Microsoft has been steadily integrating AI features into Windows 11, but recent developments mark a significant evolution in this journey. The latest build released to Windows Insider Program testers introduces an intriguing toggle for “experimental agentic features” within the Settings menu, paving the way for a feature known as Copilot Actions. Accompanying this rollout is a comprehensive support article from Microsoft that delves into the functionality of these new features.
The term “agentic” has become a cornerstone of Microsoft’s vision for Windows 11. In essence, these agents are designed to perform tasks autonomously in the background, freeing users to focus on other priorities. Microsoft envisions these agents handling a variety of everyday responsibilities, such as organizing files, scheduling meetings, and sending emails. The goal is to provide users with an “active digital collaborator” capable of executing complex tasks, thereby enhancing both efficiency and productivity.
However, like many AI systems, these agents are not without their challenges. They can occasionally misinterpret instructions or generate inaccuracies, often proceeding with confidence even when uncertain. Moreover, Microsoft acknowledges the “novel security risks” posed by these agents, particularly concerning the potential for malicious actors to manipulate them. This necessitates a careful balance in their implementation, ensuring that while agents have access to necessary files, they remain sufficiently isolated from critical system components.
Possible risks and attempted fixes
Credit: Microsoft
To mitigate potential risks, Microsoft has devised a strategy where AI agents operate under distinct user accounts, separate from personal accounts. This structure ensures that agents lack the authority to make sweeping changes across the system, while also providing them with their own “desktop” environment that does not interfere with the user’s ongoing tasks. Users retain control by needing to approve any requests for data access, and all actions performed by the agents are designed to be transparent and distinguishable from user actions. Furthermore, Microsoft emphasizes the importance of accountability, requiring agents to maintain logs of their activities and offering users a means to supervise their operations, including a detailed list of actions for multi-step tasks.
