Microsoft 365 Copilot — Prompt Flow Authoritative

Hub-and-spoke view: every step flows through the Copilot Orchestrator. Blue = forward (request) · Cyan = return (response).
Source: Microsoft 365 Copilot Chat Privacy and Protections — Microsoft Learn
User Orchestrator Bing Graph LLM Purview Logging
Microsoft 365 Service Boundary — Your Tenant 🛡 EDP ENABLED 👤 User Web · M365 App Teams · Outlook ↺ receives response 🛡 RAI coordinated throughout (input + output) 🧩 Copilot Orchestrator ◉ RAI-coordinated Central hub Routing · Model · Safety + Purview · Exchange logging Chat + M365 Copilot 🌐 Web Grounding (Bing) conditional · web grounding Copilot Chat: primary web source M365 Copilot: only if Graph lacks it Short queries · no user/tenant IDs 🗂 Microsoft Graph conditional · org content Emails · Files · Teams · Calendar Permission-scoped · Labels · DLP Semantic Index (M365 Copilot) 🧠 Azure OpenAI GPT family Microsoft-operated Inside service boundary No training on your data 🤝 Anthropic Claude Microsoft subprocessor · admin opt-in Outside Microsoft service boundary Hosted by Anthropic (AWS) No training on your data — alternative to Azure OpenAI — 🧾 Purview Compliance · DLP Sensitivity labels Citations Output enforcement 📒 Logging Exchange (your tenant) Prompt + final response Audit · eDiscovery · Retention alt · opt-in subprocessor 1 2 3 4 5 6 7 8 prompt grounded prompt raw response final response policies applied inside Orchestrator logged inside Orchestrator policy → log 🔒 Microsoft 365 Service Boundary · ✓ Prompts & responses not used to train foundation models 🤝 Subprocessor zone — Anthropic Claude is processed by Anthropic outside the M365 service boundary (admin opt-in)
1 User writes a prompt in a Microsoft 365 app. The prompt is sent to the Copilot Orchestrator.
Authoritative source: Microsoft Learn — Copilot Chat Privacy & Protections · Application card · Anthropic as Microsoft subprocessor
INTROClick play to begin the narrated tour.
0:00 / 0:00