Clinical AI, governed from the first record.
Most clinical AI was built to be fast, then asked about consent and accuracy afterwards. CIE was built the other way around. Governance isn't a setting you switch on — it's the starting line: every record carries a consent state, a time-stamped audit trail, and a clear boundary of responsibility.
Built by Design By Zen, an NZ AI lab. Evidence, not confidence.
The gap
Ambient and generative tools have moved into clinics faster than the safeguards around them.
The result is a familiar pattern: a confident output, generated quickly, recorded as fact — with the questions of did the patient consent to this, is it accurate, and who's accountable if it's wrong left to be sorted out later, if at all.
Speed without governance doesn't remove the risk in clinical AI. It just defers it to the worst possible moment — when someone asks the record to prove itself and it can't.
What "auditable" actually means
Auditable isn't a marketing word. It means a specific thing: a record you can prove is exactly what it claims to be.
For a clinical AI output, that requires three things to travel with the record, not be reconstructed after the fact —
who agreed to it (consent), when and how it was made (the trail), and whether it can be checked.
And, if wrong, shown to be wrong (falsifiability).
If any of those is missing, the record isn't auditable — it's just confident.
The three guarantees
CIE does three jobs. Each has its own page.
Every record CIE produces carries:
A consent state.
Nothing is captured, processed, or used outside what the patient agreed to — and that agreement travels with the data, not in a separate file no one can find.
An audit trail.
Each result is time-stamped and traceable, so an outcome behind a funded claim can be shown to be exactly what it represents, by whom, and when.
A falsifiable record.
Outputs are designed to be checked, not trusted on assertion. A record that could be shown to be wrong — and isn't — is stronger than one that simply sounds right. That's the difference between evidence and confidence.
Human in the loop — the LAW tier
Between the patient and any consequential output sits a human→AI→AI safeguard — the LAW (local authority weight) access tier.
No high-stakes clinical output is asserted straight from a model to the record without the governed checkpoint in between.
The point isn't to slow care down; it's to make sure the things that carry weight are the things that were checked.
The liability boundary
A governed system has to be clear about where responsibility sits — vagueness there is its own risk. CIE draws that line explicitly:
Stating the boundary plainly is part of the governance, not a disclaimer bolted on at the end.
The gap, as a contrast
AI scribes transcribe the conversation. CIE proves the outcome.
AI scribe | CIE |
|---|---|
Consent assumed or handled out-of-band | Consent state travels with every record |
Output recorded as fact | Output recorded as a checkable, falsifiable claim |
Accountability unclear after the fact | Liability boundary stated up front |
Built for speed, governed later | Governed from the first record |
Scribes aren't the villain — they're solving a narrower problem. But the gap they leave open, consent and accountability, is exactly the gap that matters most in funded, regulated care. That gap is where CIE lives.
Evidence, not confidence
This page is really one idea applied to clinical AI: in care, the currency is evidence, not confidence.
A confident wrong record is worse than no record — it accelerates the mistake. A governed, falsifiable, consent-aware record is slower to produce and far harder to argue with.
We chose the second one on purpose. It's the same standard that runs through everything Design By Zen builds. The Comfort Index →
The Comfort Index (CI), clinical edition
Everything CIE produces rests on the same governance spine:
CIE is the Comfort Index (CI) applied to clinical recovery.
The same question that runs through everything we build — how am I, and how will I know? — is, by definition, patient-reported.
PROMs are how that question is asked and answered in a clinic, measured over time. Learn about the Comfort Index →
Proof
CIE runs in clinical deployment at Node-1, Nelson — governed, stable, and in pilot clinical use since January 2026.
We describe it as deployed and governed, not as clinically proven: the independent proof ladder is work in progress, and we won't claim a result we haven't earned.
FAQ
What makes clinical AI "auditable"?
Auditable clinical AI produces records you can prove are exactly what they claim. That requires three things to accompany each record: the patient's consent state, a time-stamped audit trail, and a falsifiable output that can be verified rather than trusted on assertion.
How does CIE handle patient consent?
Consent state travels with every record in CIE. Nothing is captured, processed, or used outside what the patient agreed to, and that agreement is bound to the data rather than stored separately.
Is CIE certified or compliant for clinical use?
CIE is governed by design and in clinical deployment, but we describe its status precisely and don't claim certifications we haven't earned. The system is built around consent, audit trails, falsifiable records, and a clear liability boundary.
How is this different from an AI scribe?
A scribe transcribes the consultation, typically with consent and accountability handled outside the tool. CIE governs the record itself: consent state, audit trail, and liability boundary are built in, not bolted on.
What is the LAW tier?
The LAW access tier is a human-in-the-loop safeguard: a human→AI→AI checkpoint sits before any consequential clinical output, so high-stakes results are reviewed rather than asserted straight from a model.
What does "evidence, not confidence" mean in a clinic?
It means a record's value comes from being checkable and, if wrong, provably wrong — not from sounding certain. A confident but unverifiable output can accelerate a mistake; a governed, falsifiable one can be trusted because it can be tested.
See clinical intelligence in your clinic's workflow. Book a walkthrough — we'll show how PROMs capture, scoring, and ACC-ready filing fit your caseload.. [Book a clinic walkthrough]
From the tech labs to primetime: Trusted insights, featured stories.
