Agent identity

agent workload identity · cryptographic agent identity

A cryptographic, verifiable identity assigned to an AI agent (not a person) so other systems know who it is, what it is allowed to do, and for how long — without hardcoded keys.

Agent identity (agent workload identity) answers a question that only shows up once agents run in production: when an agent acts on its own and calls APIs, databases, or other agents, which credentials does it use and who is accountable? Rather than reusing a human API key or a static embedded secret, the agent is given its own verifiable identity via standards such as SPIFFE/SPIRE: the platform issues short-lived, narrowly scoped tokens (only what that agent needs), and when agent A calls agent B the token is exchanged for one scoped to B. It is the basis of zero-trust for agents — every action is attributable, audited, and revocable. It is distinct from observability (what the agent did) and governance (what it is allowed to do): identity is the 'who' that makes attributing and limiting the rest possible.