Clinical AI

When AI shapes care, authority must be provable before harm is alleged.

In clinical AI, authority is not a philosophical concept. It becomes evidential, legal, operational and — when a patient is harmed — adversarial. KYE Protocol™ is the identity, delegation, policy and evidence layer beneath AI-assisted clinical decisions: the layer that lets an organisation prove the authority chain from a record that existed before litigation was contemplated, not one reconstructed afterwards.

The clinical authority question

Not “what did the AI recommend?”

When an AI recommendation shapes a clinical decision, the authority question is concrete — it is the first question a plaintiff attorney asks when a patient is harmed. The defensible answer has to cover every link at once:

  • Who or what acted — the verified entity behind the recommendation, human or agent.
  • Under whose authority — the delegation that placed the AI in the decision path, and who granted it.
  • Within which clinical scope — the use case the organisation actually authorised.
  • Against which constraints — the policy and guardrails that were active at the moment of the recommendation.
  • With what oversight — the approval state and the clinician’s ability to independently review the basis of the recommendation.
  • Provable from a pre-litigation record — evidence created before harm, complaint, investigation or discovery — not after.
Contemporaneous evidence

A record that existed before litigation was contemplated.

The difference between a defensible clinical-AI decision and an indefensible one is rarely the model. It is whether the authority chain can be proven in a deposition — from a record that existed at the moment of action — rather than reconstructed once harm is alleged. KYE™ emits a signed Evidence Pack™ per governed action and a Replay-Proof™ chain derivable from public keys alone: the authority context is sealed when the recommendation is made, not assembled by counsel afterwards.

Do not reconstruct authority after harm. Preserve it at the moment of decision.

What KYE™ preserves

The clinical authority chain — sealed, link by link.

For every AI-assisted clinical decision, KYE™ preserves the full chain as structured, signed evidence:

  • Verified entity — the identity of the human or agent in the decision path.
  • Clinical role & delegated authority — the grant that authorised the AI’s involvement, and its chain back to a human principal.
  • Scope of use — the clinical use case the organisation sanctioned, bound by Purpose Permission™.
  • Model & tool capability — what the system was permitted to do, and what it was not.
  • Clinical context & policy constraints — the guardrails that were active when the recommendation was produced.
  • Approval state — the oversight and sign-off in force at the moment of action.
  • Evidence record & replayable chain — a signed Evidence Pack™ and a Replay-Proof™ chain that gives the action Authority Finality™.
Where KYE™ sits

The evidence layer underneath your clinical-AI frameworks.

KYE™ is not a clinical model and not a clinical decision support product. It is the infrastructure layer underneath them. Your governance framework defines what authority should exist across the clinical AI decision chain; your assurance framework asks for proof those constraints were present and functioning at the moment they mattered. KYE™ is the identity, delegation, policy and evidence layer that makes that chain verifiable — so both questions resolve to a record, not a narrative.

The same authority layer applies wherever an AI agent shapes a clinical decision: digital pathology, radiology and medical imaging, AI diagnostics and clinical decision support, and software functioning as a medical device (SaMD). KYE™ governs the authority around the model — not the diagnosis itself — so the provenance, scope and oversight of each AI-assisted decision are provable across every speciality. For the device-regulation view, see the MHRA Medical Devices Regulations and MHRA SaMD & AI change programme coverage maps.

Independent review

The basis of a recommendation, preserved for the clinician.

Regulators draw the line around clinical decision support partly on whether a healthcare professional can independently review the basis of a recommendation. KYE™ keeps that basis — the authority, scope, policy, context and capability around the recommendation — attached to the action as signed, replayable evidence. The clinician’s review, and the constraints they relied on, become part of the same sealed record.

Regulatory coverage

Mapped to the healthcare-AI frameworks that matter.

KYE Protocol ships per-requirement bijection maps for the frameworks UK clinical buyers + safety regulators read first:

  • HAARF v1.0 — Healthcare AI Agents Regulatory Framework. 279 requirements across 8 categories (risk lifecycle, model passport, cybersecurity, human oversight, agent registration, autonomy governance, bias/equity, tool integration). 88% weighted coverageper-requirement map →
  • MHRA Medical Devices Regulations 2002 — UK SI 2002/618 as amended. Risk classes + conformity assessment. 91% coverageper-requirement map →
  • MHRA Post-Market Surveillance Regulations 2025 — explicit PMS plan, periodic safety update, trend + incident reporting, FSCA notification. 83% coverageper-requirement map →
  • MHRA SaMD & AI Change Program (2023) — 15 work-packages covering qualification, classification, PCCP, AI Airlock, adaptive control, failure-mode analysis. 93% coverageper-requirement map →

Every requirement is bijection-mapped to a real KYE artefact (schema, engine, agent, worker, PDP, evidence pack, audit event). Live coverage dashboard →

Outcome-priced pilot · 4 weeks

KYE HAARF Readiness Pilot™.

A focused 4-week shadow-mode engagement on a single delegated clinical-AI workflow. Outputs: signed Evidence Packs™ per decision, a per-requirement HAARF v1.0 + MHRA SaMD coverage attestation, and an authority-gap report (which delegations are not currently provable from your published JWKS). Hands off cleanly to your ICB digital team or notified-body correspondent.

  • Scope — one workflow, shadow mode only. No production guardrails are toggled.
  • Frameworks bijection-mapped — HAARF v1.0 (279 requirements), MHRA MDR 2002 (23), MHRA PMS 2025 (9), MHRA SaMD & AI (15).
  • Deliverables — signed receipts, signed per-requirement attestation (90-day expiry, deterministic rebuild), clinical-safety-case binding (DCB 0129 / 0160), ICB/regulator-facing executive summary.
  • Continuity — pilot fee credits 100% against the sector healthcare licence or the high-assurance annual licence if signed within 60 days.
Outcome-priced pilot · 6 weeks · for NHS organisations

KYE NHS Readiness Pilot™.

Different from the HAARF Pilot above. The HAARF Pilot is for SaMD vendors building the AI; the NHS Readiness Pilot is for NHS organisations deploying AI — trusts, GP federations, ICBs, NHSE digital teams. 6-week shadow-mode engagement on a single NHS-deployed AI use-case, with DSPT-mapped Evidence Packs™, DCB 0129 / 0160 clinical-safety case bindings, and a Caldicott Guardian-ready summary.

  • Frameworks bijection-mapped — DSPT 2024/25, DCB 0129, DCB 0160, NHS Cyber Assessment Framework, HAARF v1.0, MHRA SaMD (where applicable).
  • Deliverables — DSPT control mapping per assertion (signed), clinical-safety case bindings, Caldicott Guardian summary, CCIO procurement template, ICB-onboarding artefact.
  • Buyer-tuned — CCIOs, Caldicott Guardians, ICB digital leads. Procurement framing, not engineering framing.
  • Continuity — pilot fee credits 100% against the sector healthcare licence or the high-assurance annual licence if signed within 60 days.
Pilot

Clinical AI does not just need explainability. It needs authority finality.

Bind KYE™ read-only to an AI-assisted clinical workflow and see the authority chain captured, sealed and replayable per action — before it is ever challenged.