Identity and Access Management Lifecycle
A former contractor can still use service and personal credentials after central account disablement. The missing work is revocation and accountable transfer.
-
-
Standing Privilege
Short-lived tokens can still support continuous access. Distinguish credential age, grant validity, and renewal authority when reviewing standing privilege.
-
Tool Authorization and Argument Validation
A model only proposes a tool call. The runtime decides whether it is well-formed, permitted for this user, valid in the current state, and safe to run without a person.
-
Prompt Injection and Untrusted Content Boundaries
A hostile sentence in a note, passage or tool result can steer a model. Track where text came from, bound what a steered model can do, and test defenses against adaptive attacks.
-
Case: Deploying AI Development Tools
A worked, fictional rollout of an AI coding assistant. Scoped credentials, a gated execution environment, a staged CI gate, and a randomized pilot are built and run locally, and a pre-agreed gate turns the evidence into a decision.
-
Least Privilege
Understand how least privilege limits actions, resources, and duration to assigned work.
-
Coding Agent Permissions and Sandboxing
A coding agent turns model output into actions. This guide shows how isolation, scoped credentials, argument-level tool rules, audit, and limits bound what a steered agent can do.