Microsoft announces Scout, an autonomous AI agent built on OpenClaw

Microsoft has launched Scout, an autonomous AI agent for Microsoft 365 that Microsoft calls an "autopilot" agent. Unlike assistants that require the user to issue explicit commands each time, Scout is designed to run continuously in the background, understand the user's workflows across different applications and systems, and execute tasks on the user's behalf. Scout is built on the OpenClaw agent framework, has a governed Entra identity, can connect to Microsoft 365 applications such as Teams, Outlook, OneDrive, and SharePoint, and can read or use data such as chats, email, calendar, and contacts. Users can access Scout through Teams; it can also interact with the browser and connect to external applications via the model context protocol (MCP); it supports cloud, desktop, and web environments. Microsoft gives examples that Scout can help coordinate meeting times, reserve calendar slots based on upcoming work arrangements, and detect risks such as stalled decisions that could turn into blockers. Currently Scout is an experimental release, offered to customers of the Microsoft Frontier program, and requires Intune policy configuration and opt-in attestation. The article also situates Scout within Microsoft's broader agentification strategy for Microsoft 365, including Microsoft 365 Copilot's Agent Mode, Copilot Cowork, and Google Workspace's recent competing product Spark. On the commercial side there is still uncertainty, as Microsoft has not yet stated whether Scout is included in the Microsoft 365 Copilot subscription or charged separately. Security is a focus of the article: a Forrester analyst argues that for organizations that have already deployed Microsoft 365 Copilot, Scout does not necessarily introduce entirely new data risks, but it amplifies existing data governance problems; the difference is that the agent not only displays sensitive data but may also take actions on the user's behalf. Related risks include expanded data exposure, agent manipulation, prompt injection, unintended tool use, and observability gaps that make it difficult to understand user intent and agent behavior. Overall, Scout shows that enterprise office AI is moving from conversational assistants toward long-running, autonomous, multi-tool execution agents, but before actual deployment, permissions, auditing, data governance, and behavioral boundaries will be more critical than feature demonstrations.