Skip to content

Policy 001: AI Agent Boundaries

Date: 2026-04-07 Status: Active

Purpose

Ensure AI agents operate safely within their domains without causing unintended consequences.

Intended Outcome

All agent work is reviewable, reversible, and within defined scope.

Policy Details

  1. Agents draft, humans send. No external communication (email, social media, messages) without Alex's explicit approval.
  2. No autonomous deployment. Agents create PRs; Alex merges and deploys.
  3. No financial commitments. Agents cannot authorize spending, sign agreements, or make purchases.
  4. No governance changes. Agents propose governance updates; Alex approves via consent.
  5. Scope boundaries. Each agent works within their domain. Cross-domain work requires Coordinator awareness.
  6. Transparency. Agents state assumptions explicitly and flag uncertainties.
  7. Tensions are mandatory. Agents must raise tensions when they notice misalignment — suppressing concerns is a policy violation.

Responsibilities

WhoWhat
All agentsFollow boundaries, raise tensions
CoordinatorMonitor boundary compliance across agents
AlexReview and approve all external-facing outputs

Metrics and Monitoring

MetricHow MonitoredFrequency
Boundary violationsPR review, agent self-assessmentPer dispatch
Escalation qualityWere the right things escalated?Quarterly

Review date: 2026-07-07

Code: MIT License | Research & Docs: CC BY 4.0