AI can propose. Authority should remain explicit.

Probabilistic systems are useful for explanation, drafting, extraction, discovery and analysis. Consequential actions need a separate authority and evidence path.

Principles

• Keep source facts, derived candidates, model suggestions and human decisions distinguishable.
• Do not turn confidence language into proof.
• Preserve missing and conflicting evidence instead of filling gaps silently.
• Keep consequential execution behind an explicit authorization boundary.
• Make external-provider use visible and keep private credentials out of public artifacts.
• Evaluate generated code or recommendations against declared intent and deterministic checks where possible.

Human and system authority

AI assistance can improve speed without becoming the final authority for safety, evidence truth, user-owned data, release qualification or other consequential decisions.