Microsoft plans to bring agentic AI to Windows 11 with a preview build toggle that lets AI-powered apps complete tasks on users’ behalf. Once enabled, every user on the device, including other administrators and standard users, is part of the agentic AI environment.
The company says that AI agents run in their own secure environment, called an agent workspace, and operate under identities distinct from those of users. These agents automate routine tasks such as sorting files, sending emails, or editing images directly from the Windows interface.
Users can invoke these agents from the taskbar using the all-new Copilot “Ask.” Typing “@” into the search bar shows available agents.
Microsoft warns that these new options carry potential risks. The company highlights threats such as cross-prompt injection, where malicious content could manipulate agents into performing harmful actions or leaking personal data. Because of the higher risk profile, Microsoft requires administrator approval before anyone enables agentic AI, and the feature is off by default, Windows Central reports.
The company will keep a clear record of all actions taken by agents in a secure audit log. This allows users to check AI activity and confirm what actions agents have performed.
Microsoft says controls will remain transparent, and agents will need user input for important actions. Because this update is still in preview, users can expect more apps, such as third-party AI assistants and workflow agents, to join the ecosystem.