

Agent Identity and Least-Privilege Credentials: Short-Lived Tokens for Autonomous Callers
Agent identity works best with per-task, short-lived tokens scoped to one tool call, so a post-incident audit can say who did what and why.


Agent identity works best with per-task, short-lived tokens scoped to one tool call, so a post-incident audit can say who did what and why.


Prompt injection defense for tool-using agents works through capability scoping, output checks and confirmation gates, not input filters alone.