AI executes, but control always stays with the enterprise
GIIP FDE Ops never grants AI agents unlimited authority. Every action runs only within execution tiers, control mechanisms, and a responsibility split — all enforced by the FDE Ops harness.
Four execution tiers
Read-only
- Collect state, logs, metrics
- Inspect configuration
- Performance analysis
- Review cost status
- Non-impacting diagnostics
Auto-execute (approved scope)
- Run approved runbooks
- Monitoring & alerting
- Run tests
- Scale within predefined limits
- Documentation & reporting
Requires pre-approval
- Production deployment
- Data deletion
- Permission & firewall changes
- Actions above cost limits
- Irreversible changes
Prohibited (humans only)
- Exporting customer data externally
- Disabling security controls
- Creating credentials without approval
- Anything policy explicitly blocks
Control mechanisms
Command audit log
Every execution is recorded auditable, with request, plan, command, approval, result, and rollback details.
Credential management
Credentials are stored and injected securely, never exposed in code or logs.
Least privilege
Each action runs with the minimum privilege needed for its scope.
Emergency stop
On anomalies, auto-execution halts immediately and hands off to human experts.
Rollback
Changes are designed to be reversible and restored to the prior state on failure.
Human FDE escalation
Novel incidents, high-risk judgment, and architecture decisions escalate to human experts.
Responsibility split
Customer responsibility
- Define policy & approval criteria
- Decide access-grant scope
- Set business priorities
- Final approval
GIIP responsibility
- Execute within policy
- Maintain audit logs & evidence
- Anomaly detection & emergency stop
- Rollback & recovery
- Escalate to human experts
This control model is the actual operating method enforced by the GIIP FDE Ops harness; the detailed scope adapts to each customer environment and policy.
Ask about adoption & security