Microsoft 365 Copilot PDF guides
Full Markdown conversions of the two supplied PDFs, reviewed on 7 September 2026.
| Guide | Source pages | Source version |
|---|---|---|
| Secure and Govern Microsoft 365 Copilot | 22 | Printed “Last updated March 6, 2026” |
| Microsoft Copilot for Microsoft 365 architecture & deployment | 7 | November 2023 footers; March 14, 2024 PDF metadata |
Each guide follows the PDF’s page order and retains its guidance, examples, outcomes, tables, diagram labels, and resource links. The blueprint includes the full E3 and E5 deployment checklists and their source entitlement markers. The architecture guide includes four Mermaid diagrams, detailed diagram transcriptions, the E3/E5 protection table, and the staged deployment table.
These are conversions of the supplied versions. Product names, licensing statements, indexing behavior, and other dated claims remain as the PDFs state them. Interpretation and source anomalies are recorded separately in Editorial notes.
Sources and link register · Verification and page coverage
The source PDFs, deployment-model collection, and earlier summaries are preserved.
Microsoft Copilot for Microsoft 365 architecture & deployment
Full Markdown conversion of the supplied PDF, reviewed on 7 September 2026. Source version: November 2023 (printed footer); PDF metadata dated March 14, 2024.
Original PDF · Conversion notes · Sources and links · Verification
This document preserves the source’s guidance as published. Page numbers correspond to the supplied PDF. Diagram descriptions and conversion notes are identified explicitly.
Contents
- Page 1 — Microsoft Copilot for Microsoft 365 architecture & deployment
- Page 2 — Microsoft Copilot for Microsoft 365 service and tenant logical architecture
- Page 3 — Semantic Index for Microsoft Copilot for Microsoft 365
- Page 4 — Microsoft Copilot for Microsoft 365 query flow
- Page 5 — Prepare your environment for Microsoft Copilot for Microsoft 365
- Page 6 — Set up secure file sharing and collaboration with Microsoft Teams
- Page 7 — Pilot and deploy Microsoft Copilot for Microsoft 365
Page 1 — Microsoft Copilot for Microsoft 365 architecture & deployment
Source: PDF page 1
Source topic marker: This topic is 1 of 7.
Copilot combines the power of large language models (LLMs) with your data in the Microsoft Graph — your calendar, emails, chats, documents, meetings, and more — and the Microsoft 365 apps to provide a powerful productivity tool.
Microsoft Copilot for Microsoft 365 logical architecture
Microsoft Copilot for Microsoft 365 or Copilot introduces several components to help users make use of content and data they already have access to. Note that only data a user has access to is returned in query responses (as illustrated).
Diagram transcription — layers and labels
| Layer | Components and labels in the source |
|---|---|
| Users and devices | User, computer and mobile-device icons |
| Microsoft 365 apps | Teams, Word, PowerPoint, Excel and Outlook icons |
| Copilot and related components | Copilot for Microsoft 365; Microsoft Graph; Semantic Index for Copilot |
| Semantic Index for Copilot | Maps your data and relationships. Provides personalized, relevant, and actionable information |
| Microsoft Graph illustration | Tasks, Teams, Groups, Calendar, Files, Messages, Meetings, People, Devices, Coworkers, Insights, Chats, User; central “MICROSOFT GRAPH” label |
| Organization data (Example data): OneDrive | User folders and files; Recordings |
| Organization data (Example data): SharePoint | Communication sites; Team sites |
| Organization data (Example data): Exchange Online | Group mailboxes; User mailboxes |
| Organization data (Example data): Teams Chat | Viva Engage groups (the label printed inside this box) |
Diagram transcription — relationships. A purple path connects the user through an app and Copilot to Microsoft Graph, with a branch to Semantic Index for Copilot. It continues to selected organization-data objects. Purple outlines select one OneDrive file and its Recordings folder, a column containing two SharePoint team-site tiles, two group mailboxes, one user mailbox, and one conversation/group icon. Other objects remain unselected. These selections illustrate the permission-limited response described above; they are examples, not quotas.
Copilot uses permission-limited organization data
Scroll sideways to read the full diagram →
View diagram source
flowchart TB
U["Users and devices"] --- Apps["Microsoft 365 apps: Teams, Word, PowerPoint, Excel, Outlook"]
Apps --- C["Copilot for Microsoft 365"]
C --- G["Microsoft Graph"]
G --- S["Semantic Index for Copilot"]
G --- O["Organization data — selected accessible examples"]
O --- OD["OneDrive: user folders and files; recordings"]
O --- SP["SharePoint: communication sites; team sites"]
O --- EX["Exchange Online: group and user mailboxes"]
O --- TC["Teams Chat box: Viva Engage groups"]
Conversion note: the Mermaid diagram reproduces the layer relationships. The table and paragraph retain the labels and selection detail from the embedded illustrations.
Source footer: November 2023. © 2023 Microsoft Corporation. All rights reserved. To send feedback about this documentation, please write to us at CloudAdopt@microsoft.com.
Page 2 — Microsoft Copilot for Microsoft 365 service and tenant logical architecture
Source: PDF page 2
Source topic marker: This topic is 2 of 7.
Diagram transcription — containment, not a query sequence
The table preserves the full source labels. The Mermaid diagram shortens the Azure Open AI subscription label to keep the boundary readable.
| Boundary / group | Contents |
|---|---|
| Outside the Microsoft 365 service boundary | Devices and users; Apps on devices (Teams, Word, Outlook, PowerPoint and Excel icons) |
| Microsoft 365 service | Your Microsoft 365 tenant; Components of Copilot for Microsoft 365 |
| Your Microsoft 365 tenant | Microsoft Graph; Semantic Index for Copilot; Your customer data |
| Your customer data — Files, mailboxes, chat data, videos, etc. | Exchange mailboxes; OneDrive files; Microsoft Teams data; SharePoint files, lists, . . . |
| Components of Copilot for Microsoft 365 | Copilot for Microsoft 365; Azure Open AI service subscription for Microsoft 365 |
| Azure Open AI service subscription for Microsoft 365 | Large Language Model (LLM) |
Microsoft 365 service and tenant boundaries
Scroll sideways to read the full diagram →
View diagram source
flowchart TB
U["Devices and users"]
Apps["Apps on devices: Teams, Word, Outlook, PowerPoint, Excel"]
subgraph Service["Microsoft 365 service — Microsoft 365 service boundary"]
subgraph Tenant["Your Microsoft 365 tenant"]
Graph["Microsoft Graph"]
Index["Semantic Index for Copilot"]
Data["Your customer data: Exchange mailboxes, OneDrive files, Microsoft Teams data, SharePoint files and lists"]
end
subgraph Components["Components of Copilot for Microsoft 365"]
Copilot["Copilot for Microsoft 365"]
subgraph Azure["Azure Open AI subscription"]
LLM["Large Language Model (LLM)"]
end
end
end
Your customer data stays within the Microsoft 365 service boundary. Your prompts, responses, and data in the Microsoft Graph is not used to train foundation LLMs that Copilot leverages. Your data is secured based on existing security, compliance, and privacy policies already deployed by your organization.
Your tenant sits inside the Microsoft 365 service boundary, where Microsoft’s commitment to security, compliance, data location, and privacy are upheld.
Copilot is a shared service just like many other services in Microsoft 365. Communication between your tenant and Copilot components is encrypted.
For more information, see Data, Privacy, and Security for Copilot for Microsoft 365.
Source footer: November 2023. © 2023 Microsoft Corporation. All rights reserved. To send feedback about this documentation, please write to us at CloudAdopt@microsoft.com.
Page 3 — Semantic Index for Microsoft Copilot for Microsoft 365
Source: PDF page 3
Source topic marker: This topic is 3 of 7.
The Semantic Index for Copilot is a separate index or map of your user and company data — identifying relationships and connections. It works with Copilot and the Microsoft Graph to create a sophisticated map of all data and content in your organization to enable Copilot to deliver personalized, relevant, and actionable responses. The Semantic Index is part of the Microsoft 365 service and is created automatically.
Semantic indexing vs keyword indexing
Semantic Index builds upon keyword matching, personalization, and social matching capabilities within Microsoft 365 by creating vectorized indices to enable conceptual understanding, which helps determine your intent and helps you find what organizational content you need.
Unlike a standard keyword index, vectors are stored multi-dimensional spaces where semantically similar data points are clustered together in the vector space, enabling Microsoft 365 to handle a much broader set of search queries beyond “exact match.”
Each dimension of a vector captures an aspect of semantic meaning of the data point being represented. This provides for fast and accurate search and retrieval of data based on vector distance or similarity. This means that instead of using traditional methods for querying based on exact matches or predefined criteria, the Semantic Index finds the most similar or relevant data based on the semantic or contextual meaning.
Diagram transcription. The Microsoft 365 service boundary contains Your Microsoft 365 tenant. The tenant contains Microsoft Graph, Semantic Index for Copilot, and Customer data (Files, mailboxes, chat data, videos, etc.). The data box lists SharePoint files, lists, . . .; Exchange mailboxes; and OneDrive files. Purple arrowheads connect Semantic Index in both directions with Microsoft Graph and with each of these three data sources; the Exchange and OneDrive paths share a branch.
Semantic Index relationships inside the tenant
Scroll sideways to read the full diagram →
View diagram source
flowchart TB
subgraph Service["Microsoft 365 service — Microsoft 365 service boundary"]
subgraph Tenant["Your Microsoft 365 tenant"]
G["Microsoft Graph"] <--> S["Semantic Index for Copilot"]
subgraph D["Customer data — Files, mailboxes, chat data, videos, etc."]
SP["SharePoint files, lists, . . ."]
EX["Exchange mailboxes"]
OD["OneDrive files"]
end
S <--> SP
S <--> EX
S <--> OD
end
end
What is currently indexed
- Semantic Index indexes text-based files in SharePoint that are shared with two or more people.
- At the user level, Semantic Index indexes all email. It also indexes all text-based files in a user’s OneDrive that have been shared, interacted with (even just by the user), or commented on.
- Current supported file types include:
- Word documents (doc/docx)
- PowerPoint (pptx)
- Web pages (html/aspx)
- OneNote (one)
- Semantic Index leverages the Microsoft Graph to better correlate relationships and understand permissions.
Excluding SharePoint Online sites
You can exclude a SharePoint Online site from being indexed by both Microsoft Search and the Semantic Index (such as payroll data or financial information). Site administrators select this option within the Site Settings page.
Additional resources
- Video — Semantic Index for Copilot: Explained by Microsoft
- Training article — Examine how Copilot uses the Semantic Index (MS-012 Prepare your organization for Copilot for Microsoft 365)
Source footer: November 2023. © 2023 Microsoft Corporation. All rights reserved. To send feedback about this documentation, please write to us at CloudAdopt@microsoft.com.
Page 4 — Microsoft Copilot for Microsoft 365 query flow
Source: PDF page 4
Source topic marker: This topic is 4 of 7.
- User issues a prompt through an app.
- Copilot references the Microsoft Graph and the Semantic Index to perform grounding (providing additional content or context).
Copilot:
- References the LLM to produce the best response for the user.
- References the Microsoft Graph and the Semantic Index to perform grounding on this response.
- Returns this optimized response to the app using only data the user has access to.
Your prompts, responses, and data in the Microsoft Graph are not used for training the foundation LLM that Copilot uses.
Diagram transcription — flow and boundaries. Users/devices and app icons (Teams, Word, Outlook, PowerPoint, Excel) sit outside the Microsoft 365 service. The user-to-app arrow points toward the app; the app exchanges information with Copilot for Microsoft 365 inside the service. Copilot exchanges information with Microsoft Graph and Semantic Index for Copilot inside Your Microsoft 365 tenant, and with the Large Language Model (LLM) inside the Azure Open AI service subscription for Microsoft 365. The tenant’s Customer data box is labeled “Files, mailboxes, chat data, videos, etc.” and contains Exchange, OneDrive, Microsoft Teams, and SharePoint. Copilot and the Azure Open AI subscription are within the service boundary, alongside the tenant.
Copilot connections across service components
Scroll sideways to read the full diagram →
View diagram source
flowchart LR
U["User and devices"] --> Apps["App: Teams, Word, Outlook, PowerPoint, Excel"]
subgraph Service["Microsoft 365 service"]
C["Copilot for Microsoft 365"]
subgraph Tenant["Your Microsoft 365 tenant"]
G["Microsoft Graph"]
S["Semantic Index for Copilot"]
D["Customer data: Exchange, OneDrive, Microsoft Teams, SharePoint — files, mailboxes, chat data, videos, etc."]
end
subgraph Azure["Azure Open AI subscription"]
L["Large Language Model (LLM)"]
end
C <--> G
C <--> S
C <--> L
end
Apps <--> C
Conversion note: arrow paths are simplified; both grounding stages remain explicit in the source text above. The diagram’s shortened Azure Open AI subscription label refers to the full source label “Azure Open AI service subscription for Microsoft 365”.
Source footer: November 2023. © 2023 Microsoft Corporation. All rights reserved. To send feedback about this documentation, please write to us at CloudAdopt@microsoft.com.
Page 5 — Prepare your environment for Microsoft Copilot for Microsoft 365
Source: PDF page 5
Source topic marker: This topic is 5 of 7.
Security and information protection recommendations
Microsoft recommends building a foundation of secure productivity to get AI-ready, including Microsoft Copilot for Microsoft 365 or Copilot.
| Area to protect | Getting started with E3 | Next steps with E5 |
|---|---|---|
| Identity and access | Configure common conditional access policies With Microsoft Entra ID P1, configure the following policies to use multi-factor authentication (MFA): • Require MFA for administrators • Require MFA for all users • Block legacy authentication See Common Conditional Access policies. Be sure Microsoft 365 Services and your other SaaS apps are included in the scope of these policies. If your environment includes hybrid identities, also enforce on-premises Microsoft Entra Password Protection for Active Directory Domain Services. |
Configure recommended policies for Zero Trust With Microsoft Entra ID P1, configure the following policies to use multi-factor authentication (MFA): • Require MFA when sign-in risk is medium or high • Block legacy authentication • Require high risk users to change their password See Common security policies for Microsoft 365 organizations. Also configure Privileged Identity Management. |
| Microsoft 365 Apps | Implement Intune App Protection policies (APP) With APP, Intune creates a wall between your organization data and personal data. Policies ensure corporate data in the apps you specify cannot be copied and pasted to other apps on the device, even if the device is not managed. See Implement App Protection policies. |
Blank cell in the source. |
| Devices | Manage devices After devices are enrolled, set up compliance policies and then require healthy and compliant devices. Finally, deploy device profiles to manage settings and features on devices. Enroll devices into management Set up compliance policies Require healthy and compliant devices Deploy device profiles |
Monitor device risk and compliance to security baselines Integrate Intune with Defender for Endpoint to monitor device risk as a condition for access. For Windows devices, monitor compliance of these devices to security baselines. See Monitor device risk and compliance to security baselines. |
| Threat protection | Configure Exchange Online Protection and endpoint protection Exchange Online Protection (EOP) helps protect your email and collaboration tools from phishing, impersonation, and other threats. You can rapidly apply these protections by configuring preset security policies. Microsoft Defender for Endpoint P1 includes Attack surface reduction and Next generation protection for antimalware and antivirus protection. See Overview of Microsoft Defender for Endpoint Plan 1. |
Pilot and deploy Microsoft 365 Defender For more comprehensive threat protection, pilot and deploy Microsoft 365 Defender, including: • Defender for Identity • Defender for Office 365 • Defender for Endpoint • Defender for Cloud Apps See Evaluate and pilot Microsoft 365 Defender. |
| Organization data | Develop your classification schema and get started with sensitivity labels and other policies Sensitivity labels form the cornerstone of protecting your data. Before you create the labels to denote the sensitivity of items and the protection actions to be applied, understand your organization's existing classification taxonomy and how it will map to labels that users will see and apply in apps. Create data loss prevention policies Create retention policies Use content explorer (to review results) |
Extend policies to more data and begin using automation with data protection policies Sensitivity labeling expands to protecting more content and more labeling methods. For example, labeling SharePoint sites and Teams by using container labels, and automatically labeling items in Microsoft 365 and beyond. For more information, see a list of common labeling scenarios and how they align to business goals. |
Diagram/icon transcription: the area column uses a user icon for Identity and access; Teams, Word, PowerPoint, Excel and Outlook icons for Microsoft 365 Apps; computer/mobile icons for Devices; a shield for Threat protection; and OneDrive, Exchange, SharePoint, file, mail and site icons for Organization data. These icons add no further policy settings.
Source footer: November 2023. © 2023 Microsoft Corporation. All rights reserved. To send feedback about this documentation, please write to us at CloudAdopt@microsoft.com.
Page 6 — Set up secure file sharing and collaboration with Microsoft Teams
Source: PDF page 6
Source topic marker: This topic is 6 of 7.
Configure Microsoft Teams with appropriate protection
Microsoft provides guidance for protecting your Teams at three different levels – baseline, sensitive, and highly sensitive. Guidance includes sensitivity labels and site sharing settings. Introducing Copilot is a good time to review your environment and ensure that appropriate protection is configured.
Diagram transcription — protection distribution
| Highest protection | Sensitive protection | Baseline protection |
|---|---|---|
| Very few Teams | Some Teams | Most Teams |
The original uses three adjacent bars of increasing width. It supplies no percentages.
- First, identify Teams or projects that warrant highly sensitive protection. Configure protections for this level. Many organizations don’t have data that requires this level of protection.
- Next, identify Teams or projects that warrant sensitive protection and apply this protection.
- Finally, ensure all Teams and projects are configured for baseline protection, at a minimum.
Resources
- Set up secure file sharing and collaboration with Microsoft Teams
- Compare levels of protection
- Configure Teams with three tiers of protection
Configure external sharing with appropriate security
Introducing Copilot is a good time to review your policies for sharing files with people outside your organization and for allowing external contributors. Note that guest accounts are not licensed to use Copilot.
Sharing with people outside your organization
You may need to share information of any sensitivity with people outside your organization. Use these resources:
- Apply best practices for sharing files and folders with unauthenticated users
- Limit accidental exposure to files when sharing with people outside your organization
- Create a secure guest sharing environment
Collaborating with people outside your organization
Use these resources for setting up your environment for collaborating with people outside your organization:
- Collaborate on documents — share individual files or folders
- Collaborate on a site — collaborate with guests in a SharePoint site
- Collaborate as a team — collaborate with guests in a team
- Collaborate with external participants in a channel — collaborate with people outside the organization in a shared channel
Source footer: November 2023. © 2023 Microsoft Corporation. All rights reserved. To send feedback about this documentation, please write to us at CloudAdopt@microsoft.com.
Page 7 — Pilot and deploy Microsoft Copilot for Microsoft 365
Source: PDF page 7
Source topic marker: This topic is 7 of 7.
Onboard users to Microsoft Copilot for Microsoft 365
Microsoft recommends deploying Copilot as you phase in protections for access, devices, and data. Whether you’re starting with Microsoft 365 E3 or you’re taking next steps with Microsoft 365 E5, use the following sequence:
- Apply identity and access protection.
- Apply device protection.
- Assign Copilot licenses to users with these protections.
Ongoing — Continue your deployment of information protection capabilities.
Begin by building a plan and then testing the plan. Then roll out new configurations and capabilities incrementally. This provides the opportunity to improve on the plan while lessons are learned.
The following diagram illustrates the recommendation to start a project with a small group to evaluate the changes. This small group can be members of your IT team or a partner team. Then, pilot the changes with a larger group. Full deployment is accomplished by gradually increasing the scope of the deployment until your whole organization is covered.
Diagram transcription — deployment ring. A ring shows Evaluate as its smallest segment, Pilot as a larger segment, and Full deployment as its largest segment. No numeric percentages are supplied. The sequence is Evaluate → Pilot → Full deployment.
Applying protections and deploying Copilot in parallel
| Evaluate | Pilot | Full deployment | |
|---|---|---|---|
| Identity and access | Identify 50 users for testing | Identify the next 50-100 users in the production environment | Apply protections to the rest of the users in larger increments |
| Devices | Test device protections with the same 50 users | Apply device protections to the same users | Enroll the rest of the endpoints in larger increments |
Copilot — spans all three stages: Assign Copilot licenses to users AFTER their account and devices are protected.
Technical adoption of information protection
The source shows six staggered, overlapping bars beneath the Evaluate / Pilot / Full deployment columns. The bars progress downward and to the right; they do not specify dates, durations, or a hard gate requiring information protection to finish before Copilot deployment.
- Discover and identify sensitive business data
- Develop a classification and protection schema
- Test and pilot the schema with data in Microsoft 365
- Deploy the classification and protection schema to data across Microsoft 365
- Extend the schema to data in other SaaS apps
- Continue to discover and protect data in other repositories based on your priorities
Source footer: November 2023. © 2023 Microsoft Corporation. All rights reserved. To send feedback about this documentation, please write to us at CloudAdopt@microsoft.com.
Secure and Govern Microsoft 365 Copilot
Full Markdown conversion of the supplied PDF, reviewed on 7 September 2026. Source version: printed “Last updated March 6, 2026”.
Original PDF · Conversion notes · Sources and links · Verification
This document preserves the source’s guidance as published. Page numbers correspond to the supplied PDF. Diagram descriptions and conversion notes are identified explicitly.
Contents
- Page 1 — Secure and Govern Microsoft 365 Copilot
- Page 2 — Microsoft 365 Copilot is built on trust
- Page 3 — Microsoft 365 Copilot is built on trust
- Page 4 — Security & Governance
- Page 5 — Tools to secure and govern Microsoft 365 Copilot
- Page 6 — Problem summary
- Page 7 — Secure and Govern Microsoft 365 Copilot
- Page 8 — Remediate Oversharing
- Page 9 — Identify high-risk sites and content
- Page 10 — Apply interim Copilot protections
- Page 11 — Fix access and permissions
- Page 12 — Set up Guardrails
- Page 13 — Establish secure defaults
- Page 14 — Establish secure Guardrails
- Page 15 — Continuously Enforce & Optimize Guardrails
- Page 16 — Meet Regulations
- Page 17 — Identify and address gaps against AI Regulations
- Page 18 — Define regulatory requirements
- Page 19 — Improve data hygiene
- Page 20 — Secure & Governed Foundation for Microsoft 365 Copilot
- Page 21 — Secure & Governed Foundation for Microsoft 365 Copilot
- Page 22 — Thank you
Page 1 — Secure and Govern Microsoft 365 Copilot
Source: PDF page 1
Remediate Oversharing, enforce Guardrails, and meet AI Regulations
A Foundational Deployment Guidance
Page 2 — Microsoft 365 Copilot is built on trust
Source: PDF page 2
The central trust statement is flanked by these two panels:
| Microsoft commitments and controls | Tools to manage Copilot + agents |
|---|---|
| Microsoft 365 Copilot is built on trust | Microsoft 365 Copilot is built on trust |
Page 3 — Microsoft 365 Copilot is built on trust
Source: PDF page 3
Microsoft commitments and controls
- We secure your data at rest and in transit
- You control your data
- Your data is not used to train or enrich foundation models
- You’re protected against AI security and copyright risks
Read more at aka.ms/EDPLearn.
Page 4 — Security & Governance
Source: PDF page 4
Microsoft 365 Copilot + Microsoft Purview
| Left panel | Right panel |
|---|---|
| Honors your existing permissions | Supports your lifecycle policies and audit requirements |
| Protects against data loss and insider risk | Detect and investigation non-compliant and unethical usage |
| Assess Oversharing risks and apply recommended corrections | Guided assistance to remain compliant with AI Regulations |
A complete solution for enterprise-ready AI
Page 5 — Tools to secure and govern Microsoft 365 Copilot
Source: PDF page 5
| Tool | Source description |
|---|---|
| Microsoft Purview | Provides security, compliance, and governance across data and files |
| SharePoint Advanced Management | Included in Microsoft 365 Copilot. Provides SharePoint site management and content governance capabilities |
Page 6 — Problem summary
Source: PDF page 6
Microsoft 365 Copilot builds on the data already in Microsoft 365, prompting organizations to strengthen oversharing, data loss, and insider risks practices to support safe, confident adoption.
If grounding data and Copilot interactions are not secured, it could lead to:
- Access to information beyond what the user needs for their role
- Unauthorized disclosure of confidential information
- Out of date or irrelevant responses
By addressing these risks, organizations can ensure that Microsoft 365 Copilot is secured end to end
Page 7 — Secure and Govern Microsoft 365 Copilot
Source: PDF page 7
Three workstreams for secure, governed Copilot
Scroll sideways to read the full diagram →
| Remediate Oversharing | Set up Guardrails | Meet Regulations | |
|---|---|---|---|
| Activities | 1. Identify high-risk sites and content 2. Apply interim Copilot protections 3. Fix access and permissions |
1. Establish secure defaults 2. Establish secure Guardrails 3. Continuously Enforce & Optimize Guardrails |
1. Identify and address gaps against AI Regulations 2. Define Regulatory Requirements 3. Improve data hygiene |
| Outcomes | High-risk sites are identified, prioritized, and contained Copilot exposure is reduced immediately, even before full cleanup Foundation is set for durable Guardrails |
Guardrails are enforced by default, not manually remediated New content and sites are protected at creation Copilot can scale safely without re-introducing Oversharing risk |
AI risks and regulatory gaps are identified and tracked Copilot usage is auditable, enforceable, and defensible Higher-quality Copilot responses and a reduced Oversharing data surface |
Last updated March 6, 2026
Page 8 — Remediate Oversharing
Source: PDF page 8
Section divider in the source.
Page 9 — Identify high-risk sites and content
Source: PDF page 9
Workstream: Remediate Oversharing
Source navigation
- Identify high-risk sites and content
- Apply interim Copilot protections
- Fix access and permissions
Actions
-
[ ] Review Microsoft Purview Data Security Posture Management (DSPM) Data Risk Assessments to identify overshared sites with sensitive data, risky sharing links, and content that is frequently accessed
-
[ ] Run the SharePoint Advanced Management (SAM) Content Management Assessment to identify sites with oversized audiences, EEEU usage, broken inheritance, inappropriate sharing, and those that are inactive or ownerless
Outcome: High-risk sites are prioritized and contained
Page 10 — Apply interim Copilot protections
Source: PDF page 10
Workstream: Remediate Oversharing
Source navigation
- Identify high-risk sites and content
- Apply interim Copilot protections
- Fix access and permissions
Actions
Apply the following interim protections:
-
[ ] Enable SAM Restricted Content Discovery (RCD) to exclude sensitive sites from Copilot discovery
-
[ ] Configure Purview Data Loss Prevention (DLP) for Copilot to exclude sensitive content from Copilot grounding
Then:
- [ ] Validate via Purview audit and reports that Copilot no longer surfaces restricted content
Outcome: Copilot exposure is reduced immediately, even before full cleanup
Page 11 — Fix access and permissions
Source: PDF page 11
Workstream: Remediate Oversharing
Source navigation
- Identify high-risk sites and content
- Apply interim Copilot protections
- Fix access and permissions
Actions
-
[ ] Initiate SAM site access reviews for sites you identified as high-risk so site owners can manage the following down to the file level:
- Remove excess users, groups, and company-wide sharing links (including EEEU)
- Correct broken permission inheritance on libraries and folders
- Assign or confirm site ownership for all remediated sites using SAM site lifecycle management
-
[ ] Review Purview DSPM Data Risk Assessment recommendations for sites you flagged as high-risk and take the following actions:
- [ ] Apply site sensitivity labels to change sites from public to private, enforce allowed sharing links, set default document labels, and more
- [ ] Remediate sharing links that are too permissive
-
[ ] Remove interim Copilot protections once access and permissions are remediated
Outcome: Foundation is set for durable Guardrails
Page 12 — Set up Guardrails
Source: PDF page 12
Section divider in the source.
Page 13 — Establish secure defaults
Source: PDF page 13
Workstream: Set up Guardrails
Source navigation
- Establish secure defaults
- Establish secure Guardrails
- Continuously Enforce & Optimize Guardrails
Actions
-
[ ] Enforce Restricted Access Control (RAC) by default for business-critical sites at provisioning time
-
[ ] Disable or restrict use of Company-wide sharing groups and Sharing with Anyone link at the tenant level
-
[ ] Use Purview Information Protection to Require site sensitivity labels at provisioning to enforce correct site privacy and sharing controls by default
Outcome: Guardrails are enforced by default, not manually remediated
Page 14 — Establish secure Guardrails
Source: PDF page 14
Workstream: Set up Guardrails
Source navigation
- Establish secure defaults
- Establish secure Guardrails
- Continuously Enforce & Optimize Guardrails
Actions
Protect sensitive data and decide how it should be used with Copilot using Microsoft Purview:
-
[ ] Configure auto-label and default sensitivity labels to ensure sensitive files and emails are protected
-
[ ] Are there files or emails that shouldn’t be used by Copilot for grounding? If yes, setup Purview DLP for Copilot policies to restrict Copilot processing of files and emails with specific sensitivity labels
-
[ ] Are there keywords or types of sensitive data that shouldn’t be processed by Copilot? If yes, enable a Purview DLP for Copilot prompts policy to restrict Copilot from responding to prompts containing specified sensitive information
- [ ] Optionally, allow sensitive data to be used for Work IQ grounding but block from web grounding
-
[ ] Enable Purview Insider Risk Management (IRM) policies to detect patterns of inappropriate or non-compliant Copilot usage and automatically add risky users to more restrictive security policies
Outcome: New content and sites are protected at creation
Page 15 — Continuously Enforce & Optimize Guardrails
Source: PDF page 15
Workstream: Set up Guardrails
Source navigation
- Establish secure defaults
- Establish secure Guardrails
- Continuously Enforce & Optimize Guardrails
Actions
Use the following signals to create new or fine tune guardrails as Copilot use evolves:
-
[ ] Use Purview DSPM Activity Explorer to review Copilot interactions, prompts and responses, web search keywords, and sensitive data activity
-
[ ] Use Purview DSPM insights and Data Risk Assessments to monitor sensitive data use by Copilot and remediate new sources of oversharing
-
[ ] Review Purview IRM and DLP alerts to detect and investigate risky AI use
Outcome: Copilot can scale safely without re-introducing Oversharing risk
Page 16 — Meet Regulations
Source: PDF page 16
Section divider in the source.
Page 17 — Identify and address gaps against AI Regulations
Source: PDF page 17
Workstream: Meet Regulations
Source navigation
- Identify and address gaps against AI Regulations
- Define regulatory requirements
- Improve data hygiene
Actions
-
[ ] Use Purview Compliance Manager to assess your tenant against checklists for 300+ continuously updated regulations, including AI regulatory requirements
-
[ ] Review regulation specific improvement actions including those completed by Microsoft as the SaaS provider and actions your organization needs to complete as the data owner
-
[ ] Assign and track your progress toward regulatory compliance using Compliance Score
Outcome: AI risks and regulatory gaps are identified and tracked
Page 18 — Define regulatory requirements
Source: PDF page 18
Workstream: Meet Regulations
Source navigation
- Identify and address gaps against AI Regulations
- Define regulatory requirements
- Improve data hygiene
Actions
-
[ ] Decide how long to keep Purview audit log activities in accordance with regulatory and internal requirements
-
[ ] Decide how long to keep or when to delete Copilot interactions based on legal risk or regulatory requirements with Purview Data Lifecycle Management
-
[ ] Use Purview eDiscovery to search, preserve, and produce Copilot-related content for audits or legal requests
Outcome: Copilot usage is auditable, enforceable, and defensible
Page 19 — Improve data hygiene
Source: PDF page 19
Workstream: Meet Regulations
Source navigation
- Identify and address gaps against AI Regulations
- Define regulatory requirements
- Improve data hygiene
Actions
For Sites
-
[ ] Maintain site lifecycle hygiene by identifying and addressing inactive or obsolete sites using SAM policies and reports
-
[ ] Use Microsoft 365 Archive to store inactive but high-value content at a lower cost while preventing Copilot from processing or reasoning over it
For Files
-
[ ] Apply Purview retention and deletion policies to remove inactive or obsolete files to improve the quality of Copilot responses
-
[ ] Use Purview retention labels + Microsoft 365 Archive to exclude files from Copilot use while preserving them for recordkeeping obligations or discovery
Outcome: Higher-quality Copilot responses and a reduced Oversharing data surface
Page 20 — Secure & Governed Foundation for Microsoft 365 Copilot
Source: PDF page 20
Customers with E5 — Optimized
Realize value quickly with Copilot by remediating Oversharing, enforcing Guardrails, and meeting AI Regulations
Select deployment services are included in your FastTrack benefit. Other critical services are available thru Microsoft Unified or our Partner Ecosystem
Legend
| Source marker | Source entitlement label |
|---|---|
| 1 | Copilot and SAM |
| 2 | OE3/ME3 |
| 3 | ME5/Purview suite |
The source uses superscript markers; these remain attached to each action below. “consumption” is retained where printed.
Deployment checklist
Remediate Oversharing
1. Identify high-risk sites and content
- Run SAM Content Management Assessment to identify sites with oversized audiences, EEEU usage, broken inheritance, inappropriate sharing, as well as sensitive content, identify sites that are inactive or ownerless1
- Review Purview Data Security Posture Management (DSPM) Data Risk Assessments to identify sites with sensitive data and risky/sharing links accessible to Copilot3
- Prioritize remediation for sites and content that are high risk
2. Apply interim Copilot protections
- Enable SAM Restricted Content Discovery (RCD) to exclude sensitive sites from Copilot discovery1
- Configure Purview Data Loss Prevention (DLP) for Copilot to exclude sensitive content from Copilot grounding3
- Validate via audit and reports that Copilot no longer surfaces restricted content
3. Fix access and permissions
- Initiate SAM site access reviews for sites flagged as high-risk so site owners can remediate down to the file level1
- Review Purview DSPM Data Risk Assessment recommendations for sites flagged as high-risk and take remediation actions3
- Remove interim Copilot restrictions once site access and permissions are fully remediated.
Setup Guardrails
1. Establish secure defaults
- Enforce Restricted Access Control (RAC) by default for business-critical sites at provisioning time1
- Disable or restrict use of Company-wide sharing groups and Sharing with Anyone link at the tenant level2
- Require site sensitivity labels at provisioning to enforce correct site privacy and sharing controls by default2
2. Establish secure Guardrails
- Configure default sensitivity labels to ensure new content is protected3
- Setup Purview DLP for Copilot policy to restrict Copilot processing of labeled sensitive contentc3
- Enable Purview DLP for Copilot prompt policy to restrict Copilot from responding to prompts containing sensitive information1
- Enable Purview Insider Risk Management (IRM) policies to detect potentially inappropriate or non-compliant AI usage3
3. Continuously Enforce & Optimize Guardrails
- Use DSPM Activity Explorer to review Copilot interactions, prompts and responses, web search keywords, and sensitive data activity3
- Use DSPM Data Risk Assessments to continuously validate that sensitive data remains protected from Copilot access3
- Review Purview IRM and DLP alerts to detect and investigate risky AI usage or potential data loss3
Meet Regulations
1. Identify and address gaps against AI Regulations
- Use Purview Compliance Manager to assess your tenant against AI-related regulatory requirements and Microsoft recommended actions3
- Review improvement actions related to data protection, auditability, and AI usage controls3
- Assign and track remediation actions to close identified compliance gaps3
- Validate improvements using Purview DSPM reports3
2. Define Regulatory Requirements
- Decide how long to keep audit logs in accordance with regulatory and internal requirements2,3
- Decide how to keep or when to delete Copilot interactions based on legal risk or regulatory requirements with Purview Data Lifecycle Management2,3
- Use Purview eDiscovery to search, preserve, and produce Copilot-related content for audits or legal requests2,3
3. Improve data hygiene
- For Sites
- Maintain lifecycle hygiene by identifying and addressing inactive or obsolete sites using SAM policies and reports1
- Use Microsoft 365 Archive to store inactive but high-value content at a lower cost while preventing Copilot from processing or reasoning over it consumption
- For Files
- Apply Purview retention and deletion policies to reduce data exposure while meeting record-keeping obligations2,3
- Use Purview retention labels + Microsoft 365 Archive to exclude files from Copilot use while preserving them for recordkeeping obligations or discovery 2, consumption
Guidance assumes Copilot technical prerequisites in place: Technical enablement of core services (Teams, SharePoint, Exchange), Office Applications deployed (modern Outlook recommended) and on current or monthly update channel
Last updated March 6, 2026
Page 21 — Secure & Governed Foundation for Microsoft 365 Copilot
Source: PDF page 21
Customers with E3 — Foundational
Realize value quickly with Copilot by remediating Oversharing, enforcing Guardrails, and meeting Regulations
Select deployment services are included in your FastTrack benefit. Other critical services are available thru Microsoft Unified or our Partner Ecosystem
Legend
| Source marker | Source entitlement label |
|---|---|
| 1 | Copilot and SAM |
| 2 | OE3/ME3 |
| 3 | ME5/Purview suite |
The source uses superscript markers; these remain attached to each action below. “consumption” is retained where printed.
Deployment checklist
Remediate Oversharing
1. Identify high-risk sites and content
- Run SAM Content Management Assessment to identify sites with oversized audiences, EEEU usage, broken inheritance, inappropriate sharing, as well as sensitive content, identify sites that are inactive or ownerless1
- Use Purview Content Explorer to view which sites and files contain sensitive information types (SITs)2
- Prioritize remediation for sites and content that are high risk
3. Apply interim Copilot protections
- Enable SAM Restricted Content Discovery (RCD) to exclude sensitive sites from Copilot discovery1
- Enable a Purview DLP for Copilot prompts policy to restrict Copilot from responding to prompts containing sensitive information1
- Validate via audit and reports that Copilot no longer surfaces restricted content
2. Fix access and permissions
- Initiate SAM site access reviews for sites flagged as high-risk so site owners can remediate down to the file level:1
- Remove interim Copilot restrictions once site access and permissions are fully remediated.
Setup Guardrails
1. Establish secure defaults
- Enforce Restricted Access Control (RAC) by default for business-critical sites at provisioning time1
- Disable or restrict use of Company-wide sharing groups and Sharing with Anyone link at the tenant level2
- Require site sensitivity labels at provisioning to enforce correct site privacy and sharing controls by default2
2. Establish secure Guardrails
- Use site labels to limit access to org-wide sharing by marking sites as ’Private’ and giving access only to site members2
- Enable Purview DLP for Copilot prompt policy to restrict Copilot from responding to prompts containing sensitive information1
3. Continuously Enforce & Optimize Guardrails
- Use DSPM Activity Explorer to review Copilot interactions and sensitive data activity2
- Review Purview DLP alerts to detect and investigate potential data loss2
Meet Regulations
1. Identify and address gaps against Regulations
- Use Purview Compliance Manager to assess your tenant against regulatory requirements and Microsoft recommended actions2
- Review improvement actions related to data protection, auditability, and AI usage controls2
- Assign and track remediation actions to close identified compliance gaps2
2. Define Regulatory Requirements
- Decide how long to keep audit logs in accordance with regulatory and internal requirements2
- Decide how to keep or when to delete Copilot interactions based on legal risk or regulatory requirements with Purview Data Lifecycle Management2
- Use Purview eDiscovery to search, preserve, and produce Copilot-related content for audits or legal requests2
3. Improve data hygiene
- For Sites
- Maintain lifecycle hygiene by identifying and addressing inactive or obsolete sites using SAM policies and reports1
- Use Microsoft 365 Archive to store inactive but high-value content at a lower cost while preventing Copilot from processing or reasoning over it consumption
- For Files
- Apply Purview retention and deletion policies to reduce data exposure while meeting record-keeping obligations2
- Use Purview retention labels + Microsoft 365 Archive to exclude files from Copilot use while preserving them for recordkeeping obligations or discovery 2, consumption
Guidance assumes Copilot technical prerequisites in place: Technical enablement of core services (Teams, SharePoint, Exchange), Office Applications deployed (modern Outlook recommended) and on current or monthly update channel
Last updated March 6, 2026
Page 22 — Thank you
Source: PDF page 22
Download this deployment guide at aka.ms/Copilot/SecureGovern.
Editorial notes
These notes describe the conversion and source anomalies. They are not part of Microsoft’s published guidance.
Version boundaries
- Blueprint: pages 7, 20 and 21 print “Last updated March 6, 2026.” Its filename ends in
vMar24, but its PDF creation/modification metadata is March 19, 2026. The conversion identifies the printed date and retains the filename; it does not infer a March 2024 content version. - Architecture: all seven pages print November 2023. PDF creation/modification metadata is March 14, 2024. Pages 2–4 are landscape, while pages 1 and 5–7 are portrait.
- This pass checks fidelity to these supplied PDFs. It does not refresh their licensing, availability, product terminology, indexing behavior, or regulatory claims against current documentation. “Current” within the architecture guide refers to the source version.
Blueprint details retained
| Source page | Observation and treatment |
|---|---|
| 4 | “Detect and investigation non-compliant and unethical usage” is printed in the source and retained. The extraction-only placeholder “Slide Title” is not visibly printed and is excluded. |
| 5 | The Microsoft 365 Copilot logo supplies the end of “Included in”; this image-only wording is transcribed in the SharePoint Advanced Management table. |
| 9–19 | Each action page’s workstream navigation and outcome are retained. The nested actions on pages 11 and 14 remain nested. Site and file hygiene actions on page 19 remain separate. |
| 11 | The source groups privacy, sharing-link controls, default document labels and more under its site-sensitivity-label action. This conversion preserves that sentence; it does not add an implementation claim that a single setting accomplishes every listed action. |
| 14 | Work IQ grounding versus web grounding is an optional nested action under the prompts policy. It is not a general replacement for the separate restriction on labeled files and emails. |
| 20 | The source literally prints “processing of labeled sensitive contentc” followed by superscript 3. The stray “c” is retained and was checked at higher resolution. |
| 20–21 | Legend wording remains “1. Copilot and SAM”, “2. OE3/ME3”, and “3. ME5/Purview suite”. Numeric and “consumption” superscripts remain attached to the relevant actions. These are source markers, not refreshed licensing advice. |
| 20–21 | “Decide how to keep or when to delete Copilot interactions” is retained. Page 18 instead says “how long to keep or when to delete”. |
| 21 | The Remediate Oversharing column prints headings in the order 1. Identify, 3. Apply interim protections, 2. Fix access and permissions. The conversion preserves both the visual order and the printed numbers. Other pages use 1–2–3. |
| 21 | The site-access-review bullet ends “file level:” with a colon before superscript 1, without sub-bullets. No actions from another page have been silently inserted there. |
The E3 and E5 checklists are separate full transcriptions. Details unique to either checklist remain in that checklist, including Content Explorer versus DSPM Data Risk Assessments, prompts policies versus labeled-content policies, the E5 validation action using DSPM reports, and the source’s technical prerequisites.
Architecture details retained
| Source page | Observation and treatment |
|---|---|
| 1 | The diagram’s “Teams Chat” area contains the caption “Viva Engage groups”. Both labels are retained. The embedded Graph illustration and data-object captions are image content that plain PDF text extraction misses; these are manually transcribed. |
| 2–4 | The diagrams place the tenant and Copilot/Azure Open AI components within the Microsoft 365 service. Copilot and the LLM subscription are outside the depicted tenant box. The conversion keeps these boundaries distinct. |
| 3 | The source says vectors are “stored multi-dimensional spaces”; the missing “in” is not silently supplied. Its indexing criteria and five supported file-type entries remain source-version statements. |
| 4 | Grounding appears both before the LLM response and on the response. Both statements are retained. |
| 5 | Both identity columns explicitly say “Microsoft Entra ID P1”, including the E5 column containing risk-based policies. This is preserved as printed and should not be read as an independently verified licensing conclusion. |
| 5 | The E5 Microsoft 365 Apps cell is blank. The conversion labels that source blank explicitly instead of inventing a recommendation. |
| 5–7 | Some PDF hyperlink annotations cover whole text blocks and occur twice at identical positions. Contextual Markdown links retain the visibly linked text and exact destinations. The source register preserves every raw annotation in its manifest, plus a readable page/destination register. |
| 6 | “Guest accounts are not licensed to use Copilot” remains a statement from this source version. “Highest protection” in the diagram and “highly sensitive” in the narrative are both retained. |
| 7 | The source prints 50 evaluation users and the next 50–100 pilot users. Copilot licenses follow account and device protection; the information-protection work proceeds in parallel. No dates or percentages have been invented for the ring or staggered bars. |
Representation choices
- Source page order, page/topic markers, visible titles and recurring architecture footers are retained. Extraction-only duplicate title text, overprinted glyph copies, and the hidden slide placeholder do not become duplicate reader-facing prose.
- Physical line wrapping is reflowed. Bullets/check boxes become Markdown lists; column and row associations become tables or labeled sections. The rotated “Deployment checklist” text on blueprint pages 20–21 becomes a normal heading.
- Brand marks are named where they carry meaning. Decorative slide backgrounds, shadows, colors, and repeated app artwork are not recreated. Original PDFs remain directly linked at every page.
- Mermaid diagrams express logical containment and connections. Prose/table transcriptions retain image labels, selected-object examples, shared branches, and rollout/protection proportions. The Graph’s illustrative network is described as a relationship map; its decorative mesh is not promoted into additional product architecture claims.
- Paragraphs explicitly labeled “Diagram transcription”, “Conversion note”, and “Source topic marker” explain the representation. They are editorial text, not quotations from Microsoft.
Sources and embedded links
Source files supplied by the user; conversion review date: 7 September 2026. The original PDFs are the authority for this conversion. No web content has been substituted for them.
Blueprint
Original PDF · Full Markdown guide
- Pages: 22
- SHA-256:
3338f24bc893715ac8e4448375794df88115943f492e4001ef9f0d80f2848109 - Extraction and raw annotation manifest
- Raw URI annotations: 4; distinct page/destination pairs: 2. Repeated annotations and multiple linked blocks can share one destination.
| PDF page | Exact embedded destination |
|---|---|
| Page 3 | https://aka.ms/EDPLearn |
| Page 22 | https://aka.ms/Copilot/SecureGovern |
The guide places these destinations beside the relevant source text. External destinations are preserved, including their original query strings; network reachability and current article contents were not revalidated.
Architecture
Original PDF · Full Markdown guide
- Pages: 7
- SHA-256:
52a9d71352232bab96afb0399e99a35a7ed9af1be7193c0d46ca84989095d0a8 - Extraction and raw annotation manifest
- Raw URI annotations: 78; distinct page/destination pairs: 32. Repeated annotations and multiple linked blocks can share one destination.
The guide places these destinations beside the relevant source text. External destinations are preserved, including their original query strings; network reachability and current article contents were not revalidated.
Conversion verification
Reviewed on 7 September 2026 against the two supplied PDFs. The review covers 29 source pages, including title/divider pages, both blueprint checklists, all architecture diagrams, and all resource sections.
Results
| Check | Result |
|---|---|
| Source preservation | Both original PDF SHA-256 hashes match the initial extraction manifests. |
| Visual source review | All 22 blueprint pages and all 7 architecture pages rendered and inspected. Blueprint page 20’s “contentc” typo also checked at higher resolution. |
| Detailed text comparisons | 25 regions pass normalized text equality: 9 blueprint action panels; 6 complete E3/E5 checklist columns; all 10 architecture protection-table cells, including the source blank. |
| Embedded destinations | All 34 distinct source-page/destination pairs appear in the corresponding guide page. The manifests retain all 82 raw URI annotations, including duplicates and overlapping text-block annotations. |
| Page coverage | 22 and 7 page anchors, respectively, in PDF order, with source-page links and contents navigation. |
| Markdown tables | All 11 tables in the two guides rendered and visually reviewed for row/column association and readable content. |
| Mermaid | All 4 architecture diagrams rendered successfully and visually reviewed against the source. |
| Layout | No horizontal page overflow in the 1440-pixel validation viewport. |
The text comparisons remove punctuation, whitespace and case and compare the remaining letter/digit sequence. They detect missing or reordered words and entitlement digits in the specified regions; they do not prove punctuation fidelity, list nesting, or diagram meaning. Those were reviewed separately against the rendered PDFs. A whole-page token coverage check supplements the region comparisons; its explicitly excluded extraction artifacts are documented below.
Blueprint page coverage
| Page | Content retained / representation |
|---|---|
| 1 | Title, subtitle and foundational-deployment description |
| 2 | Central trust statement and both surrounding panels |
| 3 | Four Microsoft commitments/controls and the EDPLearn link |
| 4 | Copilot + Purview relationship, six security/governance panels and closing banner |
| 5 | Both tool descriptions and the logo-carried “Included in Microsoft 365 Copilot” wording |
| 6 | Problem statement, three risk outcomes and end-to-end protection statement |
| 7 | All nine activities, all nine outcomes, column associations and printed date |
| 8 | Remediate Oversharing divider |
| 9 | Both risk-assessment actions, navigation and outcome |
| 10 | Both interim protections, validation action, navigation and outcome |
| 11 | Site reviews, all three review sub-actions, DSPM review and its two sub-actions, removal of interim protections, navigation and outcome |
| 12 | Set up Guardrails divider |
| 13 | Three secure-default actions, navigation and outcome |
| 14 | Introductory guidance, four main actions, optional Work IQ/web-grounding sub-action, navigation and outcome |
| 15 | Three continuous-enforcement signals, navigation and outcome |
| 16 | Meet Regulations divider |
| 17 | Three regulatory-gap actions, including 300+, SaaS/data-owner responsibilities and Compliance Score; navigation and outcome |
| 18 | Audit-log retention, interaction retention/deletion, eDiscovery, navigation and outcome |
| 19 | Both site actions, both file actions, navigation and outcome |
| 20 | Full E5 Optimized checklist, all three columns, all source markers, service-benefit statement, prerequisites and printed date |
| 21 | Full E3 Foundational checklist, all three columns, all source markers, printed 1–3–2 sequence, service-benefit statement, prerequisites and printed date |
| 22 | Closing title and download link |
Architecture page coverage
| Page | Content retained / representation |
|---|---|
| 1 | Introduction, logical layers, Graph image labels, organization-data image captions, selected-object example and Mermaid layer diagram |
| 2 | All service/tenant/component boundaries and labels, customer-data description, privacy/security paragraphs and resource link |
| 3 | Semantic-index explanation, vector/keyword comparison, indexed-data criteria, five file types, exclusion guidance, containment and arrows, both resources |
| 4 | Prompt flow, both grounding stages, LLM response, access limitation, training statement and all depicted boundaries/data labels |
| 5 | All five protection areas and their complete E3/E5 cells, explicit source blank, app/data icons and resource destinations |
| 6 | Three protection levels and qualitative proportions, ordered remediation sequence, three general resources, external-sharing narrative, three sharing resources and four collaboration resources |
| 7 | Onboarding sequence, ongoing information protection, incremental rollout narrative, qualitative ring proportions, identity/device rollout table, cross-stage licensing condition and all six linked information-protection bars |
Each architecture page retains its topic marker and November 2023 copyright/feedback footer.
Extraction artifacts and review limits
- Overprinted duplicate glyphs in the Visio PDF were deduplicated before extraction comparisons.
- The extraction-only duplicate trust title on blueprint page 3 and hidden “Slide Title” placeholder on page 4 are excluded from reader-facing text. The visible trust title is retained.
- Rotated “Deployment checklist” lettering on blueprint pages 20–21 extracts as scattered single letters. It is retained as an ordinary heading instead.
- Text embedded in pictures, graphical containment, arrow directions, shared branches, selected objects, list nesting and qualitative proportions were checked visually. Their Markdown equivalents are described in the guides and editorial notes.
- This is a source-conversion review. Current product behavior, licensing and linked-webpage contents were not independently refreshed.
Evidence and reproducibility
Initial source manifests and rendered-page evidence retain per-page extracted text, word positions, PDF metadata, SHA-256 hashes, and raw URI annotation rectangles.
- Guide construction script
- Region and embedded-link verification
- Whole-page token coverage
- Markdown/Mermaid render results
- Local-link validation
- Delivered Markdown hashes