The Twelve Principles

EDENA is not AI ethics as a statement. It is AI ethics as an operating system.

Twelve principles form the backbone of the framework. Each one is operational — written to govern a decision before it executes — and each is anchored in regulation, clinical evidence, and nursing ethics.

Twelve Operational Commitments

Static ethics asks whether a system was built responsibly. EDENA asks whether this action should be allowed right now.

Fairness, transparency, privacy, and accountability matter — but they are not enough once AI can act. The twelve principles add the missing operational question and turn it into runtime governance.

01

Human judgment is non-transferable

AI can generate, summarize, recommend, and route. It cannot own judgment, consent, accountability, or final clinical authority. Those remain with the human who is answerable for the outcome.

Anchor: ANA 2025 Code of Ethics, Provision 4.2 — nurses are accountable for their practice, and AI can erode practice authority when integrated without careful consideration of harm.
02

Tier the interaction, gate the action

Risk is not an abstract label. It is a property of what is about to happen: who is asking, what data is involved, what system is touched, how reversible the action is, and what harm could follow. The interaction is tiered; the action is gated.

Anchor: Singapore MGF and EU AI Act Article 14 — oversight must be commensurate with the risk, the level of autonomy, and the context of use.
03

Ambiguity escalates upward

When risk signals disagree, EDENA selects the higher tier. In clinical environments, uncertainty is not permission to proceed; it is a reason to get another set of eyes on the decision.

Anchor: EDENA conflict rule — where signals diverge, the gate defaults to the more conservative posture rather than the more permissive one.
04

Reversibility determines risk

A reversible draft is different from an irreversible transmission, chart entry, medication action, discharge instruction, or payer submission. The cost of being wrong, and of undoing it, sets the posture.

Anchor: ANA 2025 Code, Provision 7.5 — explicitly invokes reversibility, including the ability to withdraw data permissions, as an ethical condition of responsible AI use.
05

Externality raises the floor

Anything crossing a boundary — a third-party agent, a non-certified tool, an outbound message, an external API, a payer portal, a public channel, or a cloud model — starts at a higher governance posture, regardless of how routine it appears.

Anchor: EDENA boundary rule — actions that leave the trusted, certified environment begin elevated because their blast radius is no longer locally contained.
06

The loop closes on a named human

A tier never authorizes itself. EDENA routes work to the responsible human — nurse, physician, pharmacist, allied clinician, compliance officer, privacy lead, or technical steward — who can understand, challenge, override, and own the action.

Anchor: NAIO Values Statement — the loop must close on a named human, never on a tier; accountability has to land on a person.
07

Refusal is a governance event

A safe refusal is not system failure. EDENA resolves every candidate action to one of five outcomes — ALLOW, DENY, REQUIRE_HUMAN, THROTTLE, CONSTRAIN — and a DENY is recorded, reasoned, and routed to a safer path with the same rigor as an approval.

Anchor: Runtime governance vocabulary — a DENY carries the same audit requirements as an approval; refusal is a logged governance output, not an error state.
08

Meaningful oversight must be protected

Human review is not meaningful if the reviewer is overloaded, deskilled, rushed, or reduced to rubber-stamping. EDENA designs against automation bias and protects the human's continued competence, not just their presence.

Anchor: EU AI Act Article 14(4)(b) on automation bias; NHS skill-preserving procurement UX; 2025 JAMA randomized trials on AI and diagnostic reasoning.
09

Evidence travels with the claim

A polished AI output is not trustworthy unless it is traceable. EDENA requires source grounding, provenance, timestamps, uncertainty, missing data, and contradictory evidence to move with every recommendation.

Anchor: ONC HTI-1 structured source attributes for predictive decision support; Joint Commission & CHAI evidence, validation, and monitoring requirements.
10

Privacy is dignity in operational form

PHI protection is not only compliance — it is respect for the person. EDENA treats consent, data minimization, access control, memory scope, and re-identification risk as ethical boundaries enforced at runtime.

Anchor: HHS HIPAA Security Rule — administrative, physical, and technical safeguards for electronic PHI, with ongoing risk analysis as the first step.
11

Containment must be built before scale

Agent swarms, multi-agent coordination, persistent memory, recursive optimization, and long-horizon autonomy require safeguards before deployment, not after harm. This is precisely why the Orange tier exists.

Anchor: OWASP ASI08 Cascading Failures; Singapore MGF progressive, risk-staged rollout with monitoring from day one.
12

Nurses steward the whole environment

EDENA's most distinctive contribution is nursing epistemology: whole-person, systems-aware stewardship. The nurse sees not only the output but the patient, family, workflow, unit culture, burden, and downstream consequences.

Anchor: Frontiers in Digital Health 2025 nursing AI review; ANA Provision 7.5 and new Provision 10; Gallup — nurses the most trusted profession for 24 years.
How the Principles Operate

The twelve principles are not a creed. They resolve into the tier model and the standards.

Grouped into four clusters, each principle becomes a specific runtime behavior — a way to classify, a way to escalate, a way to record, or a way to steward.

Principles 02 · 03 · 04 · 05

Classification — how risk is read

Tiering, ambiguity escalation, reversibility, and externality define how an action is assigned to Green, Yellow, Orange, or Red. They are the logic of the EDENA tier model, where the posture rises with reversibility, externality, autonomy, and human consequence.

Principles 01 · 06 · 07 · 08

Accountability — how the loop closes

Non-transferable judgment, the named-human loop, refusal-as-event, and protected oversight govern who decides and how. They are operationalized by the Human Oversight Standard and the gate's five outcomes.

Principles 09 · 10

Evidence & privacy — what the record requires

Traceable evidence and operational privacy convert transparency into artifacts and PHI handling into gated events. They are codified by the Evidence Bundle Standard and the Florence X edge-PHI pattern.

Principles 11 · 12

Containment & stewardship — what scale demands

Containment-before-scale and whole-environment stewardship answer the agentic era. They are enforced through the Agentic Systems Standard, the Orange tier, and nurse-led oversight of the environment in which AI acts.

From Principles to Practice

Principles classify. Tiers decide. Standards make it auditable.

See how the principles resolve into action-risk and capability-scale tiers — including the Orange zone — and how the EDENA standards turn each commitment into a normative requirement.