Agentic AI Security: The Gaps That RSAC 2026 Isn’t Filling In

RSA 2026 is dominated by agentic AI. Its promise, its power, and its risk. But behind the hype is a growing reality: if your agents can reach it, they will use it. As AI agents autonomously access data, invoke tools, and chain actions across systems, traditional IAM models are breaking fast. The real gap isn’t the model, it’s authorization.

In this session, PlainID and CSA experts cut to the chase and discuss what actually matters: control. We’ll explore:

• Where legacy IAM (OAuth, SAML) fails in autonomous systems and the minimum fix
• What a real agentic identity looks like in practice
• How MAESTRO threat modeling and defense-in-depth apply to AI workflows
• The new attack surface: RAG pipelines, MCP tools, and multi-agent delegation
• Why runtime, intent-based authorization is becoming the control plane for AI

This is not theory. It’s a practical discussion on how to discover, classify, and control what your AI agents can access, do, and expose, before it becomes your next incident.