Sathyan Nandagopal | Product Leader - Tranformations
AT&T

Identity Week America - Day 2 @ 13:20

Circuit breakers, dead man switches, and state machines: Authorization boundaries for autonomous AI agents

The identity industry is racing to provision and authorize AI agents — but who holds the authority to revoke them mid-execution, and how is that authority enforced? This talk reframes "stopping a rogue agent" as an authorization-lifecycle problem and answers it with three production-tested patterns: circuit breakers that halt execution when anomaly thresholds are breached (the WHEN of revocation), dead man switches that revoke an agent's credentials at the IAM layer when heartbeat signals go silent (the HOW), and state machines that make an agent's identity lifecycle — provisioned, active, suspended, revoked — authoritative in the IAM layer rather than the agent's own code (the WHO). Attendees leave with implementable patterns for agent containment, plus the structured-testing methodology that surfaces failure modes like these before they reach production.

Sathyan Nandagopal, Product Leader - Tranformations, AT&T

×

Sathyan Nandagopal / Circuit breakers, dead man switches, and state machines: Authorization boundaries for autonomous AI agents

Sathyan Nandagopal, Product Leader - Tranformations,
AT&T

Sathyan Nandagopal is an AI product leader who builds the systems he advises on — six production agentic-AI products shipped hands-on, author of the HHH+ agent-evaluation framework, and a senior advisor to the CDP Institute / Customer Data Alliance on its AI strategy and platform modernization. He...

CloseMore