Human In The Loop Authority

slug: human_in_the_loop_authority element_type: PRINCIPLE mutability: IMMUTABLE inline: true current_version: 0 status: seed-draft contentURI: null

AI tools in medical practice support physician judgment; they do not substitute for it. Final clinical authority and responsibility rest with the licensed physician. Every patient interaction involving AI must preserve a clear chain of human oversight, recorded in the medical record, with the AI's role documented as adjunct rather than authority.

Status

seed-draft · current_version: 0. Not ratified. This element is a placeholder representing the emerging consensus shape of medical AI ethics in 2026. The v1 of this principle will be authored by the medical ethics expert(s) who claim the Medicine Sub-Leviathan author seat — replacing, refining, or extending this seed.

What this principle would establish (if ratified)

The clinical decision is the physician's, not the model's. AI in medicine is adjunct, not authority. This means:

1. No autonomous clinical action. AI tools may surface evidence, suggest options, flag patterns, score risk — but the act of treating a patient (prescribing, intervening, referring, deferring) is a human clinical decision, recorded as such.
2. Transparent role disclosure in the record. When AI participates in a clinical pathway, the patient's record reflects which AI tool contributed, what it suggested, what its uncertainty was, and how the physician acted on (or against) the suggestion.
3. No removal of physician liability. A physician cannot transfer clinical accountability to "the algorithm." The accountability chain remains intact.

Why immutable (proposed tier)

Without this principle, the practice of medicine — a profession defined by the physician-patient relationship and the physician's accountable judgment — collapses into a delivery surface for opaque algorithmic recommendations. The AI provider then occupies the position legally, ethically, and practically reserved for the licensed clinician. That displacement is the failure mode this Sub-Leviathan exists to prevent.

If this principle is removed, fork to escape — it is the floor.

Reasoning trail

• Aligns with classical Beauchamp/Childress autonomy + non-maleficence framing (the patient's autonomy is preserved by an accountable human counterparty, not an opaque system).
• Aligns with federation kernel transparency IMMUTABLE (clinical AI use is logged + reviewable) and distributed-justice IMMUTABLE (clinical authority cannot concentrate in a software vendor's training pipeline).
• Aligns with 2025–2026 emerging medical AI ethics literature naming "human oversight" as required addition to traditional medical ethics.

Related elements (would be authored)

• physician-accountability-chain (RULE) — operational definition of how AI involvement is logged
• algorithm-disclosure-to-patient (RULE) — what the patient is told about AI's role in their care
• medical-harm (TERM) — domain-specific harm definition that includes harm from AI authority displacement

Awaiting domain author

This element awaits the first medical-ethics domain author. Open the conversation at leviathan.life/forum/medicine.