微軟在 Windows 11 工作列與檔案總管展示全新 AI 體驗「Ask Copilot」

• Copilot Connectors Architecture: Microsoft 365 Copilot uses two connector types—synced connectors that index external data into Microsoft tenant databases, and federated connectors that connect to data in real time[2] • File Explorer Integration: Right-click context menu triggers AI-powered summaries and previews, leveraging Copilot's ability to analyze file content and metadata[3] • Data Access Control: Copilot respects sensitivity labels, permissions, and Rights Management Service (RMS) protections; file labels are reflected in chat responses to maintain confidentiality awareness[5] • Identity Mapping: Data source identities are mapped using Microsoft Entra IDs, enabling permission-based access control where only users with source access can view indexed content[6] • Governance Implementation: Microsoft uses Microsoft Graph Data Connect for oversharing detection and Microsoft Purview for lifecycle management, attestation protocols, and DLP standards configuration[4] • Connector Capabilities: Indexed data includes metadata properties (creation date, owners, labels, status) with customizable crawl frequencies; some connectors have limitations (e.g., Shortcut Story connector only indexes two-year-old stories)[6]

Microsoft's integration of Copilot into Windows 11's core UI (taskbar and File Explorer) signals a strategic shift toward making AI assistance ambient and always-accessible rather than application-specific. This approach aims to reduce adoption friction and establish Copilot as the default productivity layer across enterprise workflows. The expansion of 100+ connectors indicates Microsoft is positioning Copilot as a unified data access layer that abstracts complexity across disparate enterprise systems (ServiceNow, Salesforce, Google Workspace, etc.), potentially commoditizing traditional enterprise search and knowledge management tools. The emphasis on data governance and sensitivity labeling suggests Microsoft anticipates regulatory scrutiny around AI data access in regulated industries. Competitors will likely accelerate similar OS-level integrations and connector ecosystems to maintain parity. The phased deployment model Microsoft used internally (300,000+ employees) may become an industry standard for large-scale AI adoption, emphasizing change management and governance over rapid rollout.