11:08▪▪▪ ⌶ 73
AUTONOMY MODEL

Three-Tier Autonomy

Every agent declares the level of authority it operates with. Nothing acts unbounded.
LEVEL 0
Advisory
SAFEST

Agent proposes; the operator decides. The agent generates options, recommendations, and reasoning. No external action is taken.

LEVEL 1
Approval Required
DEFAULT

Agent prepares; the operator approves. Drafts are queued for review, then executed on consent. Output goes nowhere without a deliberate yes.

LEVEL 2
Auto-Execute
BOUNDED

Agent acts within a tightly scoped contract. Only routines proven safe over time graduate here. Every action is logged and auditable.