New AI agent for Microsoft 365 Copilot.
Aims for enhanced enterprise security.
Builds on existing Copilot tools.

Atlas AI
Microsoft is developing new artificial intelligence agent capabilities for integration into Microsoft 365 Copilot, targeting enterprise customers, the company confirmed on April 13, 2026. The planned features are described as similar to the open-source OpenClaw agent, but with stronger security controls designed for corporate use.
The company’s effort centers on a local-running agent that can carry out tasks directly on a user’s computer. That approach would expand Microsoft’s existing set of agentic tools by adding an option that operates on-device, rather than relying only on cloud-based execution.
Microsoft’s confirmation also pointed to security as a key differentiator. OpenClaw has been noted for security risks, and MicrosoSources said its enterprise-focused agent features will include enhanced security controls compared with the open-source tool. The company did not provide additional technical details in its confirmation.
This development follows a series of Copilot-related agent announcements earlier in 2026. In March 2026, Microsoft announced Copilot Cowork, an AI designed to take actions inside Microsoft 365 applications instead of only returning search results or operating as a chat interface. MicrosoSources said Cowork uses its “Work IQ” technology to tailor experiences for users across its applications, and it can be powered by Anthropic’s Claude AI model.
In February 2026, Microsoft introduced Copilot Tasks, another agent intended to complete tasks, which was released in preview. The company positioned Copilot Tasks toward prosumers, highlighting use cases such as organizing emails, travel, and appointments.
Microsoft drew a distinction between these existing agents and the newly described local-running concept. Both Copilot Cowork and Copilot Tasks operate in the cloud, while the proposed agent is intended to run locally on a user’s machine. For enterprise buyers, that difference can shape how organizations think about security controls and operational deployment, particularly when agent software is designed to perform actions on endpoints.
Several elements remain unclear based on Microsoft’s April 13, 2026 confirmation. The company did not specify a release timeline, availability, pricing, or the full scope of tasks the local-running agent will be able to perform. It also did not detail how the enhanced security controls will be implemented relative to OpenClaw, beyond stating that the enterprise version will be more secure.


