Skip to main content

ADR-005: Human Approval for Consequential Action

Status

Accepted — July 2026

Context

Agent and automation systems create risk when tools execute without explicit authorization, escalation, or evidence.

Decision

Public architecture for governed agents must define:

  1. What actions are consequential
  2. Who may approve them
  3. What evidence is retained
  4. How denial/missing approval blocks execution
  5. How verification confirms outcomes

Private BBIOS implements these ideas internally; public docs describe patterns without exposing private source.

Consequences

Slower autonomous demos; stronger enterprise and responsible-AI alignment.