# Safety and Guardrails - Profasee Ultra

Canonical page: https://profasee.com/safety/

## Safety model

Profasee is designed to start in observe mode. AI employees expand authority only after the operator is comfortable with the output.

## Progressive autonomy

- Level 1: Recommendations only
- Level 2: Ask me first
- Level 3: Limited handling
- Level 4: Broader handling
- Level 5: Full autonomy inside guardrails

## Core controls

- Hard guardrails for spend caps, price floors, bid limits, and blocked assets
- No-fly zones for specific ASINs, keywords, launches, or time windows
- Confidence thresholds that route low-confidence actions to review
- Double-checks before price, bid, or budget changes go live
- Anomaly pause when conditions drift
- Rollback with preserved history
- Fresh-data checks when signal quality is weak

## Failure behavior

- Detect: weak or unsafe signals are caught early
- Contain: authority narrows, actions pause, or moves route to approval
- Recover: rollback points are preserved and autonomy can be downgraded

## Data and auditability

- Workspace isolation by brand
- No cross-brand reuse of account data
- Structured action logs and append-only history
- Exportable approvals, pauses, reversals, and outcomes

## Commercial reassurance

- Run alongside an agency or internal team during rollout
- Start with recommendations or approval-required mode
- Intended failure state: no action or narrower authority, not blind automation

## Key links

- Safety page: https://profasee.com/safety/
- Pricing: https://profasee.com/pricing/
- Results: https://profasee.com/results/
- Apply: https://profasee.com/apply/