KINGAI OPS / Security
AI may propose intent. Policy controls authority.
Build safer infrastructure operations around least privilege, approvals, typed execution, verification, rollback and audit.
Policy, Approval & AuditSample environment
Security invariants
The authority boundary is engineered, not prompted.
No unrestricted model-to-root shell
Models can explain and plan; typed executors and policy decide what authority exists.
Least privilege
Observation, low-risk work and elevated helpers are separated.
Policy + approval
Risk-classified changes can require explicit approval or be denied.
Verify + rollback
Mutable operations are designed around preflight, health gates and recovery.
Outbound connectivity
Nodes establish authenticated outbound sessions; no public management port by default.
Auditability
Every mutation should answer who, what, target, reason, result and verification state.
Change contract
Intent is not execution.
01Intenthuman · playbook · agent
02Policyrisk · scope · allowlist
03Approvalwhen required
04Executortyped capability
05Verifyhealth · audit · rollback