Research seriesL2paper
An autonomous agent is a non-human principal that provisions itself, calls other services, delegates to sub-agents, and outlives any single request — and human identity systems were never designed for it. This series builds operational identity governance for agents from the ground up: why agents break human IAM and what a machine identity actually needs, how to provision and attest an identity so its bearer can be proven rather than assumed, how to run the credential lifecycle with short-lived tokens, rotation, and bounded delegation chains at runtime, how to decommission and revoke an agent's authority cleanly and provably, and how to govern a whole fleet with an identity inventory, least-privilege drift detection, and lifecycle audit. Grounded in NIST SP 800-207 Zero Trust Architecture, NIST SP 800-63 Digital Identity Guidelines, the SPIFFE workload-identity specification, the OAuth 2.0 and Token Exchange RFCs, the OWASP Agentic Security Initiative, and the NIST AI RMF.
Autonomous agents are non-human principals that provision, delegate, and act at machine speed — and the identity systems built for people fail them in specific, nameable ways.
Before an agent can hold an identity, an authority must prove the workload is what it claims to be — the attestation problem that turns a name into a credential you can trust.
An agent's credential is not a key you hand out once but a living thing that must expire, rotate without downtime, and narrow as it passes down a delegation chain.
Short lifetimes bound how long a compromised agent stays dangerous; revocation is how you end its authority now — completely, and in a way you can prove.
You cannot secure a population of agent identities you cannot enumerate; fleet governance is the discipline of knowing every identity, catching privilege that creeps, and proving the lifecycle holds.