Engagement · Meridian Trust Bank

Audit overview and Twin coverage

Autonomy: L2
Meridian Trust Bank1 Nov 2025 – 31 Oct 2026
Phase 1 Control Coverage
15/ 25
7 partial · 3 manual only
Evidence Collected
58%
7 of 12 requests accepted
Findings Drafted by Twin
5
0 confirmed by auditor yet
Clauses 4–10 Skills
7/ 7
Internal audit, risk, SoA, leadership...
Autonomy Level
L2
Recommend · human decides (§10)

Coverage Page (P1-05)

Which of the 25 Phase 1 controls the Twin can audit fully, partly, or not at all. "Manual only" means the Twin abstains and a human executes the checklist.

ControlTitleThemeTwin SupportTest StyleSkill Pinned
A.5.1Policies for information securityOrganizationalFull skillDocument review—
A.5.2Information security roles and responsibilitiesOrganizationalPartialDocument, interview—
A.5.9Inventory of information and associated assetsOrganizationalFull skillData analytics—
A.5.12Classification of informationOrganizationalPartialDocument, sample—
A.5.15Access controlOrganizationalFull skillDocument, sample—
A.5.16Identity managementOrganizationalFull skillData analytics—
A.5.18Access rightsOrganizationalFull skillData analyticsTECH-5.18-ACCESS-RIGHTS
A.5.19Information security in supplier relationshipsOrganizationalManual onlyDocument, sample—
A.5.23Information security for use of cloud servicesOrganizationalPartialDocument, data—
A.5.24Incident management planning and preparationOrganizationalFull skillDocument review—
A.5.26Response to information security incidentsOrganizationalPartialSample, data—
A.5.30ICT readiness for business continuityOrganizationalPartialDocument, test results—
A.5.31Legal, statutory, regulatory and contractual requirementsOrganizationalManual onlyDocument, interview—
A.5.36Compliance with policies, rules and standardsOrganizationalFull skillDocument review—
A.6.1ScreeningPeopleFull skillData analytics—
A.6.3Awareness, education and trainingPeopleFull skillData analytics—
A.7.2Physical entryPhysicalManual onlySample, walk-throughTECH-7.2-PHYSICAL
A.8.2Privileged access rightsTechnologicalFull skillData analytics—
A.8.5Secure authenticationTechnologicalFull skillConfig, dataTECH-8.5-AUTH
A.8.7Protection against malwareTechnologicalFull skillData analyticsTECH-8.7-MALWARE
A.8.8Management of technical vulnerabilitiesTechnologicalFull skillData analyticsTECH-8.8-VULN-MGMT
A.8.9Configuration managementTechnologicalPartialConfig, data—
A.8.13Information backupTechnologicalPartialData, sample—
A.8.15LoggingTechnologicalFull skillConfig, dataTECH-8.15-LOGGING
A.8.32Change managementTechnologicalFull skillData analytics—

Why This Build Matters — BRD Promises Mapped to Screens

Every claim in the demo traces back to a requirement in the business specification.

BRD ReqPromiseWhere You See It
SE-07/08Skills are versioned data, composed at runtimeSkill engine → composition card (A.5.18 + banking pack)
AE-03Full-population testing, not samplesFieldwork → 212 leavers matched to disable log
FR-01/02Findings with all required parts, gradedFindings → 5 drafts with condition, criteria, evidence, scope
FR-04Warn when evidence is weakFindings F-002 and F-003 carry Twin warnings
BR-1/BR-5Cite or abstainAsk the Twin → logging and physical-entry answers abstain
AE-10/BR-2No finding becomes official without a humanFindings → confirm / reject, then Report
SL-03Corrections captured with reason codesLearn & correct → correction queue → skill update
WF-07Evidence is checked at uploadEvidence portal → instant intake verdicts

Engagement Parameters

Client Organization
Meridian Trust Bank (anonymised pilot · data is synthetic)
Audit Scope
Head office, 2 data centres, 3 outsourced SaaS providers
Framework
ISO/IEC 27001:2022 · ISO/IEC 27002:2022
Audit Window
1 Nov 2025 – 31 Oct 2026
Domain Packs Active
Banking pack v0.9Cloud pack v0.4
Auditor Lead
Lead auditor: A. Menon

Ask the Twin

grounded · cite or abstain

Answers can only use the client's verified evidence and active skills. Missing evidence results in an honest abstention.

POC Honesty Box

This demo reflects the Phase 1 ISO 27001 AI Auditor Twin. Retrieval, extraction and grading follow deterministic rules and verified prompt guards. In accordance with ISO/IEC 42001 ethics guidelines, AI assistance is disclosed on all outputs.