CSE Roadmap

Core tier · about 4 hours · 13 parts

Part 11 of 13 · 1 min read

Quick notes

  • Least privilege is the main control, because it works when the model fails. Grant only the specific tools a task needs, never a general shell, and run agents on separate infrastructure.
  • Excessive agency is LLM06 in the OWASP Top 10 for LLM Applications: the agent can do more than its task needs, through excessive functionality, permissions or autonomy.
  • Blast radius is computable. Against the same compromised agent, one shared service account did hundreds of times more damage than per-task grants in the example, and it reached far more capabilities by escalation.
  • Grants should be narrow in five ways: resource, action, amount, time and number of uses, with deny by default and attenuation (a derived grant can only be narrower).
  • Least privilege has a utility cost you must measure. The tasks a narrow grant blocks (a refund above the cap, say) should go to a human approval, not to a wider grant. Autonomy is a privilege too.