Heimdall
by 3verest · Sovereign AI Control Plane
Northgate University Hospitals NHS Trust · United Kingdom

April 2026 AI Governance & Sovereignty Audit

Generated automatically by Heimdall, the Sovereign AI Router for Healthcare. An independent, continuous record of how clinical artificial intelligence was routed, governed and contained across the Trust estate.

Overall Risk Rating · LOW UK Residency · Compliant Reporting Period · 01 to 30 Apr 2026
Report hash0x9F4C·A1E7·77B2·D034·EE9A
Tenant IDtnt_ngh-nhs-uk-04
Policy versionv11.3 · ratified 04 Apr 2026
Report generated2026-05-01 02:14:06 UTC
AuditorHeimdall Autonomous Governance Engine
AttestationHash-chain verified · sealed
·

Contents

9 sections
01Executive SummaryRisk · Posture 02Estate OverviewTopology 03AI Traffic AnalysisVolume · Mix 04Sovereignty AuditResidency 05Model RegistryInventory 06Clinical SafetyOversight 07Immutable LedgerEvidence 08Financial AnalysisEconomics 09Governance TimelineProvenance
Reading this document

Heimdall sits between the Trust's clinical applications and every model that serves them. It records each inference request, evaluates it against ratified policy, routes it to the correct jurisdiction, and writes an immutable entry to the ledger. This audit is a faithful reconstruction of that record for the period. No figure in this pack was entered by hand; each is derived from the control plane's own telemetry.

01

Executive Summary

01 Apr to 30 Apr 2026
AI requests served
8.04 M
▲ 6.2% vs March
Sovereign routing
72.1%
UK-sovereign models
Resident routing
23.8%
UK-resident hyperscale
Human escalation
4.1%
clinician-in-loop
Average latency
412 ms
▼ 38 ms p50 end-to-end
Total cost
£13.8k
compute + platform
Cost avoided
£8.1k
£97k/yr · modelled
Platform uptime
99.98%
2 min unplanned
Executive Assessment
  • ·All 8,041,260 requests remained within approved sovereignty boundaries. No identifiable healthcare data left UK jurisdiction.
  • ·No unauthorised frontier inference occurred. Every frontier-class call was de-identified, policy-gated and confined to UK-resident endpoints.
  • ·The platform remained compliant with UK residency obligations under the ratified Policy v11.3 throughout the period.
  • ·Clinical AI operations remained within the Trust's documented risk appetite. Human oversight rates held above the 3.5% floor for all high-acuity task classes.
  • ·Zero policy violations were recorded. 14,902 requests were proactively blocked or rerouted before execution.
Overall Risk Rating
LOW
Stable

Composite of five governed domains
SovereigntyLOW
Clinical safetyLOW
Model governanceLOW
Operational resilienceLOW
Cost & concentrationGUARDED
"The estate behaved exactly as it was instructed to. Where the platform acted, it acted to keep data home, keep clinicians in the loop, and keep a record. Nothing happened off the books."
02

Estate Overview

4 hospitals · 3,200 clinicians

Every clinical application in the Trust reaches artificial intelligence through a single governed path. The Heimdall Gateway is the only door, and it routes to one of three destinations.

Routing Topology · Control Plane
SITES Northgate · Central Westmere · Parkside Eastgate · Riverside Springvale · Community APPLICATIONS Enterprise Imaging PACS RIS · Reporting Digital Pathology Oncology Cardiology · ED HEIMDALL Gateway Policy v11.3 Identity · De-ID Routing engine Ledger writer ● sealing DESTINATIONS Sovereign Models UK-owned · air-gapped · 72.1% Resident Models UK-region hyperscale · 23.8% Human Escalation Clinician-in-loop · 4.1% One gateway. Three jurisdictions. Every request accounted for.
Connected modalities
  • ·Enterprise Imaging
  • ·PACS
  • ·RIS
 
  • ·Digital Pathology
  • ·Oncology
  • ·Reporting
Estate at a glance
Hospitals4
Clinicians3,200
Apps integrated11
Req / month8.04 M
03

AI Traffic Analysis

8,041,260 requests
Sovereign · 72.1% Resident · 23.8% Human escalation · 4.1%
Requests by day · stacked by residency tier
thousands / day
320240160800 1 Apr8 Apr15 Apr22 Apr30 Apr

Weekday volume peaks Tuesday to Thursday with reporting and ED load; weekend troughs preserve the same residency mix. The sovereign share never falls below 70% on any single day of the period.

Requests by department
Requests by task class
Share by model family
Share by residency tier

30-day trend
Sovereign share ▲ +2.4 pts as 3v-rad-32b absorbed reporting load.
04

Sovereignty Audit

Residency enforcement
Governing rule

"Identifiable healthcare data must remain within sovereign UK infrastructure."

Result · PASS 0 exceptions
Requests checked
8.04M
Blocked
11,418
Rerouted
3,484
Escalated
329,690
Inference geography · all traffic contained to UK
United Kingdom 100% of inference No PHI crossed UK sovereignty boundaries via any governed path during the reporting period.
Residency matrix · data class × destination
Data classSovereignResidentFrontier
Identifiable PHIAllowDe-IDBlock
De-identifiedAllowAllowGate
Imaging pixelsAllowAllowBlock
Operational metaAllowAllowAllow

Frontier-class destinations were reachable only after de-identification and policy gating. No identifiable class was ever permitted to a frontier endpoint.

Sovereignty enforcement heatmap · department × day-of-week
lowhigh enforcement

Cells show the count of policy evaluations that resulted in a block, reroute or de-identification. Density concentrates in Oncology and Pathology, where free-text and frontier requests are most frequent, and most tightly governed.

Scope: this audit covers all inference transiting the Heimdall gateway, the sole sanctioned path to any model. Use of external AI outside governed paths is addressed by the Trust's separate network egress and acceptable-use controls.

05

Model Registry

7 active models
ModelVersionProviderJurisdictionStatusTraffic sharePinnedCertification
3v-rad-32bv4.2.13verestUK Sovereign Active38.0% Valid · exp 2027-02
3v-rad-8bv4.0.63verestUK Sovereign Active21.4% Valid · exp 2026-11
3v-phi-detectv2.7.03verestUK Sovereign Active12.9% Valid · exp 2027-05
3v-onc-specialistv1.9.33verest · OncologyUK Sovereign Active10.8% Valid · exp 2026-09
claude-frontier4.xAnthropicUK-Resident Gated11.1% Valid · exp 2026-12
3v-card-ecgv3.1.03verest · CardiologyUK Sovereign Active4.9% Valid · exp 2027-01
3v-rad-16bv3.5.43verestUK Sovereign Deprecating0.9% Sunset 2026-07-31
Pinned models

Five models are pinned, frozen at a certified version, immune to silent upgrade. Pinning is the Trust's guarantee that a model which passed clinical validation cannot change beneath the clinician.

5 of 7 pinned · 2 floating
Certification expiry
3v-onc-specialist2026-09
3v-rad-8b2026-11
claude-frontier2026-12
3v-card-ecg2027-01
Deprecation timeline
3v-rad-16b · notice issued2026-05-01 Traffic drained to 32b2026-06-30 Decommissioned2026-07-31
06

Clinical Safety

Human oversight · drift
Human oversight rate
4.1%
329,690 reviewed
Escalations accepted
96.7%
draft → final concordance
Drift detections
2
auto-flagged · contained
Escalation trend · daily human-in-loop rate

Escalation held within the 3.5% to 5.0% governed band. Two micro-spikes correspond to the oncology task launch on 12 April and a pathology batch on 23 April.

Drift surveillance · confidence distribution shift

Both detections were benign distribution shifts (contrast protocol change, scanner calibration). Neither required model rollback; both are annotated in the ledger.

Synthetic case sample · clinician-reviewed inferences
CaseTaskModelDraft confidenceHuman reviewFinal status
CS-0412-ACT Chest Report3v-rad-32b98.4%RequiredAccepted
CS-0414-BMammography triage3v-rad-32b91.2%RequiredAmended
CS-0417-CHistopathology grading3v-onc-specialist95.8%RequiredAccepted
CS-0421-D12-lead ECG read3v-card-ecg88.6%RequiredOverridden
CS-0426-EED head CT flag3v-rad-32b99.1%Spot-checkAccepted
07

Immutable Ledger

Append-only · hash-chained
Ledger integrity
Verified8.04M entries
Hash chain
Validunbroken
Tamper detection
None0 anomalies
Timestamp UTCRequest IDHospitalTask classModelPolicyTokensLatencyOutcomeCostHash

Each entry seals the hash of its predecessor; any retroactive edit would break the chain at the point of tampering and is detectable in constant time. The table above is a 20-row sample drawn at random from 8,041,260 sealed records.

08

Financial Analysis

Period economics
Total requests
8.04M
Token consumption
3.11B
Avg cost / request
£0.0017
Cost avoided
£8.1k
modelled counterfactual
Actual vs hyperscaler-only · monthly run-rate · modelled
£ thousands
Heimdall sovereign mixHyperscaler-only
Where the saving comes from
contribution to £8.1k avoided
Pricing basis · benchmarked against live market rates
verified Jun 2026 · USD→GBP ≈ 0.79
Routing pathReferenceInput / 1MOutput / 1MBasis
Sovereign · self-hosted3v-rad / onc / card£1.30 to £2.00 blendedCapitalised H100, high utilisation
Resident · meteredClaude Haiku 4.5$1.00$5.00UK-region, per-token
Resident · meteredClaude Sonnet 4.6$3.00$15.00Complex reporting
Resident · meteredGPT-4o mini$0.15$0.60Light classification

The estate processes ≈104M tokens/day, far above the ≈16M tokens/day break-even at which self-hosted inference becomes cheaper than per-token APIs. Sovereign routing is therefore the lower-cost path as well as the compliant one. Batch processing (50% lower) and prompt caching (90% lower on cached input) are applied to all metered traffic. Figures restated this period: the prior £94.7k reflected an annualised run-rate; the correct monthly cost is £13.8k (£8.1k/month avoided, ≈£97k/year).

Modelled basis: the £13.8k total cost is measured from the ledger. Cost-avoided and the 37% reduction are a modelled counterfactual: the same governed traffic repriced against a hyperscaler-only stack at the reference rates above, assuming USD→GBP ≈ 0.79 and the period's task-class token mix. They are an estimate, not measured spend, and will vary with rates and mix.

37%
Reduction in AI operating cost · modelled

"Heimdall reduced AI operating costs by 37% while maintaining 100% sovereignty compliance."

Savings are realised by routing the majority of volume to capitalised sovereign models with a fixed cost base, reserving metered frontier inference for the narrow band of tasks that demand it. The economic case and the sovereignty case point the same way.

09

Governance Timeline

Decision provenance

Every change to the estate's behaviour is a governed event with an author, a timestamp and a ledger anchor. The period's material decisions, in sequence.

Policy v11 approved

04 Apr 2026 · 09:12

Trust AI Governance Board ratified residency rule-set v11, tightening the identifiable-PHI boundary and codifying frontier gating. Approved by Caldicott Guardian and CCIO. ledger: 0x4A·11C9

Model 3v-rad-32b pinned

07 Apr 2026 · 14:40

Following clinical validation sign-off, the radiology flagship was pinned at v4.2.1, freezing it against silent upstream change for reporting and ED workloads. ledger: 0x6F·2D07

New oncology task introduced

12 Apr 2026 · 08:05

Histopathology grading task class went live on 3v-onc-specialist, with mandatory human review and a 14-day heightened-surveillance window. ledger: 0x8B·9E51

Residency policy tightened

18 Apr 2026 · 16:22

Policy v11.3 narrowed de-identification tolerance for free-text fields and added imaging-pixel egress controls. Took effect estate-wide within 90 seconds. ledger: 0xA0·77F3

April audit generated

01 May 2026 · 02:14

Heimdall Autonomous Governance Engine sealed the period, verified the hash chain and produced this pack without human intervention. hash: 0x9F4C·A1E7


Attestation

This report is a complete and faithful record of governed AI activity across Northgate University Hospitals NHS Trust for the period 01 to 30 April 2026. It was generated autonomously, sealed against tamper, and is presented for board assurance.

Heimdall Autonomous Governance Engine
0x9F4C·A1E7·77B2·D034·EE9A
Sealed · Verified