Skip to content

Rethinking Identity, Policy, and Access Control (IPAC) for Agentic AI - BRKETI-2623#

Cisco Live Europe 2026 (Amsterdam) digest Program Core pick Amsterdam · 9–13 February 2026

Abstract

Marcelo Yannuzzi (Cisco)

  • Why: Agentic identity is where the PM's identity work is heading, and this is one of the first vendor architectures that names the actual problem: current IAM scopes, scopes authorization around humans and static service accounts, not autonomous NL-driven entities acting at machine speed.
  • What: The technology gaps in today's IAM for agentic AI, and a proposed IPAC architecture covering: dynamic trust establishment among users, agents, and MCP servers (Just-In-Time identification, Multi-Factor Verification across domains), policy that defines what autonomous entities may do without a human in the loop, and granular authorization for entities communicating in natural language — with a demo and pointers to Cisco's open-source efforts (from abstract only).
  • How: Architecture-plus-demo: identify the gaps (static identity, human-in-loop assumptions), then layer JIT identification and policy enforcement over agent/MCP interactions.
  • Where applicable: Early-stage/forward-looking — this is where the field is going rather than where it is. Note the SPIFFE/SPIRE-shaped hole: workload-identity primitives would be the natural substrate, worth probing in Q&A. Cisco "innovations and open source efforts" means expect some product framing.