AI in education: How to control AI tools and agentic AI without limiting innovation

This advisory warns that AI and emerging agentic AI in educational institutions expand the attack surface—introducing risks like sensitive data uploads to external services, insecure AI-generated code, shadow IT, and large-scale automated actions if identities are compromised—and recommends applying practical Zero Trust controls (application allowlisting, granular data and network access restrictions, limiting impact of compromised accounts, and visibility/containment of agent behavior) to balance innovation with security.