# Agent Autonomy Tiering Guide

## Tier 0 — Observe
No recommendation or action authority. Minimum controls: inventory, data rules, user notice, retention.

## Tier 1 — Assist
Drafts or retrieves information for human use. Minimum controls: human review, source traceability, output labeling.

## Tier 2 — Recommend
Recommends a decision, prioritization, ranking, or route. Minimum controls: accountable decision owner, evaluation, evidence log, escalation rule.

## Tier 3 — Bounded Action
Executes approved actions within defined limits. Minimum controls: permitted-action register, owned identity, least privilege, action logs, rollback path.

## Tier 4 — Supervised Autonomous
Chains steps across systems under supervision. Minimum controls: continuous monitoring, exception triggers, emergency stop, incident playbook, executive review.

## Tier 5 — High Consequence
Material impact to rights, safety, finance, legal status, employment, public services, security, or trust. Minimum controls: accountable human authority, documented basis, appeal or review path, enhanced monitoring, and counsel review where appropriate.