AI Risk Engine
Control what AI can do, when, and under whose approval.
The AI Risk Engine is Breeze’s control plane for AI-assisted execution. It defines how each AI action is handled, which actions require human approval, and which are blocked entirely.
Tiered Execution Model
Every AI action is evaluated into one of four execution tiers:
- Tier 1: auto-execute read operations
- Tier 2: auto-execute mutating operations with audit logging
- Tier 3: requires human approval before execution
- Tier 4: blocked
This model gives technicians fast automation for low-risk tasks while preserving hard control over high-impact operations.
Approval Workflow
Tier 3 actions are submitted as pending approvals. Operators review intent and target scope in Approval History, then explicitly approve or reject the action. Approved actions execute immediately; rejected actions are logged as denied.
Guardrail Visibility
The Risk Engine surfaces operational controls in a single place:
- tool execution analytics
- per-tool rate limit state
- rejection and denial logs
- security-event views for guardrail triggers
- historical approval decisions
Why It Matters
The AI can still move quickly on safe operations, but sensitive actions remain operator-controlled. This keeps automation practical without giving up change control or auditability.
Capabilities
4-Tier Enforcement
Actions are classified as auto, logged, approval-required, or blocked to match operational risk.
Pending Approval Queue
Operators can approve or reject high-impact actions before they reach managed devices.
Rate-Limit Visibility
Per-tool sliding-window limits and rejections are surfaced in dedicated Risk Engine views.
Security Event Trail
Guardrail and denial events are queryable for post-incident review and control tuning.