AI agents are software applications powered by language models that can connect to enterprise tools and systems, and use both short-term and long-term memory to accomplish tasks. As enterprises increasingly adopt these agents to boost efficiency, improve decision-making, and enhance customer experience, they also introduce new security risks. Effectively onboarding AI-agent platforms into a comprehensive SaaS Agent Security framework is essential to managing these risks.

This onboarding process provides comprehensive visibility and discovery for AI agents operating within supported SaaS environments (for example, O365, Salesforce, and ServiceNow). A key objective is to mitigate “shadow AI” risks by establishing a unified and accurate inventory of agent platforms. New and existing SaaS-based agent platforms can be seamlessly integrated for security monitoring.

Furthermore, the onboarding process is designed to enable the tagging of agent platforms as Sanctioned, Tolerated, or Unsanctioned, allowing for appropriate security controls, such as blocking unsanctioned agents on the network. Specific attention is given to discovering agents within platforms like Microsoft Copilot Studio, with requirements for viewing detailed metadata such as agent name, creator, published date, users, triggers, knowledge sources, allowed actions, channels, and connections used. This comprehensive approach to onboarding lays the foundation for maintaining a secure and compliant AI agent ecosystem within the enterprise.