As AI systems evolve from passive tools into autonomous agents that initiate actions and interact with other services, traditional Identity and Access Management (IAM) models start to fail. Built for humans and static service accounts, legacy IAM cannot keep pace with the scale and dynamism of agentic AI. In this session, David Goldschlag, co-founder of Aembit, explores why securing AI agents requires a shift from secrets and over-privileged machine identities to ephemeral, policy-driven, and context-aware access. Attendees will leave with practical patterns for treating AI agents as first-class identities and moving toward dynamic, identity-centric access models designed for an autonomous future.