One Compromised Agent ID Blueprint Can Cross Tenant Boundaries
Summary
Merrill Lin interviews Datadog security researcher Katie Knowles about her deep dive into Microsoft Entra Agent ID. They discuss the complex identity model, including blueprints and multi-tenant risks, and how attackers might exploit these structures. The episode covers detection strategies, permission management, and best practices for securing agent identities against compromise.
Topics discussed
Introduction and guest background
Origin of research and Agent ID fundamentals
Understanding blueprints and identity structures
Blast radius and multi-tenant risks
Third-party agents and permission scopes
Permission restrictions and role assignments
Attack scenarios and privilege escalation
Detection challenges and logging visibility
Multi-tenant takeover attack chain
Portal differences and evolving permissions
Administrative controls and business processes
Third-party blueprint consent strategies
Secrets vs certificates and federation
Microsoft's internal cleanup and AI assistance
Monitoring, disabling, and ID protection
Future outlook and closing remarks
Listen ad-free on Castria