includes only the intended AI locations:
- Microsoft 365 Copilot (prompts and responses)
- Copilot Studio (agent interactions)
Start in Test with notifications:
- Enable user notification where appropriate for the tier (Warn/Block)
- Enable incident reports for Tier 2-3 scenarios that require evidence
Execute the U.S.-only test cases in this document and confirm:
- Expected Warn/Block behavior occurs
- Events appear in audit/incident reporting within expected propagation time
Only after the test suite passes, switch the policy to Enforce for Tier 3 use cases.

See Create DLP policies for detailed steps.

DSPM for AI Integration

DLP policies integrate with DSPM for AI for unified management:

Viewing DLP Policies in DSPM

Navigate to DSPM for AI → Policies
Expand Data Loss Prevention section
View AI-related DLP policies and status

Example AI DLP Policies

Policy Name	Solution	Purpose
DSPM for AI: Detect sensitive info added to AI sites	Data Loss Prevention	Monitor knowledge sources
DSPM for AI - Detect sensitive info shared in AI prompts in Edge	DSPM for AI	Browser-based protection

Oversharing Assessments

What is Oversharing?

Oversharing occurs when sensitive data is accessible to more users than necessary, creating risk when AI agents access that data as grounding content.

DSPM Oversharing Detection

DSPM for AI provides oversharing assessments:

Navigate to DSPM for AI → Data risk assessments
Review Assess and prevent oversharing section
Follow three-step process:
Identify: Review weekly assessment results
Protect: Apply labels and retention policies
Monitor: Conduct access reviews

Custom Oversharing Assessments

Create targeted assessments for agent knowledge sources:

Click + Create custom assessment
Select data sources (SharePoint sites used by agents)
Define user scope
Review overshared items count
Remediate excessive permissions

Sensitivity Labels

Label-Based Protection for Agents

Label	Agent Access	Recommendation
Public	Full access	All governance tiers
Internal	Full access	All governance tiers
Confidential	Restricted	Team/enterprise agents only
Highly Confidential	Blocked	Block from all agents

Recommended label taxonomy (U.S.-only)

Define a small, auditable taxonomy first; add optional sub-labels only when you can enforce and measure them.

Label	What it means	Typical content	Default AI/DLP posture
Public	Approved for public release	Marketing content, public disclosures	Allow; audit optional
Internal	Business-use only; low sensitivity	Internal procedures, org charts	Allow; audit recommended
Confidential	Customer NPI / regulated internal data	Customer contact details, KYC docs, account metadata	Warn or block depending on workflow; always log
Highly Confidential	High impact if exposed	Full customer PII sets, trading strategies/MNPI, privileged security data	Block by default; incident report

Taxonomy rules (recommended):

Use consistent naming and avoid jurisdiction-mixing (this document is U.S.-only).
Decide who can apply labels (manual vs auto-labeling) and document the decision.
Publish labels to:
- Users who create or handle the content
- Owners/operators of agents that can access labeled knowledge sources
- Compliance/SecOps roles who investigate DLP events

Configuring Label-Based DLP

Create sensitivity labels in Purview
Apply labels to content
Create DLP policy with label conditions
Configure actions:
Highly Confidential: Block agent access
Confidential: Warn user, log access
Internal/Public: Allow access

Agent Knowledge Source Labels

For SharePoint sites used as agent knowledge sources:

Content Type	Recommended Label	DLP Action
Customer PII	Confidential	Warn + Log
Financial data	Highly Confidential	Block
Internal docs	Internal	Allow
Public content	Public	Allow

Prerequisites

Primary Owner Admin Role: Purview Compliance Admin Supporting Roles: Purview Info Protection Admin

Licenses Required

License	Purpose	Required For
Microsoft 365 E5	Full DLP and sensitivity labels	All governance levels
Microsoft 365 E5 Compliance	Standalone compliance suite	Alternative to E5
Microsoft Purview Data Loss Prevention	DLP capabilities	Level 2+
Microsoft Purview Information Protection	Sensitivity labels	All levels

Permissions Required

Role	Purpose	Assignment Method
Compliance Administrator	Create and manage DLP policies	Entra ID
Information Protection Admin	Manage sensitivity labels	Entra ID
Security Administrator	View DLP reports and alerts	Entra ID
Compliance Data Administrator	Access compliance data	Purview RBAC

Dependencies

Dependency	Description	Verification
Microsoft Purview portal	DLP management interface	Access https://purview.microsoft.com
Sensitive Information Types (SITs)	SITs power DLP detection and tuning	Implement Control 1.13
Sensitivity labels	Labels must be created and published	Check Purview → Information protection
Unified audit logging	Required for DLP event logging	Verify in Purview → Audit
DSPM for AI	Optional but recommended	Check DSPM availability in tenant
Purview Endpoint DLP (optional)	Extends enforcement to Windows endpoints for exfiltration paths	Verify Data loss prevention → Endpoint DLP settings

Pre-Setup Checklist

[ ] E5 or E5 Compliance licenses assigned to users
[ ] Sensitivity label taxonomy defined for organization
[ ] Labels published to target users and groups
[ ] DLP administrator access verified
[ ] Agent inventory completed (know what agents access which data)
[ ] Control 1.13 SITs implemented and validated (built-in + any required custom SITs)
[ ] Sensitive information types selected for this DLP policy scope (e.g., SSN, ABA routing, U.S. bank account)
[ ] Decide whether Endpoint DLP is in-scope for AI-related exfiltration paths (optional)

Governance Levels

Level 1 - Baseline

Requirement	Configuration
Basic DLP	Policies for common sensitive info types
Labels defined	Sensitivity label taxonomy created
Agent awareness	Identify what data agents can access

Minimum requirements:

DLP policies for PII, financial data
Sensitivity labels published
Document agent data access patterns

Level 2-3 - Recommended

Requirement	Configuration
AI-specific DLP	Policies targeting AI applications
Label enforcement	DLP conditions based on labels
Oversharing review	Quarterly DSPM assessments
DSPM integration	DLP policies visible in DSPM

FSI recommendations:

Enable "Detect sensitive info added to AI sites" policy
Configure label-based blocking for Highly Confidential
Quarterly oversharing assessments for agent knowledge sources
Weekly review of DLP alerts related to AI

Level 4 - Regulated/High-Risk

Requirement	Configuration
Comprehensive DLP	All AI touchpoints covered
Mandatory labels	Auto-labeling for sensitive content
Real-time blocking	Block mode for enterprise-managed agents
Continuous assessment	Monthly custom oversharing assessments

FSI requirements:

Block enterprise-managed agents from Highly Confidential content
Auto-labeling for customer data repositories
Real-time DLP alerts to SOC
Monthly oversharing assessments with remediation tracking
Integration with DSPM for unified reporting

Setup & Configuration

Governance Tier DLP Configuration

Tier	DLP Mode	Label Requirement	Oversharing Review
Tier 1	Audit only	Optional	Annual
Tier 2	Warn	Recommended	Quarterly
Tier 3	Block	Mandatory	Monthly

SIT foundation (Dependency: Control 1.13)

Before enforcing DLP for AI interactions, confirm SIT readiness:

Review the SIT inventory in Purview (Data classification → Classifiers → Sensitive info types).
Validate key U.S.-only SITs with a small set of sanitized test data (see Control 1.13).
If you use custom SITs (e.g., internal account identifiers), validate and document:
- Regex/pattern logic
- Supporting keywords
- False-positive exclusions
- The minimum count/confidence you will require in DLP rules

If SITs are not validated first, DLP “false negative” troubleshooting becomes guesswork.

Create an AI-scoped DLP policy (Purview UI)

Use the smallest policy surface area that meets the requirement.

Purview → Data loss prevention → Policies → + Create policy
Choose a template closest to your need (Financial/Privacy) or start from Custom
Name the policy with an auditable convention (example): FSI-AI-DLP-Data-Protection
Select locations:
- Microsoft 365 Copilot
- Copilot Studio
- Optional: Devices (Endpoint DLP) if in scope
Add rules using:
- SIT-based conditions (preferred for detection)
- Label-based conditions (preferred for enforcement posture)
Configure actions by tier (Audit/Warn/Block), enable user notifications if appropriate, and configure incident reports
Run in Test with notifications first, execute the test cases in this document, then move to Enforce when results are acceptable

Extend to endpoints (optional; avoid overreach)

If your risk model includes endpoint exfiltration (copy/paste, print, USB, browser uploads), use Purview Endpoint DLP targeted to:

The specific devices/groups that use AI tools for work
The specific sensitive content scope (SITs/labels)

Avoid tenant-wide device blocking unless the organization has explicitly approved that operational impact.

Policy Templates for FSI

Customer Data Protection

Policy: Protect Customer PII in AI
Locations: M365 Copilot, Copilot Studio
Conditions:

- Sensitive info types: SSN, Account Numbers, Credit Card
- OR Sensitivity label: Confidential, Highly Confidential
Actions:

- Tier 1-2: Warn user, log event
- Tier 3: Block access, notify compliance

Financial Data Protection

Policy: Block Financial Data in AI Responses
Locations: Copilot Studio agents
Conditions:

- Sensitive info types: Financial statements, Trading data
- Sensitivity label: Highly Confidential
Actions:

- Tier 2: Warn, log event, incident report (as configured)
- Tier 3: Block, notify security team (incident report)

PowerShell Configuration

Connect to Security & Compliance Center

# Install module if needed
Install-Module -Name ExchangeOnlineManagement -Force

# Connect to Security & Compliance Center
Connect-IPPSSession -UserPrincipalName admin@contoso.com

# Verify connection
Get-DlpCompliancePolicy | Select-Object Name, Mode, Enabled

Create AI-Focused DLP Policy

# Create DLP policy for AI applications
$policyParams = @{
    Name = "FSI-AI-Data-Protection"
    Comment = "Protect sensitive data in AI applications"
    Mode = "Enable"  # Use "TestWithNotifications" for testing
    Priority = 1
    ExchangeLocation = "All"
    SharePointLocation = "All"
    OneDriveLocation = "All"
    TeamsLocation = "All"
}

$policy = New-DlpCompliancePolicy @policyParams

# Create rule for SSN protection
$ruleParams = @{
    Name = "Block SSN in AI Interactions"
    Policy = "FSI-AI-Data-Protection"
    ContentContainsSensitiveInformation = @{
        Name = "U.S. Social Security Number (SSN)"
        minCount = 1
    }
    BlockAccess = $true
    NotifyUser = "SiteAdmin"
    GenerateIncidentReport = "SiteAdmin"
}

New-DlpComplianceRule @ruleParams

Create Sensitivity Label-Based DLP Rule

# Rule to block Highly Confidential content in AI
$labelRuleParams = @{
    Name = "Block Highly Confidential from AI"
    Policy = "FSI-AI-Data-Protection"
    ContentPropertyContainsWords = @{
        "Document.SensitivityLabel" = "Highly Confidential"
    }
    BlockAccess = $true
    NotifyUser = "SiteAdmin,LastModifier"
    NotifyEndpointUser = "NotifyUser"
    GenerateIncidentReport = "SiteAdmin"
    IncidentReportContent = "All"
}

New-DlpComplianceRule @labelRuleParams

Audit DLP Policies

# Get all DLP policies and their status
Get-DlpCompliancePolicy | Format-Table Name, Mode, Enabled, CreatedBy, WhenCreated

# Get DLP rules for a specific policy
Get-DlpComplianceRule -Policy "FSI-AI-Data-Protection" |
    Select-Object Name, Priority, BlockAccess, Disabled

# Get DLP policy details
Get-DlpCompliancePolicy -Identity "FSI-AI-Data-Protection" |
    Select-Object * | Format-List

# Export policy configuration for documentation
Get-DlpCompliancePolicy |
    Select-Object Name, Mode, Enabled, SharePointLocation, TeamsLocation, ExchangeLocation |
    Export-Csv -Path "DLP-Policies-Report.csv" -NoTypeInformation

Monitor DLP Alerts

# Search audit log for DLP events
$startDate = (Get-Date).AddDays(-30)
$endDate = Get-Date

Search-UnifiedAuditLog -StartDate $startDate -EndDate $endDate `
    -RecordType DLP -ResultSize 5000 |
    Select-Object CreationDate, UserIds, Operations, AuditData |
    Export-Csv -Path "DLP-Audit-Log.csv" -NoTypeInformation

# Parse DLP events for AI-related incidents
$dlpEvents = Search-UnifiedAuditLog -StartDate $startDate -EndDate $endDate `
    -RecordType DLP -ResultSize 1000

foreach ($event in $dlpEvents) {
    $data = $event.AuditData | ConvertFrom-Json
    if ($data.Workload -match "Copilot|Agent") {
        Write-Host "AI DLP Event: $($data.Operation) - $($data.ObjectId)"
    }
}

Get Sensitivity Label Statistics

# Connect to Microsoft Graph for label info
Connect-MgGraph -Scopes "InformationProtectionPolicy.Read.All"

# Get sensitivity labels
Get-MgBetaInformationProtectionSensitivityPolicyLabel |
    Select-Object Id, Name, Description, IsDefault |
    Format-Table

# Check label usage via compliance search
$labelSearch = New-ComplianceSearch -Name "Highly-Confidential-Content" `
    -ContentMatchQuery 'SensitivityLabelId:<label-guid>' `
    -ExchangeLocation All -SharePointLocation All

Start-ComplianceSearch -Identity "Highly-Confidential-Content"

Financial Sector Considerations

Regulatory Mapping

Regulation	DLP Requirement	Control Implementation
FINRA 4511	Protect customer records	Block customer PII in AI responses
SEC Reg S-P	Privacy protection	Sensitivity labels on customer data
GLBA 501(b)	Safeguard customer information	DLP for financial account numbers
SOX 404	IT controls for data protection	Audit logging of DLP events
FFIEC Guidelines	Data classification	Label-based access control

Governance Tier DLP Configuration

Tier	DLP Mode	SIT Blocking	Label Enforcement	Alert Response
Tier 1	Audit Only	Log only	Optional	Weekly review
Tier 2	Warn	User notification	Recommended	Daily review
Tier 3	Block	Real-time blocking	Mandatory	Immediate escalation

FSI-Specific Sensitive Information Types

SIT	FSI Use Case	DLP Action
U.S. SSN	Customer identification	Tier 3: Block; Tier 2: Warn + log
ABA Routing Number	Bank routing	Tier 3: Block; Tier 2: Warn + log
Credit Card Number	Payment data (PCI)	Tier 3: Block; Tier 2: Warn + log
U.S. Bank Account Number	Account data	Tier 3: Block; Tier 2: Warn + log
Custom: Internal Account Identifier	Internal account format	Tier 3: Block; Tier 2: Warn + log

FSI Example: Trading Desk Agent Protection

Agent: Trading-Desk-Assistant (Enterprise Managed)
Classification: Critical - Trading Data Exposure Risk
DLP Requirements:
  - Block all credit card numbers
  - Block SSNs and account numbers
  - Block content labeled "Highly Confidential"
  - Warn on "Confidential" content access
  - Real-time alerts to compliance team
  - Audit all data access requests

Policy Configuration:
  Mode: Block (Enforce)
  Locations: M365 Copilot, Copilot Studio
  Conditions:
    - Any of: SSN, Credit Card, ABA, Account Number
    - OR: Label = Highly Confidential
  Actions:
    - Block access
    - Notify: compliance@contoso.com
    - Create incident
    - Log to audit

Regulatory Context

Primary Regulations: FINRA 4511, SEC Reg S-P, GLBA 501(b), SOX 404

Regulation	DLP Support
FINRA 4511	Protect customer records from AI exposure
SEC Reg S-P	Privacy protection in AI systems
GLBA 501(b)	Safeguard customer information
SOX 404	IT controls for data protection

Examination Considerations

Regulators may request:

DLP policy configuration documentation
Evidence of AI-specific protections
Oversharing assessment results
DLP violation reports and remediation
Sensitivity label deployment evidence

Zone-Specific Configuration

Zone 1 (Personal Productivity):

Apply a baseline minimum of DLP policies and sensitivity labels that impacts tenant-wide safety (where applicable), and document any exceptions for personal agents.
Avoid expanding scope beyond the user’s own data unless explicitly justified.
Rationale: reduces risk from personal use while keeping friction low; legal/compliance can tighten later.

Zone 2 (Team Collaboration):

Apply DLP to AI locations (Copilot/M365, Copilot Studio) and label-conditioned rules for shared agents and shared data sources; require an identified owner and an approval trail.
Validate configuration in a pilot environment before broader rollout; retain policy export + test prompts/results.
Rationale: shared agents increase blast radius; controls must be consistently applied and provable.

Zone 3 (Enterprise Managed):

Require the strictest configuration for DLP policies and sensitivity labels and enforce it via policy where possible (not manual-only).
Treat changes as controlled (change ticket + documented testing); retain policy export + test prompts/results.
Rationale: enterprise agents handle the most sensitive content and are the highest audit/regulatory risk.

Verification & Testing

Step	Action	Expected Result
1	Navigate to Purview → DLP → Policies	Policies listed
2	Verify AI locations selected	Copilot/Copilot Studio included
3	Confirm SIT readiness (Control 1.13)	SITs exist and match sanitized test data
4	Create sanitized test content + apply labels	Files/messages prepared; labels applied
5	Run AI test prompts and agent scenarios	Blocked/warned/audited per tier
6	(Optional) Run Endpoint DLP tests	Endpoint actions enforce as configured
7	Check DSPM for AI → Policies	AI-related DLP policies visible (if DSPM enabled)
8	Check audit logs / incident reports	DLP events and reports captured

Test cases (Copilot/agent; U.S.-only)

Use non-production, sanitized data. Do not use real customer data.

Test ID	Scenario	Input	Expected
AI-01	Prompt includes SSN	Prompt text includes an SSN-formatted value	Tier 2: user warning + log; Tier 3: block + incident report
AI-02	Prompt includes ABA routing	Prompt includes a routing-formatted value	Same as AI-01
AI-03	Retrieval from labeled content	Agent grounds on a SharePoint file labeled Highly Confidential	Block (per label-based rule); event logged
AI-04	Retrieval from Confidential content	Agent grounds on Confidential content	Warn or allow-with-audit (per policy); event logged
AI-05	Attempted sensitive output	Ask agent to summarize a document containing U.S. bank account numbers	Output blocked/redacted as configured; event logged

Test cases (Endpoint DLP; optional)

Run only if Devices/Endpoint DLP is in scope.

Test ID	Scenario	Action	Expected
EP-01	Copy/paste exfiltration	Copy text containing SSN and paste into an AI prompt in a browser	Block or warn per endpoint rule; event logged
EP-02	File exfiltration	Attempt to upload a Highly Confidential labeled file to an AI web experience	Block or warn per endpoint rule; event logged
EP-03	Removable media	Copy a file containing U.S. bank account numbers to USB	Block or warn per endpoint rule; event logged

Evidence to retain (audit-ready)

[ ] DLP policy export (policy + rules) and a change record (ticket or approval)
[ ] Screenshot evidence showing AI locations selected and rule conditions (SIT/label)
[ ] SIT validation evidence (sanitized test data + outcomes) per Control 1.13
[ ] Label taxonomy decision + label publication configuration (who the labels are published to)
[ ] Test execution log: test ID, timestamp, account used, expected vs actual outcome
[ ] Incident report samples (if enabled) and notification configuration
[ ] Unified audit log evidence for representative DLP events (export/query output)
[ ] DSPM for AI policy visibility and oversharing assessment outputs (if DSPM is enabled)

Troubleshooting & Validation

Issue: DLP Policy Not Detecting Sensitive Content

Symptoms: Sensitive data flows through AI without triggering DLP alerts

Solutions:

Verify policy is in "Enable" mode (not "Test" mode)
Confirm policy locations include the intended AI locations (Microsoft 365 Copilot / Copilot Studio)
Confirm SIT readiness and scope (see Control 1.13):
- The SIT exists and is enabled
- The SIT pattern actually matches your test data
- Minimum count/confidence thresholds are not too strict
Confirm you are testing a supported content path:
- For label-based rules, confirm the label is applied to the item (file/message), not only a container
- For endpoint tests, confirm Devices/Endpoint DLP is enabled and targeted to the test device
Allow for propagation time before re-testing (often hours; in some cases longer)

Issue: False negatives in AI prompts/responses

Symptoms: Some prompts/responses containing sensitive patterns are not detected.

Solutions:

Reduce reliance on a single pattern: add supporting keywords or additional SITs where appropriate
Validate formatting variations (spaces/hyphens/prefixes) in SIT patterns per Control 1.13
Review rule logic (AND/OR) and priority; ensure an allow rule is not bypassing detection
Confirm the AI location is included (a rule scoped to a different workload will not trigger)

Issue: Too Many False Positive DLP Alerts

Symptoms: Legitimate business content triggers DLP blocks

Solutions:

Review SIT confidence levels and minimum count; increase thresholds where justified
Add corroborating context (supporting keywords) to custom SITs and tighten regex patterns (Control 1.13)
Add scoped exceptions only when you can justify and evidence them (specific locations, groups, or workflows)
Prefer "Warn" + audit for ambiguous detections; reserve "Block" for high-confidence or labeled High impact content
Re-run the test suite after tuning and retain evidence of the before/after results

Issue: Sensitivity Labels Not Enforcing in AI

Symptoms: Content with "Highly Confidential" label still accessible to agents

Solutions:

Verify DLP rule includes label-based conditions
Check label is correctly applied to content (not just container)
Confirm label policy is published to affected users
Verify DSPM for AI integration is enabled
Check if agent service account is in label scope

Issue: DSPM Oversharing Assessment Shows No Results

Symptoms: Assessment completes but shows zero overshared items

Solutions:

Verify data sources are correctly specified
Check that content has sensitivity labels applied
Confirm assessment scope includes the correct sites
Wait for assessment processing (can take 24-48 hours)
Verify permissions to access assessment results

Issue: DLP Blocking Legitimate Agent Operations

Symptoms: Agent cannot access required data due to DLP blocks

Solutions:

Review incident details to understand what triggered block
Create exception for agent service account if appropriate
Adjust SIT minimum count thresholds
Use contextual conditions to allow specific scenarios
Consider separate policy for agent service accounts

Additional Resources

Control	Relationship
Control 1.6: DSPM for AI	AI monitoring and assessment
Control 1.13: Sensitive Information Types	SIT configuration
Control 1.3: SharePoint Content Governance	Knowledge source protection
Control 4.1: SharePoint IAG	Content discovery

Support & Questions

For implementation support or questions about this control, contact:

Microsoft Purview Administrator: DLP policy configuration
Information Protection Team: Sensitivity label design
Compliance Officer: Regulatory requirements and evidence
AI Governance Lead: Agent data protection strategy

Updated: Dec 2025
Version: v1.0 Beta (Dec 2025)
UI Verification Status: ❌ Needs verification

Control 1.5: Data Loss Prevention (DLP) and Sensitivity Labels

Overview

Purpose

Description

US-only scope note

Dependency: Sensitive Information Types (Control 1.13)

DLP Enforcement Status Update (2025)

Copilot Studio Channel Connectors (6 Total)

Key Capabilities

DLP for AI Applications

AI-Specific DLP Policies

Configuring AI DLP Policies

AI DLP setup steps (Copilot prompts/responses)

DSPM for AI Integration

Viewing DLP Policies in DSPM

Example AI DLP Policies

Oversharing Assessments

What is Oversharing?

DSPM Oversharing Detection

Custom Oversharing Assessments

Sensitivity Labels

Label-Based Protection for Agents

Recommended label taxonomy (U.S.-only)

Configuring Label-Based DLP

Agent Knowledge Source Labels

Prerequisites

Licenses Required

Permissions Required

Dependencies

Pre-Setup Checklist

Governance Levels

Level 1 - Baseline

Level 2-3 - Recommended

Level 4 - Regulated/High-Risk

Setup & Configuration

Governance Tier DLP Configuration

SIT foundation (Dependency: Control 1.13)

Create an AI-scoped DLP policy (Purview UI)

Extend to endpoints (optional; avoid overreach)

Policy Templates for FSI

Customer Data Protection

Financial Data Protection

PowerShell Configuration

Connect to Security & Compliance Center

Create AI-Focused DLP Policy

Create Sensitivity Label-Based DLP Rule

Audit DLP Policies

Monitor DLP Alerts

Get Sensitivity Label Statistics

Financial Sector Considerations

Regulatory Mapping

Governance Tier DLP Configuration

FSI-Specific Sensitive Information Types

FSI Example: Trading Desk Agent Protection

Regulatory Context

Examination Considerations

Zone-Specific Configuration

Verification & Testing

Test cases (Copilot/agent; U.S.-only)

Test cases (Endpoint DLP; optional)

Evidence to retain (audit-ready)

Troubleshooting & Validation

Issue: DLP Policy Not Detecting Sensitive Content

Issue: False negatives in AI prompts/responses

Issue: Too Many False Positive DLP Alerts

Issue: Sensitivity Labels Not Enforcing in AI

Issue: DSPM Oversharing Assessment Shows No Results

Issue: DLP Blocking Legitimate Agent Operations

Additional Resources

Related Controls

Support & Questions