Skip to content

Control 2.2: Sensitivity Labels and Copilot Content Classification

Control ID: 2.2 Pillar: Security & Protection Regulatory Reference: GLBA §501(b), Sarbanes-Oxley §§302/404 Last Verified: 2026-09-29 Governance Levels: Baseline / Recommended / Regulated


Objective

Implement a comprehensive sensitivity label taxonomy and auto-labeling strategy that governs how Microsoft 365 Copilot interacts with classified content. Sensitivity labels, effective user rights, DLP, and supported source/surface behavior together influence what Copilot can use, how Copilot-generated content is classified, and how label-based protections cascade to new content surfaces such as Copilot Pages. This control supports compliance with GLBA data safeguard requirements and helps meet SOX internal control obligations for information classification.

Verification note — 2026-09-29

This control and its four implementation playbooks were revalidated against current Microsoft Learn guidance for SharePoint default-library labels on existing files, auto-labeling simulation and coverage-review behavior, current Purview reporting surfaces, and current portal navigation.


Why This Matters for FSI

  • GLBA §501(b) requires financial institutions to develop, implement, and maintain a comprehensive information security program — sensitivity labels are a foundational technical control for classifying and protecting customer information across all M365 workloads, including Copilot
  • Sarbanes-Oxley §§302/404 mandates that management assess the effectiveness of internal controls over financial reporting — sensitivity labels create auditable, persistent classification markers that demonstrate information handling controls
  • SEC Reg S-P (Rule 248.30) requires safeguarding customer records — label-based encryption and access restrictions help prevent unauthorized access to customer information via Copilot summarization
  • FFIEC IT Examination Handbook expects data classification programs that identify and protect sensitive information — labels provide the classification layer that enables downstream DLP, access control, and retention policies
  • SR 11-7 / OCC Bulletin 2011-12 expects risk management practices that include data governance — labels enable consistent data handling across human and AI-assisted workflows

Control Description

Microsoft Purview sensitivity labels provide persistent, metadata-based classification that travels with content across M365. When Copilot accesses labeled content, the label influences:

  1. Whether Copilot can access the content at all (label-based DLP blocking)
  2. What protections are applied to Copilot-generated output (label inheritance)
  3. How Copilot Pages inherit labels from source content
  4. What default label is applied to newly created content

EXTRACT Permission

EXTRACT (shown to users as Copy) determines whether Copilot and agents can display text from encrypted content for the requesting user. When a user has VIEW but not EXTRACT, Microsoft documents that Copilot does not summarize the item but can return a link for the user to open outside Copilot. This is not a categorical encrypted-content block: Full Control (OWNER) includes EXTRACT, and the person who applies encryption is the Rights Management owner. Unopened SharePoint/OneDrive items encrypted with user-defined permissions, open data-in-use, active Edge tabs without Edge DLP, and external plugin/Graph connector sources have documented exceptions or limitations. See also the BlockContentAnalysisServices PowerShell advanced label setting, applied with Set-Label or New-Label, which stops Office apps from sending labeled content to connected experiences that analyze content, including Microsoft 365 Copilot and agents.

Label Taxonomy Structure: Label Groups

Microsoft Purview has replaced the parent/child label hierarchy with label groups as the modern label scheme (Microsoft 365 Roadmap ID 386900, general availability December 2025). Label groups are an organizational container that replaces the parent label concept. Migration is rolling out gradually and might not yet be available in a given tenant; it is automatic for tenants created on or after October 1, 2025, and for tenants where migrating parent labels would not create a new sublabel of the same name.

  • Migration path: Microsoft Purview portal > Solutions > Information Protection > Sensitivity labels — when the Migrate to the modern label scheme information banner appears, select Get started. Migration typically takes minutes to an hour; allow up to 24 hours
  • Migration is irreversible. Microsoft recommends testing first in a tenant that has the same sensitivity label configuration as production
  • What changes: Label groups support only name, description, color, and priority — they carry no encryption, content marking, or auto-labeling settings. Where a parent label was itself configured for label actions, published separately from its sublabels, or scoped differently, migration also creates a new sublabel with the same name, GUID, and settings, so existing services and configurations continue to resolve the label
  • Copilot impact: Label groups affect how DLP policies reference label hierarchies and how Copilot evaluates label-based access decisions. Verify that label-based DLP policies (see Control 2.1, Type 1) continue to function correctly after migration by testing with representative labeled content
  • Terminology: Use "label groups" when referring to the container; the individual labels remain "sensitivity labels"
Label Group Label Description Copilot Behavior Protection
Public Public Information approved for external distribution Full access None
Internal Internal — General Standard business information Full access Header/footer marking
Internal Internal — Employee Only HR, compensation, internal memos Full access Header/footer + no external sharing
Confidential Confidential — Standard Client data, business strategies Full access with DLP monitoring Encryption (authorized users)
Confidential Confidential — MNPI Material non-public information Restricted — DLP alerts on access Encryption + information barriers
Highly Confidential Highly Confidential — Regulatory Examination correspondence, SAR-related Blocked from Copilot processing Encryption + DKE + no forwarding
Highly Confidential Highly Confidential — Board/Executive Board materials, M&A documents Blocked from Copilot processing Encryption + DKE + scoped access

Governance tier recommendations for label groups migration: - Baseline: Plan migration timeline; inventory existing parent labels before migrating; document which DLP policies reference parent/sub-label conditions - Recommended: Migrate to label groups; test all Copilot DLP policies against the new label group structure before completing migration; update any DLP policy conditions that reference old parent label names - Regulated: Complete migration with documented validation that all DLP policies function correctly with label groups before moving to production; maintain pre-migration label taxonomy documentation for audit trail

Label Inheritance in Copilot

When Copilot generates content based on multiple source documents, label inheritance selects the highest-priority label (typically the most restrictive). Copilot in Word, Copilot in PowerPoint, and Copilot in Outlook support sensitivity label inheritance for newly created content, and the inherited label carries the source label's protection settings:

Source Document A (Confidential) ─┐
                                   ├─→ Copilot Output → Confidential (highest priority wins)
Source Document B (Internal)     ─┘

Source Document A (Highly Confidential) ─┐
                                          ├─→ Copilot does not summarize the HC source
Source Document B (Confidential)        ─┘   (label-based DLP for the Copilot location)

Blocking is not automatic from the label alone. Copilot and agents stop summarizing labeled content when a DLP policy scoped to the Microsoft 365 Copilot location identifies it, or when the label's encryption withholds EXTRACT from the requesting user. In both cases Copilot can still return a link to the item. Test the documented owner, data-in-use, Edge, and external-source cases before representing either mechanism as complete coverage.

Copilot Studio Agent Label Behavior

Microsoft Purview sensitivity labels in Copilot Studio are in preview and are on by default for agents with supported knowledge sources. Supported knowledge sources include SharePoint, OneDrive for Business, SQL, Dataverse, Cosmos, Azure Blob Storage, Word, Excel, Outlook, Office groups, and Office users.

  • How it works: An agent response displays a shield icon carrying the highest-priority sensitivity label applied to the content the agent used to generate that response, and each cited file shows its own label. The label is surfaced per response — Microsoft does not assign a persistent "effective label" to the agent object itself
  • Encryption scope: Label-applied encryption is honored for the SharePoint knowledge source. For that supported source, Copilot Studio agents return encrypted-item text only where the requesting user holds EXTRACT; validate each additional knowledge-source type rather than assuming external-source encryption is recognized
  • Copilot governance impact: Because responses can surface high-sensitivity labels drawn from connected knowledge, label-based DLP policies (Control 2.1, Type 1) can engage on agent interactions even when the user's prompt does not reference sensitive content directly
  • Agent deployment review: Before activating a Copilot Studio agent, review the sensitivity labels on all configured knowledge sources. The highest label reachable through those sources should be documented as part of the deployment approval process

Governance tier recommendations for agent label review: - Baseline: Audit all Copilot Studio agent knowledge source labels; document the highest label reachable through each deployed agent - Recommended: Enforce knowledge source label review as part of the agent deployment approval workflow; agents that can surface Highly Confidential labels require compliance sign-off before activation - Regulated: Mandatory knowledge source label assessment with documented compliance approval before agent activation; quarterly re-audit of deployed agents as knowledge sources are added or changed

Copilot Pages Label Behavior

Copilot Pages are .page files stored in a user-owned SharePoint Embedded container and are labeled through the Microsoft Loop labeling surface. Applying labels to Copilot Pages requires that SharePoint and OneDrive are enabled for sensitivity labels, and that labels are published with the Files & other data assets scope. Label behavior for Pages:

Scenario Label Applied to Page Notes
Page created from Microsoft Copilot Chat response Inherits the highest-priority label from the referenced labeled files As with all automatic labeling, a user can replace the inherited label unless mandatory labeling and downgrade justification are configured
Unlabeled components placed on a labeled page Inherit the page's label The label is retained if the component moves to another page
Page shared with additional users Label protections enforced Users without label permissions cannot access
No labeled sources used Default document label from the label policy applied Configurable — recommend "Internal" minimum

Labeling capabilities that are not supported for Loop components and Copilot Pages: automatic and recommended labeling, headers and footers, watermark variables and dynamic watermarks, and labels configured for user-defined permissions, content expiry, or Double Key Encryption.

Auto-Labeling Configuration

Auto-labeling applies sensitivity labels automatically based on content inspection. Service-side auto-labeling policies for SharePoint, OneDrive, and Exchange support nested rule logic (AND/OR/NOT): as in DLP policy design, conditions can be arranged into groups with logical operators between the groups. The client-side label setting auto-labeling for files and emails does not support nested rule logic — use an auto-labeling policy where combined condition logic is required.

For FSI classification, the nested condition logic enables rules such as: "Apply Confidential IF (contains SSN AND contains account number) OR (contains MNPI keyword) NOT (from public disclosure folder)." This expresses complex financial data classification requirements that single-condition rules cannot.

Secure by Default Label Derivation

Microsoft's Secure by Default deployment model provides an alternative approach to sensitivity labeling at scale by deriving file labels from site (container) labels rather than relying on manual user labeling or auto-labeling alone:

  • Default label for a document library: Container labels applied to sites, Teams, and Microsoft 365 Groups are not inherited by the items inside them. The mechanism that derives file labels at scale is the default sensitivity label for a SharePoint document library — location-based labeling that applies the library's label to new files uploaded to the library and to existing files edited there when they have no label or a lower-priority label. Files that were already in the library before the default label was configured can remain unlabeled until they are edited or separately processed. Unlike the default label from a label policy, a library default label overrides a lower-priority label unless that label was applied manually
  • Scale advantage: Organizations can reach high labeling volumes quickly by setting library defaults across Teams and SharePoint sites, reducing reliance on manual user labeling and per-file SIT inspection
  • User training shift: Rather than training users on when to apply labels, this model trains users on how to request exceptions when the derived label is too restrictive for a specific use case
  • Copilot governance impact: Because Copilot cannot read a container label from an item, file-level labels are what Copilot displays and inherits. Combining library default labels with sensitivity labels for Teams, Groups, and sites — which govern site privacy and sharing — helps establish a baseline where Copilot grounding respects classification without requiring users to label every file individually

FSI consideration: The Secure by Default model is particularly useful for large financial institutions with thousands of SharePoint sites, where achieving comprehensive file-level labeling through auto-labeling or manual labeling alone may take extended time. Organizations should evaluate this deployment model alongside the auto-labeling approach documented above and choose the strategy — or combination of strategies — that best fits their tenant scale and classification maturity.

Applying library defaults to existing files at rest: When a SharePoint document library already has a default sensitivity label and you want that label applied to eligible files that predate the configuration, create the dedicated auto-labeling policy that uses the default library label rather than content-inspection conditions. Simulation for this policy type identifies the SharePoint sites, libraries, default labels, and files in scope before enforcement. Microsoft currently supports one policy of this type per tenant, with up to 10 SharePoint sites per run; rotate the site selection in batches if you need broader coverage.

Auto-Label Trigger Target Label SIT/Condition Scope
SSN detected (high confidence) Confidential — Standard U.S. Social Security Number All workloads
5+ SSNs detected Highly Confidential — Regulatory U.S. SSN (instance count >= 5) SharePoint, OneDrive
ABA routing + account number combo Confidential — Standard ABA Routing AND Bank Account (nested AND) Exchange, SharePoint
MNPI keyword match Confidential — MNPI Custom MNPI keyword dictionary All workloads
SAR/CTR keywords Highly Confidential — Regulatory Custom BSA/AML dictionary Exchange, SharePoint
Contains CUSIP/ISIN AND context keyword, NOT in public disclosure folder Confidential — MNPI Nested AND/OR/NOT condition SharePoint, OneDrive

Configuring nested auto-labeling conditions: In Microsoft Purview portal > Solutions > Information Protection > Policies > Auto-labeling policies > + Create auto-labeling policy, the condition builder supports grouped conditions with AND/OR logic and NOT exclusions. Nest conditions by creating condition groups and selecting the operator between groups, the same way DLP complex rule design works. This is part of the existing auto-labeling policy workflow — not a separate feature.

Auto-Labeling Override for Manually Applied Labels (April 2026)

Auto-labeling policies can override manually applied labels for files in SharePoint and OneDrive (Microsoft 365 Roadmap ID 558342, general availability April 2026, covering Word, Excel, PowerPoint, and PDF files). Before this enhancement, the override option applied to emails only.

  • What changed: An auto-labeling policy can replace a manually applied label with the policy's label when file content matches the policy conditions. Override applies only when the policy's label has a higher priority than the existing label — a manually applied higher-priority label is never overridden. The same release also allows auto-labeling policies to remove a label from files in SharePoint and OneDrive
  • FSI impact: This closes a governance gap where users could manually label a document containing MNPI or client PII with a lower classification (e.g., "Internal") and the auto-labeling engine would not correct it. Auto-labeling rules that detect high-confidence SIT matches (such as bulk SSNs or MNPI keyword patterns) can now reclassify those files
  • Configuration: In the auto-labeling policy wizard, on the Additional label settings page, select All locations (emails plus files in SharePoint and OneDrive). Selecting Emails only limits the override to email
  • Caution: Organizations should verify that auto-labeling rules have a low false positive rate (<3%) before enabling override for manually applied labels, since incorrect auto-labeling upgrades may disrupt user access to files via encryption changes. Run in simulation mode first. Where one active policy is configured to replace a manually applied label and another is not, the label is not replaced

Permission Level Renames

Microsoft updated the sensitivity label permission level names shown in the Microsoft Purview portal and in the custom permissions dialog in Word, Excel, and PowerPoint for Windows (version 2411 and later). Other applications continue to use the original naming.

Previous Name New Name Usage rights included
Reviewer Restricted Editor View, Open, Read; Save; Edit Content, Edit; View Rights; Reply; Reply All; Forward; Allow Macros — no Copy, Print, or Save As/Export
Co-Author Editor Restricted Editor rights plus Copy and Print. Save As, Export is not included when the level is configured in the Microsoft Purview portal
Co-Owner Owner All usage rights, including Change Rights and Full Control
  • FSI impact: Organizations that relied on the Co-Author level to let users save copies or export labeled content should review their label configurations. Where the level is configured from the Microsoft Purview portal, Editor does not grant Save As/Export, which may affect workflows that create derivative documents from labeled sources
  • Action required: Audit existing sensitivity labels that use the Co-Author (now Editor) permission level and verify that the available Save As/Export behavior aligns with the intended data handling. If Save As/Export is required, assign it through custom permissions

Default Label Policy

Workload Default Label Justification Required to Downgrade
Word, Excel, PowerPoint Internal — General Yes
Outlook (email) Internal — General Yes
Teams meetings Internal — General No (may block adoption) — a label policy setting applies a default label to meetings and calendar events
Copilot Pages Internal — General Yes
Power BI Internal — General Yes

Copilot Surface Coverage

M365 Application Label Inheritance Auto-Labeling Default Label Label-Based Blocking
Microsoft Copilot Chat Yes N/A N/A Yes
Word Yes Yes Yes Yes
Excel Yes Yes Yes Yes
PowerPoint Yes Yes Yes Yes
Outlook Yes Yes Yes Yes
Teams Yes Limited Yes — a label policy can apply a default label to meetings and calendar events Yes
OneNote Yes Limited No Yes
Loop Yes No — automatic and recommended labeling are not supported for components and pages Yes Yes
Copilot Pages Yes No — automatic and recommended labeling are not supported for components and pages Yes Yes
SharePoint (Agents) Yes Yes N/A Yes
Copilot Studio Agents Response shows the highest-priority label from the content used (preview) N/A N/A Yes

Governance Levels

Level Requirement Rationale
Baseline Deploy 4-tier label taxonomy using label groups (Public, Internal, Confidential, Highly Confidential); configure default label of "Internal" for Office apps; enable label inheritance for Copilot-generated content; plan label groups migration timeline; audit Copilot Studio agent knowledge source labels Establishes classification foundation — minimum viable protection for Copilot deployment; migration awareness prevents unexpected DLP behavior changes
Recommended Add sub-labels (MNPI, Regulatory, Board/Executive) within label groups; enable auto-labeling policies with nested AND/OR/NOT condition groups for FSI classification rules; configure label-based DLP to block Copilot from Highly Confidential content; enforce justification for label downgrades; migrate to label groups and test Copilot DLP policies against new structure; require agent knowledge source label review as part of agent deployment approval; enable DSPM for AI reporting; review auto-labeling coverage and failures with the policy review surfaces after simulation and enforcement Comprehensive label strategy with agent governance — suitable for most FSI production environments; nested auto-labeling reduces manual labeling burden for complex financial data patterns
Regulated Mandate labeling (no unlabeled content allowed); deploy auto-labeling policies with nested conditions for all FSI SIT categories; enable double key encryption (DKE) for Highly Confidential sub-labels; complete label groups migration with full DLP policy validation before production; mandatory agent knowledge source label assessment with compliance sign-off; quarterly label accuracy audits; integrate label telemetry with SIEM; DSPM for AI reporting reviewed monthly by compliance; review Information Protection reports monthly and use Content Explorer exports for current labeled-item evidence, supplemented by auto-labeling policy review pages, Activity Explorer, or audit evidence for encrypted SharePoint and OneDrive items that Content Explorer doesn't surface Maximum classification rigor for firms facing frequent regulatory examinations or handling the most sensitive financial data

Setup & Configuration

Step 1: Create Label Taxonomy Using Label Groups

Portal: Microsoft Purview portal > Solutions > Information Protection > Sensitivity labels

  1. Select + Create > Label group for each group: Public, Internal, Confidential, Highly Confidential. A label group carries only a name, display name, priority, description, and color
  2. Select a label group, then ... (Actions) > Create label in group to create labels within each group per the taxonomy table above. Existing standalone labels can be moved with ... > Move into group
  3. If migrating from the parent/child hierarchy: on the Sensitivity labels page, use the Migrate to the modern label scheme banner > Get started. Review the proposed scheme, resolve naming conflicts, then select Migrate. Migration is irreversible — test in a non-production tenant first
  4. Configure visual markings (headers, footers, watermarks) for each label
  5. Configure encryption settings for Confidential and above

Step 2: Configure Label Policies

Portal: Microsoft Purview portal > Solutions > Information Protection > Publishing policies > Publish label

  1. Publish all labels to the appropriate user groups
  2. Set default label to "Internal — General" for Word, Excel, PowerPoint, Outlook
  3. Enable "Require users to apply a label" (Recommended and Regulated levels)
  4. Enable "Require justification for removing a label or lowering its classification"
  5. To cover meetings, select the policy settings for calendar items and Teams meetings — Apply a default label to meetings and calendar events and, where appropriate, Inherit label from files shared to meetings

Step 3: Configure Auto-Labeling Policies with Nested Conditions

Portal: Microsoft Purview portal > Solutions > Information Protection > Policies > Auto-labeling policies > + Create auto-labeling policy

  1. Create auto-labeling rules per the auto-labeling configuration table above
  2. For nested conditions: use the condition builder to create condition groups with logical operators between groups
  3. On the Additional label settings page, choose All locations where the policy should override manually applied lower-priority labels on SharePoint and OneDrive files
  4. New policies save in simulation mode — run simulation for 14 days minimum before enforcement
  5. Review simulation results for false positive rate and note that simulation evaluates a single policy, so overlapping active policies can still change the enforced outcome
  6. If you use SharePoint document-library default labels and need them applied to preexisting files at rest, create the dedicated auto-labeling policy for default-library labels rather than rebuilding those locations as content-inspection rules
  7. Enable enforcement after validation (target <3% false positive rate)
  8. After enforcement, review Coverage by simulation context plus the policy's Labeled items and Failed views to confirm which files were labeled, are still pending, or could not be labeled

Step 4: Configure Label-Based Copilot Blocking

  1. In DLP policy configuration (see Control 2.1, Type 1), add condition: "Content contains sensitivity label"
  2. Select "Highly Confidential" (all labels in the Highly Confidential label group)
  3. Scope the policy to the Microsoft 365 Copilot location and set the action that stops Copilot from summarizing matched content
  4. Copilot and agents then do not summarize Highly Confidential content, though they can still return a link to the item for the user to open outside Copilot
  5. After label groups migration: verify the DLP policy condition correctly references labels by their new group structure

Step 5: Review Copilot Studio Agent Knowledge Source Labels

Portal: Microsoft 365 admin center > Agents > All agents (Agent inventory tab, within the Copilot Control System)

For each deployed Copilot Studio agent: 1. Review all configured knowledge sources and their sensitivity labels 2. Determine the highest-priority label reachable through those knowledge sources — this is the label an agent response can display 3. Document that label in the agent deployment record 4. Confirm that DLP policies covering that label tier are in place before activating the agent

Step 6: Copilot Pages Label Settings

Copilot Pages label behavior is configured through Microsoft Purview, not through a dedicated Pages labeling page:

  1. Enable sensitivity labels for Office files in SharePoint and OneDrive — a prerequisite for labeling Copilot Pages
  2. Publish labels with the Files & other data assets scope so users can apply them to Pages, and set the default document label in the label policy for Pages created without labeled sources
  3. Exclude labels configured for user-defined permissions, content expiry, or Double Key Encryption from the Pages scope — those settings are not supported for Loop components and pages
  4. Control whether users can create Copilot Pages and Copilot Notebooks with the Cloud Policy setting Create and view Copilot Pages and Copilot Notebooks at https://config.office.com/ > Customization > Policy Management

Key PowerShell Commands

# Connect to Security & Compliance PowerShell
Connect-IPPSSession

# List all sensitivity labels (includes label group membership)
Get-Label | Format-Table Name, DisplayName, Priority, ContentType, ParentId

# List label policies
Get-LabelPolicy | Format-Table Name, Labels, Settings

# View auto-labeling policies
Get-AutoSensitivityLabelPolicy | Format-Table Name, Mode, WhenCreated

# Check label activity for a specific user.
# Service-side events (Office for the web, SharePoint details pane, Teams Files tab,
# auto-labeling policies) use the File* operations; Microsoft 365 apps use the others.
Search-UnifiedAuditLog -StartDate (Get-Date).AddDays(-7) -EndDate (Get-Date) `
  -Operations "FileSensitivityLabelApplied","SensitivityLabelApplied", `
              "FileSensitivityLabelChanged","SensitivityLabelUpdated" `
  -UserIds "user@firm.com"

Financial Sector Considerations

  • Broker-Dealers: The MNPI label within the Confidential label group is critical for firms with investment banking and trading operations. When Copilot encounters MNPI-labeled content, it should be blocked from including that content in responses to users who may be on the public side of the wall. Coordinate MNPI labeling with the firm's restricted list procedures. When migrating to label groups, verify that MNPI label conditions in DLP policies continue to function correctly.
  • Registered Investment Advisers: Client portfolio documents and investment recommendations should be labeled at minimum "Confidential — Standard." Auto-labeling rules with nested conditions should detect client account number patterns specific to the firm's account numbering scheme combined with context keywords (AND logic), reducing false positives compared to single-SIT rules.
  • Banking (Deposit-Taking): Suspicious Activity Report (SAR) documentation and Bank Secrecy Act (BSA) files must be labeled "Highly Confidential — Regulatory" to prevent any Copilot access. The "tipping off" prohibition under 31 USC 5318(g)(2) makes it critical that SAR-related content is never surfaced by Copilot to unauthorized individuals. Use nested auto-labeling conditions combining SAR/CTR keywords with BSA context to accurately classify these documents without over-labeling.
  • Insurance Carriers: Policy documents containing personal health information may need label protections aligned with both GLBA and state insurance data security laws. Consider a label for "PHI-Insurance" within the Confidential label group.
  • Copilot Studio Agent Governance: Financial institutions building Copilot Studio agents that query internal data (loan origination systems, client databases, research repositories) should assess knowledge source labels before deployment. An agent connected to a research database labeled "Confidential — MNPI" can surface that label on its responses and engage the matching DLP policies — this is correct behavior, but it must be explicitly planned and documented.
  • Examination Readiness: Maintain documentation showing label taxonomy design decisions, label groups migration records, deployment dates, auto-labeling accuracy metrics, and periodic review records. Examiners may ask for evidence that the firm's data classification program extends to AI-assisted workflows and that agent deployments include label governance.
  • Label Sprawl Prevention: Avoid creating more than 8-10 labels total. Financial firms often want dozens of sub-labels — this creates user confusion and reduces labeling accuracy. Consolidate where possible using label groups and use DLP policies for fine-grained enforcement within label tiers.

Verification Criteria

  1. Label Taxonomy Deployed: Confirm all required labels (Public, Internal, Confidential, Highly Confidential) are published and visible to users in Office applications
  2. Label Groups Migration Status: Verify current taxonomy structure — confirm whether migration from parent labels to label groups has been completed or is scheduled
  3. Default Label Active: Open a new Word document and verify the "Internal — General" label is automatically applied
  4. Label Inheritance: Use Copilot to summarize a Confidential-labeled document and verify the output inherits the "Confidential" label
  5. Highly Confidential Blocking: Attempt to reference a Highly Confidential document in a Copilot prompt — verify Copilot does not summarize the content (a link to the item may still be returned) and a DLP notification is generated
  6. Agent Knowledge Source Labels: For at least one deployed Copilot Studio agent, verify the label shown on an agent response matches the highest-priority label across the content the agent used
  7. Auto-Labeling Accuracy: Review auto-labeling simulation results and confirm <3% false positive rate for deployed SIT-based rules; after enforcement, review Coverage by simulation context plus the policy's Labeled items and Failed views to verify nested condition groups function as designed
  8. Downgrade Justification: Attempt to change a label from Confidential to Internal — verify the system requires justification and logs the event
  9. Default-Library Existing Files (if used): Run the dedicated auto-labeling policy for SharePoint library default labels in simulation, then turn it on and verify the expected existing files in scope are labeled while unsupported, open, or unavailable files are surfaced through the policy review experience
  10. Copilot Pages Label: Create a Copilot Page from content that references a Confidential document — verify the Page inherits the Confidential label
  11. Visual Markings: Open a Confidential-labeled document and confirm headers/footers/watermarks appear as configured
  12. Audit Trail: Verify that label application and modification events appear in the Unified Audit Log under both the service-side (FileSensitivityLabelApplied, FileSensitivityLabelChanged) and Microsoft 365 apps (SensitivityLabelApplied, SensitivityLabelUpdated) operations
  13. Policy Documentation: Confirm that label taxonomy documentation, label groups migration records, deployment records, and review cadence are maintained and accessible for examination

Additional Resources