What is an Agent Authentication?

This is a different problem than user authentication. A human logs in once and a session token represents them for a while; an agent may need to authenticate itself to dozens of different tools and APIs over the course of a single task, sometimes acting on its own service identity and sometimes acting "as" a delegated user. Getting this wrong in either direction is dangerous: too weak, and any process can impersonate a trusted agent; too rigid, and the agent can't do its job without a human re-entering credentials at every step. Common patterns include short-lived service tokens minted per session, OAuth delegation where the agent presents credentials on behalf of a user with a defined scope, and mutual verification where both the agent and the calling system confirm each other's identity before executing sensitive operations. The credential itself should never be something the agent can read or leak in its own output — it's injected at the runtime boundary, not embedded in the agent's context.