Microsoft has raised concerns regarding security risks associated with its new agentic AI capabilities in Windows, particularly through the introduction of the agent workspace feature which allows AI agents to operate in isolated environments. While these agents enhance user interaction by performing tasks autonomously, they also open up vulnerabilities such as cross-prompt injection attacks, which could lead to data breaches or malware installation. The company is addressing these risks with stringent security principles focused on non-repudiation, confidentiality, and authorization, ensuring actions by agents are distinguishable from user actions. As Microsoft cautiously rolls out these capabilities, it emphasizes the need for ongoing security measures to adapt to the evolving challenges posed by autonomous AI systems.