Trust / Responsible autonomy
Autonomy bounded by explicit checkpoints
Autonomy is not the absence of human involvement; it is a system that knows exactly when to ask. Omnacta OS defines fixed escalation triggers rather than leaving judgment calls to an individual agent's discretion.
Escalation triggers below are implemented for the engineering reference workflow. Cross-domain escalation policy is in development.
Fixed escalation triggers
These triggers are checked structurally, not left to the executing agent to decide case by case.
| Trigger | Response |
|---|---|
| Authority boundary reached | Mission pauses and requests an explicit grant before proceeding. |
| Ambiguous acceptance criteria | Mission surfaces the ambiguity for clarification rather than guessing at intent. |
| Conflicting evidence | Mission blocks and reports the conflict instead of picking one signal silently. |
| External-effect authority requested | Requires the strictest approval gate regardless of mission progress so far. |
Authority levels, recap
Observe, propose, execute isolated, integrate governed, and external effects. Each level is a distinct scoped grant. The full definitions live on the Execution page.
What the human role actually is
A human sets acceptance criteria, grants authority, and makes the final accept/reject decision on delivered evidence. The system does not accept its own work on the user's behalf.
- Escalation policy currently covers the engineering Mission Pack; other domains will define their own trigger sets.
- There is no current claim that autonomous acceptance without human review is available for consequential outcomes.
Back to the trust overview
Review all four trust pillars together, or go deeper on any one of them.