Identity is the distinct account, credential, or service principal used to tell which agent performed an action. In security systems, identity is what ties a tool call, API request, file change, or approval back to a specific actor, so the activity can be logged, audited, scoped, and revoked.
Identity matters because permissions are assigned to identities, not to abstract “intelligence.” If multiple agents share one account, defenders lose attribution and attackers can hide behind broad access. In real environments, a compromised or over-privileged identity can let an agent read data, write to systems, trigger workflows, or delete records far beyond its intended task. Good defenses give each agent a separate identity, apply least privilege, use short-lived credentials, and require explicit approval for high-impact actions. That way, if an agent is misused or receives a bad instruction, the blast radius stays narrow and the action remains traceable.