Microsoft introduced Autopilot on September 25, 2026, as part of its new Copilot experience and said a preview is rolling out to its first customers. The company describes Autopilot, previously called Microsoft Scout, as a persistent, proactive personal agent that keeps working when its user is away. Microsoft says the agent is built on OpenClaw and is expanding to private preview at the end of September. Omar Shahine, who leads the Autopilot team, said Microsoft is working with OpenClaw creator Peter Steinberger and the OpenClaw Foundation to build an enterprise-grade runtime.

Microsoft contributors are also adding policy conformance checks to OpenClaw. Its Policy plugin lets operators specify requirements, compare them with an agent’s actual configuration and record the result. Contributions extend checks to model providers, networks, MCP servers, secrets and authentication settings. Shahine also added tests to confirm that sample incoming messages reach the intended agent, helping organizations check which channels, providers and resources an agent can access and whether its setup still follows their rules.

Microsoft and the Foundation have described a long-term collaboration to make Windows a leading platform for OpenClaw. A native Windows companion includes guided setup, WinUI chat, inline command approvals, model selection, sandbox integration and support for displaying user and agent media. Contributors also added the MXC sandbox backend for supported Windows environments, giving operators another way to limit agent command execution through Microsoft’s execution-container technology.

OpenClaw updates target reliability during background jobs, restarts, long conversations and failures. Changes prioritize queued user requests over background work, fix scheduler behavior that could hang the Gateway, and improve responsiveness during database recovery. A streaming fix prevents duplicate Codex replies in conversation history when streaming and completion identifiers differ. Other changes guard against repeated tool loops after context compaction, reduce session-store memory retention, and preserve uncertainty when an agent cannot tell whether a command ran, preventing automatic retries in that situation.

Additional contributions let agents react to Teams messages without copying an internal message ID while keeping the action within the current conversation; redact recognizable API keys, tokens and passwords from command text in approval prompts; and improve Azure OpenAI provider support by connecting Azure endpoints to the appropriate Responses implementation. Shahine also contributed iMessage approvals, native polls and changes that keep replies in the correct conversation, benefiting OpenClaw users on Apple devices. The article says the public record shows substantial Microsoft contributions, while noting that not every Microsoft employee contribution is necessarily an Autopilot feature or company-sponsored project.

Microsoft introduced Autopilot on September 25, 2026, as part of its new Copilot experience and said a preview is rolling out to its first customers. The company describes Autopilot, previously called Microsoft Scout, as a persistent, proactive personal agent that keeps working when its user is away. Microsoft says the agent is built on OpenClaw and is expanding to private preview at the end of September. Omar Shahine, who leads the Autopilot team, said Microsoft is working with OpenClaw creator Peter Steinberger and the OpenClaw Foundation to build an enterprise-grade runtime.

Microsoft contributors are also adding policy conformance checks to OpenClaw. Its Policy plugin lets operators specify requirements, compare them with an agent’s actual configuration and record the result. Contributions extend checks to model providers, networks, MCP servers, secrets and authentication settings. Shahine also added tests to confirm that sample incoming messages reach the intended agent, helping organizations check which channels, providers and resources an agent can access and whether its setup still follows their rules.

Microsoft and the Foundation have described a long-term collaboration to make Windows a leading platform for OpenClaw. A native Windows companion includes guided setup, WinUI chat, inline command approvals, model selection, sandbox integration and support for displaying user and agent media. Contributors also added the MXC sandbox backend for supported Windows environments, giving operators another way to limit agent command execution through Microsoft’s execution-container technology.

OpenClaw updates target reliability during background jobs, restarts, long conversations and failures. Changes prioritize queued user requests over background work, fix scheduler behavior that could hang the Gateway, and improve responsiveness during database recovery. A streaming fix prevents duplicate Codex replies in conversation history when streaming and completion identifiers differ. Other changes guard against repeated tool loops after context compaction, reduce session-store memory retention, and preserve uncertainty when an agent cannot tell whether a command ran, preventing automatic retries in that situation.

Additional contributions let agents react to Teams messages without copying an internal message ID while keeping the action within the current conversation; redact recognizable API keys, tokens and passwords from command text in approval prompts; and improve Azure OpenAI provider support by connecting Azure endpoints to the appropriate Responses implementation. Shahine also contributed iMessage approvals, native polls and changes that keep replies in the correct conversation, benefiting OpenClaw users on Apple devices. The article says the public record shows substantial Microsoft contributions, while noting that not every Microsoft employee contribution is necessarily an Autopilot feature or company-sponsored project.