Microsoft is testing OpenClaw-inspired autonomous agent capabilities for Microsoft 365 Copilot, targeting enterprise customers with tighter security controls than the open-source project offers. Omar Shahine, Microsoft’s corporate vice president, confirmed to The Information that the company is “exploring the potential of technologies like OpenClaw in an enterprise context.”
Always-On Agents With Role-Based Permissions
The planned features would make Copilot “run autonomously around the clock” to complete tasks on behalf of users, according to The Verge. An always-on version of 365 Copilot could monitor Outlook inboxes and calendars, surfacing suggested tasks each day without waiting for a user prompt.
Microsoft is also exploring role-specific agents tailored to marketing, sales, and accounting functions, according to The Information. The approach siloes each agent’s permissions by job function rather than granting broad system access, a direct response to the security incidents that have plagued OpenClaw deployments.
The Third Agentic Layer in Four Months
This marks Microsoft’s third distinct agentic push since February. In February, the company released Copilot Tasks in preview, an agent for completing multi-step work ranging from email organization to travel planning. In March, Microsoft announced Copilot Cowork, which takes actions inside Microsoft 365 apps using its “Work IQ” intelligence layer. Cowork also integrated Anthropic’s Claude as a model option after a partnership announced late last year.
Both Copilot Tasks and Cowork run in the cloud. Whether the new OpenClaw-style features will run locally on user hardware or remain cloud-based has not been confirmed, according to TechCrunch. The distinction matters: OpenClaw’s local execution model is a core part of its appeal for users who want agents with direct access to their filesystem, apps, and peripherals.
Build Conference Preview
Microsoft plans to demonstrate the new capabilities at its Build developer conference starting June 2, according to The Verge. The company declined to share specifics on how this effort relates to its existing agent products or whether it will run locally.
“Across our work, we are continuously experimenting as we bring broader orchestration and autonomy to our enterprise and consumer AI experiences while staying anchored in security, governance, and trust,” a Microsoft spokesperson told TechCrunch.
Competitive Pressure
The timing reflects growing competition from Anthropic, which has moved aggressively into enterprise agent territory with Claude Cowork and Managed Agents. OpenAI shipped its updated Agents SDK with native sandboxing on April 15. And the open-source OpenClaw project continues to gain users despite recurring security concerns, with Mac Mini sales surging as the go-to hardware for OpenClaw deployments.
Microsoft’s bet is that enterprises want OpenClaw’s autonomy without its risk profile, delivered through infrastructure they already pay for.