Microsoft is enhancing Windows 11 to embrace the agentic AI era, allowing AI agents to autonomously manage tasks based on user commands. For example, users can ask the AI to order a pizza, effectively automating web browsing and payment processes. However, this innovation raises significant security concerns. Users can activate the new “Experimental agentic features” in Build 26220.7262, but Microsoft warns that these features could affect device performance and security. Vulnerabilities, including cross-prompt injection attacks, pose risks as attackers can embed malicious instructions within benign documents, leading AI agents to execute harmful actions like malware installation or unauthorized data sharing. While these AI agents operate within a controlled “Agentic Workspace,” allowing limited and auditable access to certain folders, they still present substantial security challenges. Microsoft’s solution, reminiscent of the Windows Sandbox, allows ongoing operations but requires addressing prompt injection vulnerabilities to ensure user safety. For detailed insights, refer to Windows Latest.
Source link
Share
Read more