Microsoft has detailed new AI containment strategies to help organizations control autonomous systems, limit potential damage, and build a stronger foundation for secure AI adoption. The company notes that many AI risks originate from existing security weaknesses, including excessive user permissions, outdated software, vulnerable code, exposed internet-facing systems, and limited visibility into system activity. As AI agents become more autonomous, they can act faster and at a larger scale than humans, meaning these weaknesses can lead to greater consequences, such as agents unintentionally accessing sensitive resources, crossing security boundaries, or performing unauthorized actions. Microsoft’s guidance focuses on placing clear limits on agent access, permissions, and actions, encouraging organizations to control which tools agents can use, minimize the impact of compromise, strengthen identity and access management, improve monitoring, and establish rapid shutdown mechanisms. The Secure Now framework outlines six security priorities: ensuring AI agents operate within defined boundaries; maintaining strong cyber hygiene by promptly applying security updates to Microsoft products and addressing vulnerabilities in open-source components and third-party dependencies; secure application development with continuous scanning of source code and configurations; reducing exposure to external threats by identifying internet-facing assets and eliminating unnecessary exposure; and enabling multifactor authentication, limiting privileges to only what is required, and retiring outdated authentication methods.