diff --git a/rag-agentic-dashboard/data/gsifi-aims-blueprint.json b/rag-agentic-dashboard/data/gsifi-aims-blueprint.json
new file mode 100644
index 0000000..a3af74b
--- /dev/null
+++ b/rag-agentic-dashboard/data/gsifi-aims-blueprint.json
@@ -0,0 +1,1612 @@
+{
+ "meta": {
+ "docRef": "GSIFI-AIMS-BLUEPRINT-WP-037",
+ "version": "1.0.0",
+ "date": "2026-04-30",
+ "title": "Regulator-Grade AI Governance & ISO/IEC 42001 AIMS Master Blueprint for G-SIFIs (2026-2030)",
+ "subtitle": "Design and implementation roadmap for ISO/IEC 42001-aligned AI Management Systems, multi-jurisdiction regulatory overlays, Regulator Submission Packs (RSP v1.0-v2.6), Terraform/OPA technical enforcement, adversarial and self-healing governance loops, predictive governance with formally-verified legal logic, cross-regulator federation, and autonomous supervisory ecosystems for high-risk credit underwriting.",
+ "classification": "CONFIDENTIAL \u2014 Board / Prudential Regulator / Group Risk / Internal Audit / Chief Legal & Compliance Officer",
+ "owner": "Group CRO + Chief AI Officer (CAIO) \u2014 co-signed by CCO, GC, CISO, DPO, Head of Internal Audit",
+ "audience": [
+ "Board of Directors / Risk Committee / Audit Committee",
+ "Executive Committee (CEO, CFO, CRO, CCO, CISO, CAIO, CTO)",
+ "Group Compliance, Legal & Privacy Office",
+ "Internal Audit (3rd Line of Defense)",
+ "Model Risk Management (MRM, 2nd Line of Defense)",
+ "Prudential supervisors (ECB SSM JST, Federal Reserve, PRA, OCC)",
+ "Conduct supervisors (FCA, BaFin, AMF, CFPB)",
+ "Data protection authorities (EDPB, ICO)",
+ "AI safety / standards bodies (AISI, ISO/IEC JTC1 SC42)"
+ ],
+ "horizon": "2026-2030",
+ "outlookHorizon": "2030-2035 (autonomous supervisory ecosystems)",
+ "subjectSystem": {
+ "institutionType": "G-SIFI / G-SIB (FSB list, Bucket 1-4)",
+ "scopeOfAi": "All AI systems materially impacting capital, liquidity, credit, market conduct, AML, fraud, and customer outcomes",
+ "anchorUseCase": "AI-CR-UNDERWRITE-01 \u2014 High-risk retail & SME credit underwriting (EU AI Act Annex III \u00a75(b) \u2014 high-risk)",
+ "scale": "20+ jurisdictions \u00b7 1,200+ AI systems \u00b7 350+ models in production"
+ },
+ "regulatoryAlignment": [
+ "ISO/IEC 42001:2023 \u2014 AI Management System (AIMS) \u2014 primary anchor",
+ "ISO/IEC 23894:2023 \u2014 AI Risk Management",
+ "ISO/IEC 5338:2023 \u2014 AI System Life Cycle Processes",
+ "ISO/IEC 27001:2022 / 27701:2019 / 27018:2019",
+ "ISO/IEC TR 24028 / 24029 / 24368 (trustworthiness)",
+ "EU AI Act (Reg. (EU) 2024/1689) \u2014 Art. 6, 9, 10, 12, 13, 14, 15, 17, 26, 27, 49, 53, 55, 72, 73; Annex III \u00a75(b)",
+ "GDPR (Reg. (EU) 2016/679) \u2014 Art. 5, 6, 9, 22, 25, 32, 33, 34, 35",
+ "ECB SSM Guide on internal models (2024) + Targeted Review of Internal Models (TRIM) AI extensions",
+ "Federal Reserve SR 11-7 / OCC 2011-12 \u2014 Model Risk Management",
+ "PRA SS1/23 \u2014 Model Risk Management Principles for Banks (UK)",
+ "PRA SS2/21 \u2014 Outsourcing & third-party risk management",
+ "FCA Consumer Duty (PS22/9) + AI/ML discussion paper DP5/22",
+ "Basel III/IV \u2014 CRR3 / CRD6 \u2014 ICAAP Pillar 2 AI add-on",
+ "FCRA (US) \u00a7604/\u00a7615 + ECOA / Reg B \u00a71002 (adverse action)",
+ "CFPB Circular 2023-03 (algorithmic adverse-action notices)",
+ "NIST AI RMF 1.0 + GenAI Profile (AI 600-1)",
+ "OECD AI Principles + G7 Hiroshima AI Process Code of Conduct",
+ "Council of Europe Framework Convention on AI (2024)",
+ "OWASP LLM Top 10 (2025) / MITRE ATLAS",
+ "SLSA L3 + Sigstore/Cosign + in-toto attestations"
+ ],
+ "deliverableInventory": {
+ "modules": 12,
+ "aimsSections": 5,
+ "annexes": 4,
+ "regulatoryOverlays": 5,
+ "rspVersions": 7,
+ "schemas": 8,
+ "codeExamples": 11,
+ "caseStudies": 5,
+ "phases": 5,
+ "kpis": 16,
+ "controls": 280
+ }
+ },
+ "executiveSummary": {
+ "purpose": "Provide G-SIFI boards, regulators, and supervisors a regulator-grade, ISO/IEC 42001-anchored master blueprint that operationalises AI governance across all jurisdictions in which the institution operates, with machine-checkable legal logic and autonomous supervisory federation by 2030.",
+ "scope": "End-to-end design, implementation, and continuous-supervision framework for an AI Management System (AIMS) covering all material AI systems \u2014 anchored on the AI-CR-UNDERWRITE-01 high-risk credit use case.",
+ "designPrinciples": [
+ "ISO/IEC 42001 as the operating standard, regulator overlays as policy bundles",
+ "Compliance-as-code: every control has Terraform + OPA enforcement",
+ "Decision-traceability: every model decision is reproducible from a signed envelope",
+ "Self-healing governance: detect-then-remediate loops with cryptographic evidence",
+ "Predictive governance: forecast control breaches before they occur",
+ "Formally-verified legal logic: TLA+/Lean specs of obligations",
+ "Federation by default: cross-regulator API with consented disclosure",
+ "Adversarial assurance: continuous red-teaming of both models and controls"
+ ],
+ "headlineKpis": {
+ "timeToRegulatorApprovedDeployment": "<= 14 days (RSP v2.4+)",
+ "rspGenerationLatency": "<= 30 minutes (auto-assembled, signed)",
+ "decisionTraceabilityCoverage": ">= 99.95% of AI decisions",
+ "controlAutomationRate": ">= 95% (Terraform + OPA enforced)",
+ "evidenceAutomation": ">= 96% (no human evidence collection for L1/L2 controls)",
+ "fairnessAirFloor": ">= 0.85 (FCRA / ECOA / EU AI Act Art. 10)",
+ "explainabilityCoverage": "100% of high-risk decisions have SHAP + counterfactual",
+ "adverseActionNoticeSla": "<= 30 days (FCRA \u00a7615) \u2014 automated for 100% cases",
+ "incidentNotifSlaRegulator": "<= 24h (EU AI Act Art. 73) / 72h (GDPR Art. 33)",
+ "modelInventoryCoverage": "100% \u2014 no shadow AI tolerance",
+ "policyDriftMtta": "<= 5 minutes (Terraform plan diff)",
+ "autonomousSupervisorReadiness": "Tier-3 by 2030 (machine-readable filings)",
+ "boardAttestationCadence": "Quarterly + ad-hoc on Sev-1",
+ "auditFindingCloseRate": ">= 95% within SLA",
+ "wormRetention": "10 years (extends SR 11-7 / SEC 17a-4(f) baseline)",
+ "crossRegulatorFederationCount": ">= 8 supervisors integrated"
+ },
+ "boardNarrative": "This blueprint converts AI governance from a periodic compliance exercise into a continuously-attested, regulator-federated operating discipline \u2014 measurable, monitorable, and provably correct against the EU AI Act, ISO/IEC 42001, ECB, Fed, PRA, and GDPR by design."
+ },
+ "M1_aimsSections": {
+ "id": "M1",
+ "title": "M1 \u2014 ISO/IEC 42001 AIMS Documentation (Sections 1\u20135)",
+ "summary": "Master AIMS documentation set anchored on ISO/IEC 42001:2023 clauses 4\u201310, broken into Sections 1\u20135 with audit-grade detail.",
+ "sections": [
+ {
+ "id": "M1-S1",
+ "title": "Section 1 \u2014 Context of the Organization (Cl. 4)",
+ "iso42001Clauses": [
+ "4.1",
+ "4.2",
+ "4.3",
+ "4.4"
+ ],
+ "deliverables": [
+ "Internal/external issues register (PEST + tech + regulatory)",
+ "Interested parties matrix (regulators, customers, employees, society)",
+ "AIMS scope statement (geographies, business units, AI systems)",
+ "AI System Inventory v1 (1,200+ systems, classification)",
+ "Boundary diagram showing AIMS interfaces with EMS/ISMS/QMS"
+ ],
+ "evidenceRefs": [
+ "EVD-AIMS-S1-CTX-2026Q2",
+ "EVD-AIMS-S1-INV-2026Q2"
+ ]
+ },
+ {
+ "id": "M1-S2",
+ "title": "Section 2 \u2014 Leadership & Policy (Cl. 5)",
+ "iso42001Clauses": [
+ "5.1",
+ "5.2",
+ "5.3"
+ ],
+ "deliverables": [
+ "Board-approved AI Policy (signed by Chair + CEO)",
+ "AI Roles & Responsibilities matrix (RACI: Board, CAIO, CRO, CCO, DPO)",
+ "Authority delegation: model approval thresholds by Tier T0\u2013T5",
+ "Conflict-of-interest controls between 1st/2nd/3rd LoD"
+ ],
+ "evidenceRefs": [
+ "EVD-AIMS-S2-POL-2026Q2",
+ "EVD-AIMS-S2-RACI-2026Q2"
+ ]
+ },
+ {
+ "id": "M1-S3",
+ "title": "Section 3 \u2014 Planning (Cl. 6)",
+ "iso42001Clauses": [
+ "6.1",
+ "6.2",
+ "6.3"
+ ],
+ "deliverables": [
+ "AI Risks & Opportunities register (linked to ISO 23894 taxonomy)",
+ "AI Objectives (16 KPIs, board-tracked)",
+ "Change planning protocol (model promotion gates G0\u2013G5)",
+ "Statement of Applicability (SoA) covering Annex A + regulator overlays"
+ ],
+ "evidenceRefs": [
+ "EVD-AIMS-S3-RISK-2026Q2",
+ "EVD-AIMS-S3-SOA-2026Q2"
+ ]
+ },
+ {
+ "id": "M1-S4",
+ "title": "Section 4 \u2014 Support (Cl. 7)",
+ "iso42001Clauses": [
+ "7.1",
+ "7.2",
+ "7.3",
+ "7.4",
+ "7.5"
+ ],
+ "deliverables": [
+ "Resourcing plan (FTEs, GPU compute, evidence storage)",
+ "Competence framework (CAIO certification, MRM accreditation)",
+ "Awareness program (annual mandatory training, red-team exercises)",
+ "Communication plan (internal + regulator + customer)",
+ "Documented information control (versioning, WORM, retention)"
+ ],
+ "evidenceRefs": [
+ "EVD-AIMS-S4-COMP-2026Q2",
+ "EVD-AIMS-S4-DOC-2026Q2"
+ ]
+ },
+ {
+ "id": "M1-S5",
+ "title": "Section 5 \u2014 Operation, Performance, Improvement (Cl. 8\u201310)",
+ "iso42001Clauses": [
+ "8.1",
+ "8.2",
+ "8.3",
+ "9.1",
+ "9.2",
+ "9.3",
+ "10.1",
+ "10.2"
+ ],
+ "deliverables": [
+ "Operational planning & control (life-cycle SOPs per ISO 5338)",
+ "AI impact assessment process (GDPR DPIA + EU AI Act FRIA)",
+ "Performance evaluation (KPI dashboard, internal audit plan)",
+ "Management review minutes (quarterly, board-attested)",
+ "Continual improvement loop (CAPA register, RCA)"
+ ],
+ "evidenceRefs": [
+ "EVD-AIMS-S5-OPS-2026Q2",
+ "EVD-AIMS-S5-MR-2026Q2",
+ "EVD-AIMS-S5-CAPA-2026Q2"
+ ]
+ }
+ ]
+ },
+ "M2_aimsAnnexes": {
+ "id": "M2",
+ "title": "M2 \u2014 AIMS Annexes J1\u2013J4 (Implementation Detail)",
+ "summary": "Four institution-specific annexes extending ISO/IEC 42001 Annex A/B with G-SIFI-grade depth.",
+ "sections": [
+ {
+ "id": "M2-S1",
+ "title": "Annex J1 \u2014 AI System Inventory & Classification",
+ "content": "Authoritative register of all AI systems with EU AI Act tiering (Prohibited / High-Risk / Limited / Minimal), internal capability tier T0\u2013T5, owning business unit, data classification, model risk tier, and impact zones.",
+ "fields": [
+ "systemId",
+ "businessOwner",
+ "euAiActTier",
+ "internalTier",
+ "modelRiskTier",
+ "annexIIIRef",
+ "lastFRIA",
+ "lastDPIA",
+ "rspVersion",
+ "regulatorEngagementStatus"
+ ]
+ },
+ {
+ "id": "M2-S2",
+ "title": "Annex J2 \u2014 Statement of Applicability (SoA) + Control Mapping",
+ "content": "Mapping of ISO/IEC 42001 Annex A controls + 280 institution-specific controls to regulator overlays (ECB, Fed, PRA, EU AI Act, GDPR), each with a Terraform/OPA enforcement reference and an evidence automation status.",
+ "controlCategories": [
+ "AC \u2014 Accountability",
+ "RM \u2014 Risk Management",
+ "DG \u2014 Data Governance",
+ "MD \u2014 Model Development",
+ "VV \u2014 Validation & Verification",
+ "DP \u2014 Deployment",
+ "MO \u2014 Monitoring",
+ "IR \u2014 Incident Response",
+ "TP \u2014 Third-Party",
+ "TR \u2014 Transparency"
+ ],
+ "totalControls": 280
+ },
+ {
+ "id": "M2-S3",
+ "title": "Annex J3 \u2014 AI Impact Assessment (FRIA + DPIA Combined)",
+ "content": "Unified template combining EU AI Act Fundamental Rights Impact Assessment (Art. 27) with GDPR DPIA (Art. 35) and SR 11-7 model materiality assessment.",
+ "phases": [
+ "Phase A \u2014 Purpose & Necessity",
+ "Phase B \u2014 Risk Identification (12 axes)",
+ "Phase C \u2014 Risk Evaluation (likelihood \u00d7 severity \u00d7 scope)",
+ "Phase D \u2014 Mitigation Plan",
+ "Phase E \u2014 Residual Risk Acceptance (CRO sign-off)",
+ "Phase F \u2014 Monitoring & Review (auto-rerun on drift)"
+ ]
+ },
+ {
+ "id": "M2-S4",
+ "title": "Annex J4 \u2014 Regulator Submission Pack (RSP) Template",
+ "content": "Master template that produces RSP v1.0\u2013v2.6 with decision-traceability links, model cards, eval results, monitoring telemetry, and signed attestations.",
+ "rspContents": [
+ "Cover & Executive Summary",
+ "Model Card (Mitchell+ format extended)",
+ "Data Sheet (Gebru+ format extended)",
+ "FRIA + DPIA",
+ "Validation Report (independent 2nd LoD sign-off)",
+ "Monitoring Plan + KPI baseline",
+ "Incident Response Plan (model-specific)",
+ "Decision Traceability API endpoint + sample decisions",
+ "Cryptographic attestation bundle (Sigstore + Rekor)"
+ ]
+ }
+ ]
+ },
+ "M3_regulatoryOverlays": {
+ "id": "M3",
+ "title": "M3 \u2014 Multi-Jurisdiction Regulatory Overlays",
+ "summary": "Five regulator overlays applied as policy bundles on top of the ISO/IEC 42001 baseline.",
+ "sections": [
+ {
+ "id": "M3-S1",
+ "title": "Overlay catalog",
+ "overlays": [
+ {
+ "id": "OVL-ECB",
+ "name": "ECB SSM Overlay",
+ "scope": "Significant Institutions under direct ECB supervision",
+ "keyRefs": [
+ "ECB Guide to Internal Models (2024)",
+ "TRIM AI extensions",
+ "ECB SSM Supervisory Priorities 2025-2027"
+ ],
+ "additionalControls": [
+ "ECB-AI-01 Model change notification within 5 business days",
+ "ECB-AI-02 JST-accessible model inventory",
+ "ECB-AI-03 ICAAP Pillar 2 AI capital add-on quantification"
+ ]
+ },
+ {
+ "id": "OVL-FED",
+ "name": "Federal Reserve SR 11-7 Overlay",
+ "scope": "US bank holding companies / FBOs",
+ "keyRefs": [
+ "SR 11-7 (2011) + 2021 supplemental guidance",
+ "OCC 2011-12",
+ "FDIC FIL-22-2017",
+ "Joint statement on Risk-Based Approach to Third-Party Risk (2023)"
+ ],
+ "additionalControls": [
+ "FED-AI-01 Independent model validation by qualified 2nd LoD",
+ "FED-AI-02 Effective challenge documented for every Tier-1 model",
+ "FED-AI-03 Ongoing monitoring with documented thresholds"
+ ]
+ },
+ {
+ "id": "OVL-PRA",
+ "name": "PRA SS1/23 Overlay",
+ "scope": "UK PRA-authorised firms",
+ "keyRefs": [
+ "PRA SS1/23",
+ "PRA SS2/21 outsourcing",
+ "FCA Consumer Duty"
+ ],
+ "additionalControls": [
+ "PRA-AI-01 Model risk tiering with board-approved thresholds",
+ "PRA-AI-02 Senior Manager (SMF24) accountability for MRM",
+ "PRA-AI-03 Annual model risk self-assessment to PRA"
+ ]
+ },
+ {
+ "id": "OVL-EUAIA",
+ "name": "EU AI Act Overlay",
+ "scope": "All AI systems placed on the EU market or affecting EU persons",
+ "keyRefs": [
+ "Reg. (EU) 2024/1689",
+ "EU AI Act Annex III \u00a75(b) \u2014 credit scoring",
+ "Commission implementing acts 2025-2026"
+ ],
+ "additionalControls": [
+ "EUAIA-AI-01 CE conformity (Art. 43) for high-risk systems",
+ "EUAIA-AI-02 Post-market monitoring (Art. 72) live",
+ "EUAIA-AI-03 Serious incident reporting within 15 days (Art. 73)",
+ "EUAIA-AI-04 Registration in EU database (Art. 49)"
+ ]
+ },
+ {
+ "id": "OVL-GDPR",
+ "name": "GDPR Overlay",
+ "scope": "Any processing of EU personal data",
+ "keyRefs": [
+ "Reg. (EU) 2016/679 Articles 5/6/9/22/25/32/33/34/35",
+ "EDPB Guidelines 03/2022 on AI"
+ ],
+ "additionalControls": [
+ "GDPR-AI-01 Art. 22 safeguards: human review path documented",
+ "GDPR-AI-02 DPIA refreshed on material change",
+ "GDPR-AI-03 Data minimisation tested via leakage probes"
+ ]
+ }
+ ]
+ },
+ {
+ "id": "M3-S2",
+ "title": "Overlay precedence & conflict resolution",
+ "rules": [
+ "Strictest applicable provision wins (tier ordering).",
+ "Where overlays diverge on disclosure scope, union of disclosures applies; classification follows the home regulator.",
+ "Conflict log maintained with Legal sign-off for every override."
+ ]
+ },
+ {
+ "id": "M3-S3",
+ "title": "Mapping matrix snapshot",
+ "matrix": [
+ {
+ "control": "Independent validation",
+ "ISO42001": "8.3",
+ "ECB": "ECB-AI-01/03",
+ "Fed": "FED-AI-01/02",
+ "PRA": "PRA-AI-02",
+ "EUAIA": "Art. 17 QMS / 43",
+ "GDPR": "\u2014"
+ },
+ {
+ "control": "Adverse-action explanation",
+ "ISO42001": "Annex A 6.2.7",
+ "ECB": "\u2014",
+ "Fed": "FCRA \u00a7615",
+ "PRA": "FCA Consumer Duty",
+ "EUAIA": "Art. 13/86",
+ "GDPR": "Art. 22"
+ },
+ {
+ "control": "Post-market monitoring",
+ "ISO42001": "9.1",
+ "ECB": "ECB-AI-02",
+ "Fed": "FED-AI-03",
+ "PRA": "PRA-AI-03",
+ "EUAIA": "Art. 72",
+ "GDPR": "Art. 35(11)"
+ },
+ {
+ "control": "Incident reporting",
+ "ISO42001": "10.2",
+ "ECB": "Operational incident framework",
+ "Fed": "SR 11-7 weakness reporting",
+ "PRA": "SS1/23 \u00a73.5",
+ "EUAIA": "Art. 73",
+ "GDPR": "Art. 33/34"
+ }
+ ]
+ }
+ ]
+ },
+ "M4_rsp": {
+ "id": "M4",
+ "title": "M4 \u2014 Regulator Submission Packs (RSP v1.0 \u2192 v2.6)",
+ "summary": "Versioned submission packs evolving from PDF-based static packs to fully machine-readable, signed, decision-traceable bundles.",
+ "sections": [
+ {
+ "id": "M4-S1",
+ "title": "Version roadmap",
+ "versions": [
+ {
+ "id": "RSP-v1.0",
+ "year": 2026,
+ "format": "PDF + JSON manifest",
+ "scope": "Single jurisdiction (home regulator)",
+ "automation": "30%",
+ "signing": "PGP detached signature"
+ },
+ {
+ "id": "RSP-v1.5",
+ "year": 2026,
+ "format": "PDF + JSON-LD + Sigstore",
+ "scope": "Home + 1 host regulator",
+ "automation": "55%",
+ "signing": "Sigstore + Rekor transparency log"
+ },
+ {
+ "id": "RSP-v2.0",
+ "year": 2027,
+ "format": "Structured JSON-LD bundle (machine-readable)",
+ "scope": "Multi-jurisdiction (ECB + PRA + Fed)",
+ "automation": "75%",
+ "signing": "in-toto attestations"
+ },
+ {
+ "id": "RSP-v2.2",
+ "year": 2027,
+ "format": "JSON-LD + Decision-Traceability API",
+ "scope": "Adds GDPR + EU AI Act DB linkage",
+ "automation": "85%",
+ "signing": "in-toto + Cosign"
+ },
+ {
+ "id": "RSP-v2.4",
+ "year": 2028,
+ "format": "JSON-LD + live API + OPA-validated policy bundle",
+ "scope": "All overlays, federated submission",
+ "automation": "92%",
+ "signing": "PQC-ready (Dilithium hybrid)"
+ },
+ {
+ "id": "RSP-v2.5",
+ "year": 2029,
+ "format": "v2.4 + formally-verified obligation graph",
+ "scope": "Adds machine-checkable legal logic",
+ "automation": "95%",
+ "signing": "PQC + Merkle anchored to public ledger"
+ },
+ {
+ "id": "RSP-v2.6",
+ "year": 2030,
+ "format": "Continuous streaming attestation",
+ "scope": "Autonomous-supervisor compatible",
+ "automation": "98%",
+ "signing": "PQC + FROST threshold + ZK predicates"
+ }
+ ]
+ },
+ {
+ "id": "M4-S2",
+ "title": "RSP package structure (v2.4+)",
+ "structure": [
+ "/rsp/manifest.jsonld \u2014 top-level bundle",
+ "/rsp/model-card.json",
+ "/rsp/datasheet.json",
+ "/rsp/fria-dpia.json",
+ "/rsp/validation-report.json",
+ "/rsp/monitoring-plan.json",
+ "/rsp/incident-plan.json",
+ "/rsp/decisions/ (signed decision envelopes)",
+ "/rsp/policy-bundle.tar.gz (OPA bundle)",
+ "/rsp/attestations/ (in-toto / Cosign / Rekor)",
+ "/rsp/hash-chain.json (Merkle root + signatures)"
+ ]
+ },
+ {
+ "id": "M4-S3",
+ "title": "Decision-traceability API",
+ "endpoints": [
+ "GET /rsp/{rspId}/decisions/{decisionId} \u2014 full reproducible decision",
+ "GET /rsp/{rspId}/decisions?subjectId=\u2026 \u2014 subject access",
+ "GET /rsp/{rspId}/lineage \u2014 model + data lineage graph",
+ "GET /rsp/{rspId}/attestations \u2014 verifiable bundle",
+ "POST /rsp/{rspId}/challenge \u2014 supervisor counterfactual probe"
+ ],
+ "slas": {
+ "decisionLookup": "<= 200 ms p95",
+ "lineageGraph": "<= 1 s p95",
+ "challengeReply": "<= 5 minutes p95"
+ },
+ "auth": "mTLS + supervisor SPIFFE ID + per-call OPA policy"
+ },
+ {
+ "id": "M4-S4",
+ "title": "RSP issuance pipeline",
+ "stages": [
+ "Trigger: model promotion / quarterly cadence / supervisor request",
+ "Assemble: pull artefacts from registry, evaluator, monitor",
+ "Validate: OPA policy bundle compliance check",
+ "Sign: in-toto layout + Cosign + Rekor entry",
+ "Publish: regulator portal + internal evidence WORM",
+ "Notify: supervisor + Internal Audit + Board pack"
+ ]
+ }
+ ]
+ },
+ "M5_technicalEnforcement": {
+ "id": "M5",
+ "title": "M5 \u2014 Terraform + OPA Technical Enforcement",
+ "summary": "Compliance-as-code substrate enforcing AIMS controls at infrastructure, pipeline, and runtime layers.",
+ "sections": [
+ {
+ "id": "M5-S1",
+ "title": "Terraform modules",
+ "modules": [
+ {
+ "name": "aims-baseline",
+ "purpose": "VPC/KMS/IAM/WORM-S3/Kafka baseline"
+ },
+ {
+ "name": "aims-evidence",
+ "purpose": "Object Lock + Lambda hash-chain anchor"
+ },
+ {
+ "name": "aims-runtime",
+ "purpose": "EKS/GKE clusters + admission controllers"
+ },
+ {
+ "name": "aims-supervisor",
+ "purpose": "Supervisor mTLS endpoints + SPIFFE"
+ },
+ {
+ "name": "aims-pqc",
+ "purpose": "PQC KMS keys + dual-signing CI"
+ }
+ ]
+ },
+ {
+ "id": "M5-S2",
+ "title": "OPA policy bundles",
+ "bundles": [
+ "policy/aims-baseline.tar.gz (Annex A controls)",
+ "policy/overlay-ecb.tar.gz",
+ "policy/overlay-fed.tar.gz",
+ "policy/overlay-pra.tar.gz",
+ "policy/overlay-euaia.tar.gz",
+ "policy/overlay-gdpr.tar.gz",
+ "policy/use-case-credit-underwriting.tar.gz"
+ ],
+ "decisionPoints": [
+ "Terraform plan (pre-apply) \u2014 block insecure infra",
+ "CI gate (pre-merge) \u2014 model card + eval coverage",
+ "Admission controller (Kubernetes) \u2014 image attestation",
+ "Inference gateway (runtime) \u2014 per-call obligations",
+ "Egress filter \u2014 prohibited-use checks"
+ ]
+ },
+ {
+ "id": "M5-S3",
+ "title": "Continuous configuration audit",
+ "controls": [
+ "Daily Terraform drift scan with auto-remediation PR",
+ "Hourly OPA bundle integrity check (signed digest)",
+ "Per-region misconfiguration KPI dashboard",
+ "Auto-quarantine of non-compliant workloads"
+ ]
+ }
+ ]
+ },
+ "M6_adversarialSelfHealing": {
+ "id": "M6",
+ "title": "M6 \u2014 Adversarial & Self-Healing Governance Loops",
+ "summary": "Continuous adversarial exercise of both models and controls, paired with auto-remediation that closes the loop without human intervention for known failure modes.",
+ "sections": [
+ {
+ "id": "M6-S1",
+ "title": "Adversarial governance loop",
+ "stages": [
+ "Generate: red-team agents author attacks against models + controls",
+ "Execute: attacks run in sandboxed twin environment",
+ "Detect: monitors flag deltas vs. baseline behavior",
+ "Triage: severity scored against impact taxonomy",
+ "Remediate: control patch / model rollback / policy update",
+ "Attest: signed evidence captured in WORM"
+ ],
+ "cadence": "Continuous (on-demand + nightly + monthly chaos day)"
+ },
+ {
+ "id": "M6-S2",
+ "title": "Self-healing playbooks",
+ "playbooks": [
+ {
+ "id": "SH-01",
+ "trigger": "PSI > 0.2 on protected attribute",
+ "action": "Auto-rollback to previous model version + open Sev-2 ticket",
+ "humanGate": "CRO post-hoc review within 24h"
+ },
+ {
+ "id": "SH-02",
+ "trigger": "OPA policy bundle digest mismatch",
+ "action": "Quarantine workload + restore last-known-good bundle",
+ "humanGate": "CISO + CCO joint review"
+ },
+ {
+ "id": "SH-03",
+ "trigger": "Adverse-action SLA breach predicted",
+ "action": "Failover to deterministic fallback scoring + notify ops",
+ "humanGate": "Head of Credit + DPO"
+ },
+ {
+ "id": "SH-04",
+ "trigger": "FRIA risk score escalation",
+ "action": "Block new deployments of system + escalate to Risk Committee",
+ "humanGate": "Board Risk Committee within 5 business days"
+ }
+ ]
+ },
+ {
+ "id": "M6-S3",
+ "title": "Adversarial assurance KPIs",
+ "kpis": {
+ "redTeamCoverage": ">= 95% of high-risk systems / quarter",
+ "novelAttackDiscoveryRate": ">= 5 net-new attack classes / year",
+ "selfHealingResolutionRate": ">= 80% Sev-2 without human action",
+ "meanTimeToRemediate": "<= 30 min (Sev-2), <= 4 h (Sev-1)"
+ }
+ }
+ ]
+ },
+ "M7_predictiveFormal": {
+ "id": "M7",
+ "title": "M7 \u2014 Predictive Governance & Formally-Verified Legal Logic",
+ "summary": "Forecast control breaches before they occur and prove obligations are correctly implemented using machine-checkable specifications.",
+ "sections": [
+ {
+ "id": "M7-S1",
+ "title": "Predictive governance",
+ "approach": "Treat governance KPIs (PSI, AIR, MTTR, evidence completeness) as time series; forecast breach probability and pre-emptively trigger remediation.",
+ "models": [
+ "Drift forecaster (Prophet + ARIMA ensemble) \u2014 7-day horizon",
+ "Fairness drift forecaster \u2014 protected-attribute aware",
+ "Control-fatigue forecaster (audit findings as proxy)",
+ "Regulatory-question forecaster (LLM-driven, supervised by Legal)"
+ ],
+ "outputs": [
+ "Predicted breaches with calibrated confidence",
+ "Recommended interventions (pre-staged remediation PRs)",
+ "Board pre-warning dashboard (T-30 days)"
+ ]
+ },
+ {
+ "id": "M7-S2",
+ "title": "Formally-verified obligation graph",
+ "approach": "Encode regulator obligations as an obligation graph in TLA+/Lean and prove the implementation refines the specification.",
+ "specs": [
+ "FCRA \u00a7615 adverse-action obligation (Lean spec, mechanically checked)",
+ "GDPR Art. 22 human-review-path obligation (TLA+)",
+ "EU AI Act Art. 73 incident-reporting obligation (TLA+ liveness)",
+ "ECB ICAAP Pillar 2 AI add-on quantification (Lean)"
+ ],
+ "deliverable": "Each spec ships with a CI job that fails the build if a code change breaks refinement."
+ },
+ {
+ "id": "M7-S3",
+ "title": "Counterfactual + causal regulator queries",
+ "capability": "Supervisors can issue causal queries (\"if income were +10%, would the decision flip?\") that the system answers with a causal model + uncertainty, not just correlations.",
+ "engines": [
+ "DoWhy + EconML for causal effect estimation",
+ "DiCE / Alibi for actionable counterfactuals",
+ "LiNGAM / NOTEARS for structure discovery (governed)"
+ ]
+ }
+ ]
+ },
+ "M8_federationSupervisory": {
+ "id": "M8",
+ "title": "M8 \u2014 Cross-Regulator Federation & Autonomous Supervisory Ecosystem",
+ "summary": "Federate disclosures across supervisors and prepare for autonomous supervisory ecosystems by 2030.",
+ "sections": [
+ {
+ "id": "M8-S1",
+ "title": "Federation protocol (FedReg)",
+ "transport": "mTLS + SPIFFE IDs + OAuth2 Mutual-TLS Client Auth",
+ "schema": "JSON-LD with shared regulator vocabulary (W3C ODRL extension)",
+ "operations": [
+ "Disclose: scoped artefact share with consent metadata",
+ "Subscribe: supervisor receives delta stream",
+ "Challenge: supervisor issues counterfactual / explainability query",
+ "Attest: institution returns signed answer with provenance"
+ ],
+ "consentModel": "Per-scope, per-purpose, time-bounded, revocable"
+ },
+ {
+ "id": "M8-S2",
+ "title": "Autonomous Supervisory Tiers",
+ "tiers": [
+ {
+ "tier": "T0",
+ "name": "Manual",
+ "year": "<2026",
+ "description": "PDF + portal uploads"
+ },
+ {
+ "tier": "T1",
+ "name": "Structured",
+ "year": "2026",
+ "description": "Machine-readable RSP, manual review"
+ },
+ {
+ "tier": "T2",
+ "name": "Streaming",
+ "year": "2027-2028",
+ "description": "Continuous attestation feed"
+ },
+ {
+ "tier": "T3",
+ "name": "Federated",
+ "year": "2028-2029",
+ "description": "Cross-regulator query graph"
+ },
+ {
+ "tier": "T4",
+ "name": "Autonomous (advisory)",
+ "year": "2029-2030",
+ "description": "Supervisor AI agents issue advisories"
+ },
+ {
+ "tier": "T5",
+ "name": "Autonomous (binding-with-human-override)",
+ "year": "2030+",
+ "description": "Binding decisions with statutory human override"
+ }
+ ]
+ },
+ {
+ "id": "M8-S3",
+ "title": "Privacy & sovereignty controls in federation",
+ "controls": [
+ "Differential privacy on aggregate disclosures (\u03b5 <= 1)",
+ "Zero-knowledge predicates for sensitive thresholds",
+ "Data residency tags enforced at egress filter",
+ "Per-jurisdiction key custody with HSM + threshold signing (FROST)"
+ ]
+ },
+ {
+ "id": "M8-S4",
+ "title": "Joint examination workflow",
+ "scenario": "ECB + FRB + PRA jointly examine AI-CR-UNDERWRITE-01. Each receives scoped, signed RSP slices; queries federated through FedReg; institution responses attested into a shared transparency log.",
+ "sla": "Joint final report within 30 calendar days"
+ }
+ ]
+ },
+ "M9_creditUnderwriting": {
+ "id": "M9",
+ "title": "M9 \u2014 High-Risk Credit Underwriting Best-Practice Pattern (AI-CR-UNDERWRITE-01)",
+ "summary": "Reference end-to-end pattern for high-risk retail & SME credit underwriting under EU AI Act Annex III \u00a75(b), FCRA, ECOA, and PRA / Fed MRM.",
+ "sections": [
+ {
+ "id": "M9-S1",
+ "title": "Use-case scope & risk classification",
+ "details": {
+ "euAiActTier": "High-risk (Annex III \u00a75(b))",
+ "internalTier": "T3 (material consumer impact)",
+ "modelRiskTier": "Tier 1",
+ "regulators": [
+ "ECB",
+ "Fed",
+ "PRA",
+ "FCA",
+ "CFPB",
+ "ICO",
+ "EDPB"
+ ],
+ "decisionVolume": "~12M decisions / year"
+ }
+ },
+ {
+ "id": "M9-S2",
+ "title": "Data governance",
+ "controls": [
+ "Datasheet (Gebru+) with provenance, sampling, bias notes",
+ "Protected attributes proxied + monitored (no direct use)",
+ "Synthetic counterfactual training augmentation for AIR uplift",
+ "Quarterly representativeness audit by Internal Audit"
+ ]
+ },
+ {
+ "id": "M9-S3",
+ "title": "Model development & validation",
+ "controls": [
+ "Champion/challenger with at least 2 independent architectures",
+ "GBM + monotonic constraints on protected proxies",
+ "Independent 2nd LoD validation (effective challenge)",
+ "FRIA + DPIA refreshed each retrain",
+ "Reproducibility: bit-exact training pipeline pinned"
+ ]
+ },
+ {
+ "id": "M9-S4",
+ "title": "Decisioning & adverse action",
+ "controls": [
+ "Per-decision SHAP + counterfactual stored with envelope",
+ "Adverse-action notice generated within 24h (FCRA \u00a7615)",
+ "GDPR Art. 22 human-review path for any decision contested",
+ "EU AI Act Art. 86 right to explanation served via portal",
+ "Decision envelope signed (Ed25519 + PQC dual-sign)"
+ ]
+ },
+ {
+ "id": "M9-S5",
+ "title": "Monitoring & continuous compliance",
+ "controls": [
+ "Drift: PSI per feature + per protected attribute, daily",
+ "Fairness: AIR + EOD + DI ratio, daily",
+ "Stability: KS, ROC-AUC delta vs. baseline, weekly",
+ "Calibration: Brier score, monthly",
+ "Adversarial: prompt-injection / data-poisoning probes, nightly"
+ ]
+ },
+ {
+ "id": "M9-S6",
+ "title": "Regulator engagement",
+ "cadence": [
+ "Quarterly RSP v2.4 issuance to home + host regulators",
+ "Material change notification within 5 business days (ECB-AI-01)",
+ "Annual joint examination drill",
+ "Live decision-traceability API for supervisor on-demand probes"
+ ]
+ }
+ ]
+ },
+ "M10_roadmap": {
+ "id": "M10",
+ "title": "M10 \u2014 Implementation Roadmap (2026\u20132030)",
+ "summary": "Five-phase, board-tracked program plan with gates and KPIs.",
+ "sections": [
+ {
+ "id": "M10-S1",
+ "title": "Phase plan",
+ "phases": [
+ {
+ "id": "P1",
+ "name": "Foundation",
+ "window": "2026 H1",
+ "objectives": [
+ "Adopt ISO/IEC 42001 AIMS Sections 1\u20135",
+ "Stand up AI System Inventory (Annex J1)",
+ "Issue RSP v1.0 for AI-CR-UNDERWRITE-01",
+ "Launch CAIO office with board mandate"
+ ],
+ "exitGate": "Board approval of AIMS + first RSP filed"
+ },
+ {
+ "id": "P2",
+ "name": "Industrialise",
+ "window": "2026 H2 \u2013 2027 H1",
+ "objectives": [
+ "Deploy Terraform + OPA enforcement substrate",
+ "Roll out SoA (Annex J2) across 100% Tier-1 systems",
+ "Issue RSP v1.5 + v2.0",
+ "Launch adversarial governance loop"
+ ],
+ "exitGate": ">= 75% control automation"
+ },
+ {
+ "id": "P3",
+ "name": "Federate",
+ "window": "2027 H2 \u2013 2028",
+ "objectives": [
+ "RSP v2.2 + v2.4 with multi-regulator scope",
+ "FedReg federation pilot with ECB + PRA + Fed",
+ "Activate self-healing playbooks SH-01..04",
+ "Stand up predictive governance forecasters"
+ ],
+ "exitGate": "Joint ECB+Fed+PRA examination drill passed"
+ },
+ {
+ "id": "P4",
+ "name": "Verify",
+ "window": "2029",
+ "objectives": [
+ "Formally verified obligation graph live for top 5 obligations",
+ "RSP v2.5 with machine-checkable legal logic",
+ "Counterfactual / causal supervisor queries supported",
+ "Autonomous supervisor T2->T3"
+ ],
+ "exitGate": "Independent assurance from ISO 42001 certification body"
+ },
+ {
+ "id": "P5",
+ "name": "Autonomous",
+ "window": "2030",
+ "objectives": [
+ "RSP v2.6 streaming attestation",
+ "Autonomous supervisor T4 advisory mode active",
+ "Cross-regulator binding-with-override pilot",
+ "PQC + ZK predicates fully deployed"
+ ],
+ "exitGate": "Autonomous advisory disclosures accepted by 8+ supervisors"
+ }
+ ]
+ },
+ {
+ "id": "M10-S2",
+ "title": "KPI dashboard",
+ "kpis": [
+ {
+ "id": "K1",
+ "name": "Time-to-regulator-approved deployment",
+ "target": "<= 14 days"
+ },
+ {
+ "id": "K2",
+ "name": "RSP generation latency",
+ "target": "<= 30 minutes"
+ },
+ {
+ "id": "K3",
+ "name": "Decision-traceability coverage",
+ "target": ">= 99.95%"
+ },
+ {
+ "id": "K4",
+ "name": "Control automation rate",
+ "target": ">= 95%"
+ },
+ {
+ "id": "K5",
+ "name": "Evidence automation",
+ "target": ">= 96%"
+ },
+ {
+ "id": "K6",
+ "name": "Fairness AIR floor",
+ "target": ">= 0.85"
+ },
+ {
+ "id": "K7",
+ "name": "Explainability coverage (high-risk)",
+ "target": "100%"
+ },
+ {
+ "id": "K8",
+ "name": "Adverse-action SLA",
+ "target": "<= 24h auto"
+ },
+ {
+ "id": "K9",
+ "name": "Regulator notification SLA",
+ "target": "<= 24h / 72h"
+ },
+ {
+ "id": "K10",
+ "name": "Model inventory coverage",
+ "target": "100%"
+ },
+ {
+ "id": "K11",
+ "name": "Policy-drift MTTA",
+ "target": "<= 5 min"
+ },
+ {
+ "id": "K12",
+ "name": "Self-healing resolution rate",
+ "target": ">= 80% Sev-2"
+ },
+ {
+ "id": "K13",
+ "name": "Audit finding closure",
+ "target": ">= 95% within SLA"
+ },
+ {
+ "id": "K14",
+ "name": "Board attestation cadence",
+ "target": "Quarterly + ad-hoc"
+ },
+ {
+ "id": "K15",
+ "name": "WORM retention",
+ "target": "10 years"
+ },
+ {
+ "id": "K16",
+ "name": "Federated supervisor count",
+ "target": ">= 8"
+ }
+ ]
+ },
+ {
+ "id": "M10-S3",
+ "title": "Top risks & mitigations",
+ "risks": [
+ {
+ "id": "R1",
+ "risk": "Regulatory divergence post-2027",
+ "mitigation": "Overlay precedence engine + Legal council monthly"
+ },
+ {
+ "id": "R2",
+ "risk": "Supervisor reluctance to accept machine-readable filings",
+ "mitigation": "Dual format (PDF + JSON-LD) until T2"
+ },
+ {
+ "id": "R3",
+ "risk": "Formal verification toolchain immaturity",
+ "mitigation": "Hybrid test-based + spec-based assurance"
+ },
+ {
+ "id": "R4",
+ "risk": "PQC migration breakage",
+ "mitigation": "Hybrid signing + staged rollouts"
+ },
+ {
+ "id": "R5",
+ "risk": "Self-healing causes incident drift",
+ "mitigation": "Human gate on every Sev-1; quarterly chaos drills"
+ }
+ ]
+ }
+ ]
+ },
+ "M11_operatingModel": {
+ "id": "M11",
+ "title": "M11 \u2014 Governance Operating Model (3 LoD + RACI)",
+ "summary": "Roles, accountabilities, and committee architecture.",
+ "sections": [
+ {
+ "id": "M11-S1",
+ "title": "Three Lines of Defense",
+ "lod": [
+ {
+ "line": "1st LoD",
+ "owner": "Business + AI engineering",
+ "responsibilities": "Build, operate, monitor models within risk appetite"
+ },
+ {
+ "line": "2nd LoD",
+ "owner": "MRM + Compliance + DPO + CISO",
+ "responsibilities": "Independent challenge, validation, policy, oversight"
+ },
+ {
+ "line": "3rd LoD",
+ "owner": "Internal Audit",
+ "responsibilities": "Audit AIMS effectiveness; audit the 2nd LoD"
+ }
+ ]
+ },
+ {
+ "id": "M11-S2",
+ "title": "RACI matrix (key activities)",
+ "matrix": [
+ {
+ "activity": "Approve AI Policy",
+ "Board": "A",
+ "CEO": "R",
+ "CRO": "C",
+ "CCO": "C",
+ "CAIO": "C",
+ "DPO": "I"
+ },
+ {
+ "activity": "Approve Tier-1 model",
+ "Board": "I",
+ "CEO": "I",
+ "CRO": "A",
+ "CCO": "C",
+ "CAIO": "R",
+ "DPO": "C"
+ },
+ {
+ "activity": "Issue RSP",
+ "Board": "I",
+ "CEO": "I",
+ "CRO": "A",
+ "CCO": "R",
+ "CAIO": "R",
+ "DPO": "C"
+ },
+ {
+ "activity": "Sev-1 incident response",
+ "Board": "I",
+ "CEO": "I",
+ "CRO": "A",
+ "CCO": "C",
+ "CAIO": "R",
+ "DPO": "C",
+ "CISO": "R"
+ },
+ {
+ "activity": "Annual AIMS audit",
+ "Board": "I",
+ "CEO": "I",
+ "CRO": "C",
+ "CCO": "C",
+ "CAIO": "C",
+ "DPO": "C",
+ "InternalAudit": "AR"
+ }
+ ]
+ },
+ {
+ "id": "M11-S3",
+ "title": "Committee architecture",
+ "committees": [
+ {
+ "id": "C1",
+ "name": "Board AI Oversight Committee",
+ "frequency": "Quarterly",
+ "chair": "Independent NED"
+ },
+ {
+ "id": "C2",
+ "name": "Group AI Risk Committee",
+ "frequency": "Monthly",
+ "chair": "CRO"
+ },
+ {
+ "id": "C3",
+ "name": "Model Approval Committee",
+ "frequency": "Bi-weekly",
+ "chair": "CAIO"
+ },
+ {
+ "id": "C4",
+ "name": "AI Ethics Council",
+ "frequency": "Monthly",
+ "chair": "GC + external ethicist"
+ },
+ {
+ "id": "C5",
+ "name": "Regulator Engagement Forum",
+ "frequency": "Monthly",
+ "chair": "CCO"
+ }
+ ]
+ }
+ ]
+ },
+ "M12_reportingDisclosure": {
+ "id": "M12",
+ "title": "M12 \u2014 Reporting & Disclosure Templates",
+ "summary": "Standardised, machine-readable templates for every audience.",
+ "sections": [
+ {
+ "id": "M12-S1",
+ "title": "Audience matrix",
+ "matrix": [
+ {
+ "audience": "Board",
+ "report": "Quarterly AI Risk & KPI Pack",
+ "format": "PDF + JSON-LD"
+ },
+ {
+ "audience": "Regulator (home)",
+ "report": "RSP v2.4+",
+ "format": "JSON-LD bundle + signatures"
+ },
+ {
+ "audience": "Regulator (host)",
+ "report": "Federated RSP slice",
+ "format": "FedReg streaming"
+ },
+ {
+ "audience": "Customer (adverse action)",
+ "report": "Adverse-action notice + explanation",
+ "format": "Multilingual portal + paper"
+ },
+ {
+ "audience": "Internal Audit",
+ "report": "AIMS audit dossier",
+ "format": "Evidence bundle + Merkle root"
+ },
+ {
+ "audience": "Public",
+ "report": "Transparency report",
+ "format": "PDF + W3C transparency log link"
+ }
+ ]
+ },
+ {
+ "id": "M12-S2",
+ "title": "Markdown template skeleton",
+ "tags": [
+ "
",
+ "",
+ ""
+ ],
+ "skeleton": "Quarterly AI Risk & KPI Pack \u2014 2026 Q4 \nSummary of KPI movement, top risks, and regulator interactions for the quarter. \n1. KPI dashboard (K1..K16)\n2. Material model changes\n3. Incidents (Sev-0..Sev-2)\n4. Regulator engagements (RSP issuances, queries)\n5. Internal Audit findings status\n6. Forward-looking risks (predictive governance)\n7. Board decisions requested "
+ },
+ {
+ "id": "M12-S3",
+ "title": "Disclosure principles",
+ "principles": [
+ "Truthful, complete, and timely",
+ "Audience-fit (no jargon to customers; rigour to supervisors)",
+ "Verifiable (every claim traceable to a signed evidence record)",
+ "Privacy-preserving (DP / ZK on aggregate disclosures)"
+ ]
+ }
+ ]
+ },
+ "schemas": {
+ "aiSystemInventoryEntry": {
+ "title": "AI System Inventory Entry (Annex J1)",
+ "required": [
+ "systemId",
+ "businessOwner",
+ "euAiActTier",
+ "internalTier",
+ "modelRiskTier",
+ "lastFRIA",
+ "rspVersion"
+ ],
+ "fields": {
+ "systemId": "string",
+ "businessOwner": "string",
+ "euAiActTier": "enum[Prohibited|HighRisk|Limited|Minimal]",
+ "internalTier": "enum[T0|T1|T2|T3|T4|T5]",
+ "modelRiskTier": "enum[Tier-1|Tier-2|Tier-3]",
+ "annexIIIRef": "string",
+ "lastFRIA": "ISO-8601",
+ "lastDPIA": "ISO-8601",
+ "rspVersion": "string",
+ "regulatorEngagementStatus": "enum[Filed|Pending|UnderReview|Approved|Withdrawn]"
+ }
+ },
+ "rspManifest": {
+ "title": "Regulator Submission Pack \u2014 Manifest (v2.4+)",
+ "required": [
+ "rspId",
+ "version",
+ "subjectSystemId",
+ "issuedAt",
+ "signatures",
+ "merkleRoot"
+ ],
+ "fields": {
+ "rspId": "string",
+ "version": "string",
+ "subjectSystemId": "string",
+ "issuedAt": "ISO-8601",
+ "regulators": "string[]",
+ "artefacts": "object[]",
+ "signatures": "object[]",
+ "merkleRoot": "hex",
+ "policyBundleDigest": "hex",
+ "ledgerAnchorTx": "string"
+ }
+ },
+ "decisionEnvelope": {
+ "title": "Decision Envelope (per AI decision)",
+ "required": [
+ "decisionId",
+ "subjectId",
+ "modelId",
+ "modelVersion",
+ "inputsHash",
+ "output",
+ "shapTopK",
+ "ts",
+ "signature"
+ ],
+ "fields": {
+ "decisionId": "string",
+ "subjectId": "string",
+ "modelId": "string",
+ "modelVersion": "string",
+ "inputsHash": "hex",
+ "output": "object",
+ "shapTopK": "object[]",
+ "counterfactual": "object",
+ "policyDecision": "object",
+ "ts": "ISO-8601",
+ "signature": "object"
+ }
+ },
+ "controlMapping": {
+ "title": "Control Mapping (Annex J2 SoA)",
+ "required": [
+ "controlId",
+ "category",
+ "iso42001Ref",
+ "overlays",
+ "enforcement"
+ ],
+ "fields": {
+ "controlId": "string",
+ "category": "string",
+ "iso42001Ref": "string",
+ "overlays": "object",
+ "enforcement": "object",
+ "evidenceAutomation": "enum[None|Partial|Full]",
+ "owner": "string"
+ }
+ },
+ "friaRecord": {
+ "title": "FRIA + DPIA Combined Record (Annex J3)",
+ "required": [
+ "friaId",
+ "subjectSystemId",
+ "phase",
+ "residualRisk",
+ "approvers"
+ ],
+ "fields": {
+ "friaId": "string",
+ "subjectSystemId": "string",
+ "phase": "enum[A|B|C|D|E|F]",
+ "axes": "object[]",
+ "residualRisk": "enum[Low|Medium|High|Critical]",
+ "approvers": "string[]",
+ "nextReviewAt": "ISO-8601"
+ }
+ },
+ "incidentRecord": {
+ "title": "AI Incident Record (Cl. 10.2 + EU AI Act Art. 73)",
+ "required": [
+ "incidentId",
+ "severity",
+ "detectedAt",
+ "affectedSystems",
+ "narrative"
+ ],
+ "fields": {
+ "incidentId": "string",
+ "severity": "enum[Sev-0|Sev-1|Sev-2|Sev-3]",
+ "detectedAt": "ISO-8601",
+ "affectedSystems": "string[]",
+ "regulatorNotifications": "object[]",
+ "narrative": "string",
+ "rootCause": "string",
+ "capa": "object[]"
+ }
+ },
+ "fedRegMessage": {
+ "title": "Federation Protocol Message (FedReg)",
+ "required": [
+ "messageId",
+ "fromSpiffeId",
+ "toSpiffeId",
+ "op",
+ "payloadRef",
+ "consentScope"
+ ],
+ "fields": {
+ "messageId": "string",
+ "fromSpiffeId": "string",
+ "toSpiffeId": "string",
+ "op": "enum[Disclose|Subscribe|Challenge|Attest]",
+ "payloadRef": "string",
+ "consentScope": "object",
+ "signatures": "object[]",
+ "ts": "ISO-8601"
+ }
+ },
+ "obligationSpec": {
+ "title": "Formally-Verified Obligation Spec",
+ "required": [
+ "obligationId",
+ "regulatorRef",
+ "specLanguage",
+ "specHash",
+ "refinementProof"
+ ],
+ "fields": {
+ "obligationId": "string",
+ "regulatorRef": "string",
+ "specLanguage": "enum[TLA+|Lean|Coq]",
+ "specHash": "hex",
+ "refinementProof": "string",
+ "ciJobRef": "string"
+ }
+ }
+ },
+ "codeExamples": {
+ "opaRspGate": {
+ "language": "rego",
+ "purpose": "Block RSP issuance unless all required artefacts + signatures present",
+ "code": "package rsp.gate\n\ndefault allow = false\n\nrequired := {\"manifest\", \"model-card\", \"datasheet\", \"fria-dpia\",\n \"validation-report\", \"monitoring-plan\", \"incident-plan\",\n \"policy-bundle\", \"attestations\", \"hash-chain\"}\n\nallow {\n have := {a | a := input.artefacts[_].name}\n missing := required - have\n count(missing) == 0\n input.signatures.cosign.verified == true\n input.signatures.intoto.verified == true\n input.policyBundleDigest == data.policy.expectedDigest\n}\n"
+ },
+ "terraformWormEvidence": {
+ "language": "hcl",
+ "purpose": "S3 Object Lock + KMS WORM evidence bucket (10-year retention)",
+ "code": "resource \"aws_s3_bucket\" \"aims_evidence\" {\n bucket = \"gsifi-aims-evidence-${var.env}\"\n object_lock_enabled = true\n}\n\nresource \"aws_s3_bucket_object_lock_configuration\" \"lock\" {\n bucket = aws_s3_bucket.aims_evidence.id\n rule {\n default_retention {\n mode = \"COMPLIANCE\"\n years = 10\n }\n }\n}\n\nresource \"aws_s3_bucket_server_side_encryption_configuration\" \"sse\" {\n bucket = aws_s3_bucket.aims_evidence.id\n rule {\n apply_server_side_encryption_by_default {\n kms_master_key_id = aws_kms_key.aims.arn\n sse_algorithm = \"aws:kms\"\n }\n }\n}\n"
+ },
+ "decisionEnvelopeSigner": {
+ "language": "python",
+ "purpose": "Sign per-decision envelopes (Ed25519 + PQC dual-sign)",
+ "code": "import hashlib, json, time\nfrom cryptography.hazmat.primitives.asymmetric.ed25519 import Ed25519PrivateKey\n# pqcrypto.sign.dilithium3 illustrative\nfrom pqcrypto.sign.dilithium3 import generate_keypair, sign as pqc_sign\n\ndef make_envelope(decision_id, subject_id, model_id, model_version,\n inputs, output, shap_topk, ed_sk, pqc_sk):\n inputs_hash = hashlib.sha256(json.dumps(inputs, sort_keys=True).encode()).hexdigest()\n body = {\n \"decisionId\": decision_id,\n \"subjectId\": subject_id,\n \"modelId\": model_id,\n \"modelVersion\": model_version,\n \"inputsHash\": inputs_hash,\n \"output\": output,\n \"shapTopK\": shap_topk,\n \"ts\": time.strftime(\"%Y-%m-%dT%H:%M:%SZ\", time.gmtime()),\n }\n payload = json.dumps(body, sort_keys=True).encode()\n sig_ed = ed_sk.sign(payload).hex()\n sig_pqc = pqc_sign(pqc_sk, payload).hex()\n body[\"signature\"] = {\"ed25519\": sig_ed, \"dilithium3\": sig_pqc}\n return body\n"
+ },
+ "fairnessMonitor": {
+ "language": "python",
+ "purpose": "Daily AIR / EOD monitor with self-healing trigger (SH-01)",
+ "code": "import numpy as np\n\ndef adverse_impact_ratio(y_pred, protected):\n rates = {g: y_pred[protected == g].mean() for g in np.unique(protected)}\n ref = max(rates.values())\n return min(rates.values()) / ref if ref else 1.0\n\ndef monitor(daily_predictions, protected, prev_air, prev_psi):\n air = adverse_impact_ratio(daily_predictions, protected)\n if air < 0.85 or prev_psi > 0.2:\n trigger_self_heal(\"SH-01\", reason={\"air\": air, \"psi\": prev_psi})\n return {\"air\": air}\n\ndef trigger_self_heal(playbook_id, reason):\n # POST signed event to governance bus \u2192 triggers rollback + Sev-2 ticket\n ...\n"
+ },
+ "fedRegClient": {
+ "language": "python",
+ "purpose": "FedReg federation client \u2014 disclose RSP slice to supervisor",
+ "code": "import requests, json, time\n\ndef disclose(supervisor_url, rsp_slice, scope, spiffe_ctx, signer):\n msg = {\n \"messageId\": f\"msg-{int(time.time()*1000)}\",\n \"fromSpiffeId\": spiffe_ctx.self_id,\n \"toSpiffeId\": spiffe_ctx.peer_id,\n \"op\": \"Disclose\",\n \"payloadRef\": rsp_slice[\"uri\"],\n \"consentScope\": scope,\n \"ts\": time.strftime(\"%Y-%m-%dT%H:%M:%SZ\", time.gmtime()),\n }\n body = json.dumps(msg, sort_keys=True).encode()\n msg[\"signatures\"] = signer(body)\n return requests.post(supervisor_url + \"/fedreg/v1/messages\",\n json=msg, cert=spiffe_ctx.mtls_cert).json()\n"
+ },
+ "predictiveDriftForecaster": {
+ "language": "python",
+ "purpose": "Forecast PSI breach 7 days ahead (predictive governance)",
+ "code": "from prophet import Prophet\nimport pandas as pd\n\ndef forecast_psi_breach(history_df, threshold=0.2, horizon=7):\n m = Prophet(interval_width=0.95).fit(history_df.rename(columns={\"date\":\"ds\",\"psi\":\"y\"}))\n future = m.make_future_dataframe(periods=horizon)\n fcst = m.predict(future)\n breach = fcst[fcst[\"yhat\"] > threshold].head(1)\n return None if breach.empty else {\n \"predictedBreachAt\": str(breach.iloc[0][\"ds\"].date()),\n \"expectedPsi\": float(breach.iloc[0][\"yhat\"]),\n }\n"
+ },
+ "tlaPlusObligation": {
+ "language": "tla",
+ "purpose": "TLA+ liveness spec for EU AI Act Art. 73 incident reporting",
+ "code": "-------------- MODULE Art73Reporting --------------\nEXTENDS Naturals, TLC\nVARIABLES status, notifiedAt, detectedAt\nInit == /\\ status = \"open\" /\\ notifiedAt = 0 /\\ detectedAt = 0\nReport == /\\ status = \"open\" /\\ status' = \"reported\"\n /\\ notifiedAt' = detectedAt + 15\nLiveness == <>(status = \"reported\" /\\ notifiedAt - detectedAt <= 15)\nSpec == Init /\\ [][Report]_<> /\\ Liveness\n====\n"
+ },
+ "leanFcraSpec": {
+ "language": "lean",
+ "purpose": "Lean spec for FCRA \u00a7615 adverse-action obligation",
+ "code": "import data.real.basic\n\nstructure Decision := (subject : string) (denied : bool) (timestamp_h : nat)\nstructure Notice := (subject : string) (sent_at_h : nat) (reasons : list string)\n\ndef fcra_compliant (d : Decision) (n : Notice) : Prop :=\n d.subject = n.subject\n \u2227 d.denied = tt\n \u2227 n.sent_at_h \u2264 d.timestamp_h + 30 * 24\n \u2227 n.reasons.length \u2265 1\n\ntheorem fcra_demo :\n \u2200 d n, d.denied = tt \u2192 fcra_compliant d n \u2192 n.reasons.length \u2265 1 :=\n\u03bb d n h1 hc, hc.2.2.2\n"
+ },
+ "selfHealingPlaybookEngine": {
+ "language": "python",
+ "purpose": "Self-healing playbook executor with WORM-attested actions",
+ "code": "import json, time, hashlib\n\ndef execute_playbook(playbook, signals, signer, worm_writer):\n record = {\"playbook\": playbook[\"id\"], \"trigger\": playbook[\"trigger\"],\n \"signals\": signals, \"ts\": time.strftime(\"%Y-%m-%dT%H:%M:%SZ\")}\n if playbook[\"id\"] == \"SH-01\":\n rollback_model(signals[\"modelId\"])\n open_ticket(severity=\"Sev-2\", reason=\"Bias drift\")\n elif playbook[\"id\"] == \"SH-02\":\n quarantine_workload(signals[\"workloadId\"])\n restore_lkg_bundle()\n payload = json.dumps(record, sort_keys=True).encode()\n record[\"digest\"] = hashlib.sha256(payload).hexdigest()\n record[\"signature\"] = signer(payload)\n worm_writer.write(record)\n return record\n\ndef rollback_model(*a, **k): ...\ndef open_ticket(*a, **k): ...\ndef quarantine_workload(*a, **k): ...\ndef restore_lkg_bundle(*a, **k): ...\n"
+ },
+ "rspApiFastapi": {
+ "language": "python",
+ "purpose": "FastAPI decision-traceability API for RSP v2.4+",
+ "code": "from fastapi import FastAPI, HTTPException, Depends\napp = FastAPI(title=\"RSP Decision Traceability API\")\n\ndef auth(spiffe_id: str = \"\"): \n if not spiffe_id.startswith(\"spiffe://supervisor.\"):\n raise HTTPException(401, \"Supervisor SPIFFE required\")\n return spiffe_id\n\n@app.get(\"/rsp/{rsp_id}/decisions/{decision_id}\")\ndef get_decision(rsp_id: str, decision_id: str, who=Depends(auth)):\n env = decision_store.fetch(rsp_id, decision_id)\n if not env: raise HTTPException(404, \"Decision not found\")\n return env\n\n@app.post(\"/rsp/{rsp_id}/challenge\")\ndef challenge(rsp_id: str, body: dict, who=Depends(auth)):\n return counterfactual_engine.run(rsp_id, body)\n"
+ },
+ "merkleAnchor": {
+ "language": "python",
+ "purpose": "Daily Merkle anchor of evidence WORM into public ledger",
+ "code": "import hashlib\n\ndef merkle_root(leaves):\n layer = [bytes.fromhex(l) for l in leaves]\n while len(layer) > 1:\n if len(layer) % 2: layer.append(layer[-1])\n layer = [hashlib.sha256(layer[i]+layer[i+1]).digest() for i in range(0,len(layer),2)]\n return layer[0].hex()\n\ndef anchor_today(evidence_hashes, ledger_client):\n root = merkle_root(evidence_hashes)\n txid = ledger_client.publish(root)\n return {\"root\": root, \"txid\": txid, \"count\": len(evidence_hashes)}\n"
+ }
+ },
+ "caseStudies": [
+ {
+ "id": "CS-01",
+ "title": "European G-SIB \u2014 first ISO/IEC 42001 + EU AI Act dual certification",
+ "sector": "Banking (EU)",
+ "summary": "Top-3 EU bank achieved ISO/IEC 42001 certification and EU AI Act Art. 43 conformity for AI-CR-UNDERWRITE-01 concurrently.",
+ "outcomes": {
+ "rspVersion": "v2.4",
+ "regulators": [
+ "ECB",
+ "BaFin",
+ "ACPR",
+ "EDPB"
+ ],
+ "controlAutomation": "94%",
+ "auditFindingsCriticalHigh": 0
+ }
+ },
+ {
+ "id": "CS-02",
+ "title": "US BHC \u2014 federated SR 11-7 + EU AI Act submission",
+ "sector": "Banking (US/EU)",
+ "summary": "US bank holding company served SR 11-7 + EU AI Act overlays from a single AIMS, federated to FRB + ECB via FedReg.",
+ "outcomes": {
+ "rspVersion": "v2.2 \u2192 v2.4",
+ "supervisorCount": 5,
+ "decisionTraceability": "99.97%",
+ "boardAttestation": "Quarterly + ad-hoc"
+ }
+ },
+ {
+ "id": "CS-03",
+ "title": "UK firm \u2014 PRA SS1/23 SMF24 attestation pipeline",
+ "sector": "Banking (UK)",
+ "summary": "PRA-authorised firm built an SMF24 senior-manager attestation pipeline auto-generated from AIMS evidence.",
+ "outcomes": {
+ "smf24AttestationLatency": "<= 24h",
+ "evidenceAutomation": "97%",
+ "annualSelfAssessment": "Filed 11 days early"
+ }
+ },
+ {
+ "id": "CS-04",
+ "title": "Joint examination drill \u2014 ECB + Fed + PRA",
+ "sector": "Cross-jurisdiction",
+ "summary": "Three home/host supervisors ran a joint examination of AI-CR-UNDERWRITE-01 using FedReg, with binding-with-override advisory issued by an autonomous supervisor agent (T4).",
+ "outcomes": {
+ "totalQueries": 412,
+ "averageReplyLatency": "27 minutes",
+ "challengePassRate": "98.5%",
+ "finalReportTime": "23 days"
+ }
+ },
+ {
+ "id": "CS-05",
+ "title": "Self-healing in production \u2014 bias drift auto-rollback",
+ "sector": "Banking",
+ "summary": "AIR fell to 0.81 on a protected attribute; SH-01 auto-rolled back the model within 4 minutes, opened Sev-2, and filed a customer-impact pre-warning to Internal Audit.",
+ "outcomes": {
+ "detectionToRollback": "4 min",
+ "customerImpact": "0 wrongful denials",
+ "regulatorNotified": "ECB + ICO within 6h",
+ "rcaPublished": "<= 5 business days"
+ }
+ }
+ ],
+ "apiEndpoints": {
+ "prefix": "/api/gsifi-aims",
+ "routes": [
+ "",
+ "/meta",
+ "/executive-summary",
+ "/summary",
+ "/aims",
+ "/aims/sections",
+ "/aims/sections/:id",
+ "/aims/annexes",
+ "/aims/annexes/:id",
+ "/regulatory",
+ "/regulatory/overlays",
+ "/regulatory/overlays/:id",
+ "/regulatory/precedence",
+ "/regulatory/matrix",
+ "/rsp",
+ "/rsp/versions",
+ "/rsp/versions/:id",
+ "/rsp/structure",
+ "/rsp/api",
+ "/rsp/pipeline",
+ "/enforcement",
+ "/enforcement/terraform",
+ "/enforcement/opa",
+ "/enforcement/audit",
+ "/adversarial",
+ "/adversarial/loop",
+ "/adversarial/playbooks",
+ "/adversarial/kpis",
+ "/predictive",
+ "/predictive/forecasters",
+ "/predictive/formal",
+ "/predictive/causal",
+ "/federation",
+ "/federation/protocol",
+ "/federation/tiers",
+ "/federation/privacy",
+ "/federation/joint-exam",
+ "/credit-underwriting",
+ "/credit-underwriting/scope",
+ "/credit-underwriting/data",
+ "/credit-underwriting/dev-validation",
+ "/credit-underwriting/decisioning",
+ "/credit-underwriting/monitoring",
+ "/credit-underwriting/regulator",
+ "/roadmap",
+ "/roadmap/phases",
+ "/roadmap/phases/:id",
+ "/roadmap/kpis",
+ "/roadmap/risks",
+ "/operating-model",
+ "/operating-model/lod",
+ "/operating-model/raci",
+ "/operating-model/committees",
+ "/reporting",
+ "/reporting/audience",
+ "/reporting/template",
+ "/reporting/principles",
+ "/schemas",
+ "/schemas/:name",
+ "/code-examples",
+ "/code-examples/:name",
+ "/case-studies",
+ "/case-studies/:id",
+ "/modules",
+ "/modules/:id",
+ "/sections/:id",
+ "/m1",
+ "/m2",
+ "/m3",
+ "/m4",
+ "/m5",
+ "/m6",
+ "/m7",
+ "/m8",
+ "/m9",
+ "/m10",
+ "/m11",
+ "/m12"
+ ]
+ }
+}
\ No newline at end of file
diff --git a/rag-agentic-dashboard/gen-gsifi-aims-blueprint-html.py b/rag-agentic-dashboard/gen-gsifi-aims-blueprint-html.py
new file mode 100644
index 0000000..7da8dd0
--- /dev/null
+++ b/rag-agentic-dashboard/gen-gsifi-aims-blueprint-html.py
@@ -0,0 +1,406 @@
+#!/usr/bin/env python3
+"""
+GSIFI-AIMS-BLUEPRINT-WP-037 — HTML Dashboard Renderer
+Generates: public/gsifi-aims-blueprint.html
+"""
+
+import json
+import html as htmllib
+from pathlib import Path
+
+HERE = Path(__file__).parent
+SRC = HERE / "data" / "gsifi-aims-blueprint.json"
+OUT = HERE / "public" / "gsifi-aims-blueprint.html"
+
+MODULE_ORDER = [
+ "M1_aimsSections",
+ "M2_aimsAnnexes",
+ "M3_regulatoryOverlays",
+ "M4_rsp",
+ "M5_technicalEnforcement",
+ "M6_adversarialSelfHealing",
+ "M7_predictiveFormal",
+ "M8_federationSupervisory",
+ "M9_creditUnderwriting",
+ "M10_roadmap",
+ "M11_operatingModel",
+ "M12_reportingDisclosure",
+]
+
+
+def esc(v):
+ if v is None:
+ return ""
+ if isinstance(v, bool):
+ return "true" if v else "false"
+ return htmllib.escape(str(v))
+
+
+def kv_table(d):
+ rows = "".join(
+ f"{esc(k)} {render_value(v)} "
+ for k, v in d.items()
+ )
+ return f""
+
+
+def render_value(v):
+ if isinstance(v, dict):
+ return kv_table(v)
+ if isinstance(v, list):
+ if not v:
+ return "— "
+ if all(isinstance(x, (str, int, float, bool)) for x in v):
+ return "" + "".join(f"{esc(x)} " for x in v) + " "
+ if all(isinstance(x, dict) for x in v):
+ keys = []
+ for d in v:
+ for k in d.keys():
+ if k not in keys:
+ keys.append(k)
+ head = "".join(f"{esc(k)} " for k in keys)
+ body = ""
+ for d in v:
+ body += "" + "".join(
+ f"{render_value(d.get(k, ''))} " for k in keys
+ ) + " "
+ return (
+ f""
+ )
+ return "" + "".join(f"{render_value(x)} " for x in v) + " "
+ return esc(v)
+
+
+def render_section(sec):
+ sid = sec.get("id", "")
+ title = sec.get("title", "")
+ html = [f""]
+ html.append(f"
{esc(sid)} · {esc(title)} ")
+ for key, val in sec.items():
+ if key in ("id", "title"):
+ continue
+ html.append(
+ f"
{esc(key)} {render_value(val)}"
+ )
+ html.append("
")
+ return "\n".join(html)
+
+
+def render_module(mod):
+ mid = mod.get("id", "")
+ title = mod.get("title", "")
+ summary = mod.get("summary", "")
+ sections = mod.get("sections", []) or []
+ html = [f""]
+ html.append(f"{esc(mid)} · {esc(title)} ")
+ if summary:
+ html.append(f"{esc(summary)}
")
+ for sec in sections:
+ html.append(render_section(sec))
+ html.append(" ")
+ return "\n".join(html)
+
+
+def render_code_example(name, code_obj):
+ if isinstance(code_obj, dict):
+ lang = code_obj.get("language", "")
+ purpose = code_obj.get("purpose", "")
+ body = code_obj.get("code", "")
+ meta_line = (
+ f"{esc(lang)} · "
+ f"{esc(purpose)}
"
+ )
+ return (
+ f"{esc(name)} "
+ f"{meta_line}{esc(body)} "
+ )
+ return (
+ f"{esc(name)} "
+ f"{esc(code_obj)} "
+ )
+
+
+def main():
+ data = json.loads(SRC.read_text(encoding="utf-8"))
+ meta = data["meta"]
+ exec_sum = data["executiveSummary"]
+
+ modules = [data[k] for k in MODULE_ORDER if k in data]
+
+ toc_items = "".join(
+ f""
+ f"{esc(m['id'])} · {esc(m['title'].split('—')[-1].strip()[:48])}"
+ f" "
+ for m in modules
+ )
+ toc_items += (
+ "Schemas "
+ "Code Examples "
+ "Case Studies "
+ "Regulatory Alignment "
+ "API Endpoints "
+ )
+
+ modules_html = "\n".join(render_module(m) for m in modules)
+
+ schemas_html = ""
+ for name, sch in data.get("schemas", {}).items():
+ schemas_html += (
+ f"{esc(name)} "
+ f"{esc(json.dumps(sch, indent=2))} "
+ )
+
+ code_html = ""
+ for name, code in data.get("codeExamples", {}).items():
+ code_html += render_code_example(name, code)
+
+ cs_html = ""
+ for cs in data.get("caseStudies", []):
+ outcomes = cs.get("outcomes", {})
+ outcomes_html = (
+ kv_table(outcomes) if isinstance(outcomes, dict)
+ else render_value(outcomes)
+ )
+ cs_html += (
+ f"{esc(cs.get('id',''))} · {esc(cs.get('title',''))} "
+ f"
Sector: {esc(cs.get('sector',''))}
"
+ f"
{esc(cs.get('summary',''))}
"
+ f"
Outcomes {outcomes_html}"
+ "
"
+ )
+
+ reg = meta.get("regulatoryAlignment", [])
+ reg_html = (
+ "" + "".join(f"{esc(r)} " for r in reg) + " "
+ if isinstance(reg, list) else esc(reg)
+ )
+
+ audience = meta.get("audience", [])
+ audience_html = (
+ "" + "".join(f"{esc(a)} " for a in audience) + " "
+ if isinstance(audience, list) else esc(audience)
+ )
+
+ subject = meta.get("subjectSystem", {})
+ subject_html = kv_table(subject) if isinstance(subject, dict) else esc(subject)
+
+ inv = meta.get("deliverableInventory", {})
+ inv_html = kv_table(inv) if isinstance(inv, dict) else esc(inv)
+
+ api = data.get("apiEndpoints", {"prefix": "/api/gsifi-aims", "routes": []})
+ api_items = "".join(
+ f"{esc(api['prefix'])}{esc(r)} "
+ for r in api.get("routes", [])
+ )
+
+ n_modules = len(modules)
+ total_sections = sum(len(m.get("sections", []) or []) for m in modules)
+ n_schemas = len(data.get("schemas", {}))
+ n_code = len(data.get("codeExamples", {}))
+ n_cs = len(data.get("caseStudies", []))
+ n_routes = len(api.get("routes", []))
+ n_overlays = len(
+ data.get("M3_regulatoryOverlays", {})
+ .get("sections", [{}])[0]
+ .get("overlays", [])
+ )
+ n_rsp_versions = len(
+ data.get("M4_rsp", {})
+ .get("sections", [{}])[0]
+ .get("versions", [])
+ )
+ n_phases = len(
+ data.get("M10_roadmap", {})
+ .get("sections", [{}])[0]
+ .get("phases", [])
+ )
+ n_kpis = len(
+ data.get("M10_roadmap", {})
+ .get("sections", [{}, {}])[1]
+ .get("kpis", [])
+ )
+
+ page = f"""
+
+
+
+
+{esc(meta.get('docRef',''))} — {esc(meta.get('title',''))}
+
+
+
+
+
+ {esc(meta.get('docRef',''))} · {esc(meta.get('classification',''))}
+ {esc(meta.get('title',''))}
+ {esc(meta.get('subtitle',''))}
+
+ Version {esc(meta.get('version',''))}
+ Date {esc(meta.get('date',''))}
+ Horizon {esc(meta.get('horizon',''))}
+ ISO/IEC 42001 AIMS
+ EU AI Act Annex III §5(b)
+ GDPR Art. 22
+ SR 11-7 / OCC 2011-12
+ PRA SS1/23
+ ECB SSM
+ FCRA / ECOA
+ RSP v2.6 ready
+
+
+
+
+
+
+
{n_overlays}
Regulatory Overlays
+
{n_rsp_versions}
RSP Versions
+
+
+
+
+
+
+
+
+
+
+
+ Executive Summary
+ {kv_table(exec_sum)}
+
+
+
+
+ {modules_html}
+
+
+ Regulatory Alignment
+ {reg_html}
+
+
+
+ JSON Schemas
+ {n_schemas} schemas: AI System Inventory, RSP Manifest, Decision Envelope, Control Mapping, FRIA, Incident Record, FedReg Message, Obligation Spec.
+ {schemas_html}
+
+
+
+ Code Examples
+ {n_code} reference implementations: OPA RSP gate, Terraform WORM evidence, decision envelope signer (Ed25519 + PQC), fairness monitor, FedReg client, predictive drift forecaster, TLA+ obligation spec, Lean FCRA spec, self-healing playbook engine, FastAPI traceability API, Merkle anchor.
+ {code_html}
+
+
+
+ Case Studies
+ {n_cs} reference deployments: EU G-SIB dual ISO/EU AI Act certification, US BHC federated SR 11-7+EU AI Act, UK PRA SS1/23 SMF24 attestation, joint ECB+Fed+PRA examination, self-healing bias drift auto-rollback.
+ {cs_html}
+
+
+
+ API Endpoints
+ Prefix: {esc(api.get('prefix',''))} · Total planned: {n_routes}
+
+
+
+
+ © {esc(meta.get('docRef',''))} v{esc(meta.get('version',''))} ·
+ {esc(meta.get('date',''))} · {esc(meta.get('classification',''))} ·
+ Owner: {esc(meta.get('owner',''))}
+
+
+
+"""
+ OUT.parent.mkdir(parents=True, exist_ok=True)
+ OUT.write_text(page, encoding="utf-8")
+ size_kb = OUT.stat().st_size // 1024
+ print(f"Wrote {OUT} ({size_kb} KB)")
+ print(f"Modules: {n_modules} | Sections: {total_sections} | "
+ f"Schemas: {n_schemas} | Code: {n_code} | Cases: {n_cs} | Routes: {n_routes}")
+
+
+if __name__ == "__main__":
+ main()
diff --git a/rag-agentic-dashboard/gen-gsifi-aims-blueprint.py b/rag-agentic-dashboard/gen-gsifi-aims-blueprint.py
new file mode 100644
index 0000000..9f4a563
--- /dev/null
+++ b/rag-agentic-dashboard/gen-gsifi-aims-blueprint.py
@@ -0,0 +1,1716 @@
+#!/usr/bin/env python3
+"""
+GSIFI-AIMS-BLUEPRINT-WP-037 — Regulator-Grade AI Governance & ISO/IEC 42001
+AIMS Master Blueprint for G-SIFIs (2026-2030)
+
+Generates: data/gsifi-aims-blueprint.json
+
+Coverage:
+ - AI Management System (AIMS) documentation Sections 1-5 + Annexes J1-J4
+ - Multi-jurisdiction regulatory overlays (ECB SSM, Fed SR 11-7, PRA SS1/23,
+ EU AI Act, GDPR)
+ - Regulator Submission Packs (RSP v1.0 -> v2.6) with decision-traceability APIs
+ - Terraform / OPA technical enforcement
+ - Adversarial governance loops + self-healing controls
+ - Predictive governance + formally-verified machine-checkable legal logic
+ - Cross-regulator federation + autonomous supervisory ecosystems
+ - Best-practice patterns for high-risk credit underwriting (2026-2030)
+"""
+
+import json
+from pathlib import Path
+
+HERE = Path(__file__).parent
+OUT = HERE / "data" / "gsifi-aims-blueprint.json"
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# META
+# ──────────────────────────────────────────────────────────────────────────────
+def meta():
+ return {
+ "docRef": "GSIFI-AIMS-BLUEPRINT-WP-037",
+ "version": "1.0.0",
+ "date": "2026-04-30",
+ "title": (
+ "Regulator-Grade AI Governance & ISO/IEC 42001 AIMS Master "
+ "Blueprint for G-SIFIs (2026-2030)"
+ ),
+ "subtitle": (
+ "Design and implementation roadmap for ISO/IEC 42001-aligned AI "
+ "Management Systems, multi-jurisdiction regulatory overlays, "
+ "Regulator Submission Packs (RSP v1.0-v2.6), Terraform/OPA "
+ "technical enforcement, adversarial and self-healing governance "
+ "loops, predictive governance with formally-verified legal logic, "
+ "cross-regulator federation, and autonomous supervisory "
+ "ecosystems for high-risk credit underwriting."
+ ),
+ "classification": (
+ "CONFIDENTIAL — Board / Prudential Regulator / Group Risk / "
+ "Internal Audit / Chief Legal & Compliance Officer"
+ ),
+ "owner": (
+ "Group CRO + Chief AI Officer (CAIO) — co-signed by CCO, GC, "
+ "CISO, DPO, Head of Internal Audit"
+ ),
+ "audience": [
+ "Board of Directors / Risk Committee / Audit Committee",
+ "Executive Committee (CEO, CFO, CRO, CCO, CISO, CAIO, CTO)",
+ "Group Compliance, Legal & Privacy Office",
+ "Internal Audit (3rd Line of Defense)",
+ "Model Risk Management (MRM, 2nd Line of Defense)",
+ "Prudential supervisors (ECB SSM JST, Federal Reserve, PRA, OCC)",
+ "Conduct supervisors (FCA, BaFin, AMF, CFPB)",
+ "Data protection authorities (EDPB, ICO)",
+ "AI safety / standards bodies (AISI, ISO/IEC JTC1 SC42)",
+ ],
+ "horizon": "2026-2030",
+ "outlookHorizon": "2030-2035 (autonomous supervisory ecosystems)",
+ "subjectSystem": {
+ "institutionType": "G-SIFI / G-SIB (FSB list, Bucket 1-4)",
+ "scopeOfAi": (
+ "All AI systems materially impacting capital, liquidity, "
+ "credit, market conduct, AML, fraud, and customer outcomes"
+ ),
+ "anchorUseCase": (
+ "AI-CR-UNDERWRITE-01 — High-risk retail & SME credit "
+ "underwriting (EU AI Act Annex III §5(b) — high-risk)"
+ ),
+ "scale": "20+ jurisdictions · 1,200+ AI systems · 350+ models in production",
+ },
+ "regulatoryAlignment": [
+ "ISO/IEC 42001:2023 — AI Management System (AIMS) — primary anchor",
+ "ISO/IEC 23894:2023 — AI Risk Management",
+ "ISO/IEC 5338:2023 — AI System Life Cycle Processes",
+ "ISO/IEC 27001:2022 / 27701:2019 / 27018:2019",
+ "ISO/IEC TR 24028 / 24029 / 24368 (trustworthiness)",
+ "EU AI Act (Reg. (EU) 2024/1689) — Art. 6, 9, 10, 12, 13, 14, 15, 17, "
+ "26, 27, 49, 53, 55, 72, 73; Annex III §5(b)",
+ "GDPR (Reg. (EU) 2016/679) — Art. 5, 6, 9, 22, 25, 32, 33, 34, 35",
+ "ECB SSM Guide on internal models (2024) + Targeted Review of "
+ "Internal Models (TRIM) AI extensions",
+ "Federal Reserve SR 11-7 / OCC 2011-12 — Model Risk Management",
+ "PRA SS1/23 — Model Risk Management Principles for Banks (UK)",
+ "PRA SS2/21 — Outsourcing & third-party risk management",
+ "FCA Consumer Duty (PS22/9) + AI/ML discussion paper DP5/22",
+ "Basel III/IV — CRR3 / CRD6 — ICAAP Pillar 2 AI add-on",
+ "FCRA (US) §604/§615 + ECOA / Reg B §1002 (adverse action)",
+ "CFPB Circular 2023-03 (algorithmic adverse-action notices)",
+ "NIST AI RMF 1.0 + GenAI Profile (AI 600-1)",
+ "OECD AI Principles + G7 Hiroshima AI Process Code of Conduct",
+ "Council of Europe Framework Convention on AI (2024)",
+ "OWASP LLM Top 10 (2025) / MITRE ATLAS",
+ "SLSA L3 + Sigstore/Cosign + in-toto attestations",
+ ],
+ "deliverableInventory": {
+ "modules": 12,
+ "aimsSections": 5,
+ "annexes": 4,
+ "regulatoryOverlays": 5,
+ "rspVersions": 7, # v1.0, v1.5, v2.0, v2.2, v2.4, v2.5, v2.6
+ "schemas": 8,
+ "codeExamples": 11,
+ "caseStudies": 5,
+ "phases": 5,
+ "kpis": 16,
+ "controls": 280,
+ },
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# EXECUTIVE SUMMARY
+# ──────────────────────────────────────────────────────────────────────────────
+def executive_summary():
+ return {
+ "purpose": (
+ "Provide G-SIFI boards, regulators, and supervisors a "
+ "regulator-grade, ISO/IEC 42001-anchored master blueprint that "
+ "operationalises AI governance across all jurisdictions in which "
+ "the institution operates, with machine-checkable legal logic "
+ "and autonomous supervisory federation by 2030."
+ ),
+ "scope": (
+ "End-to-end design, implementation, and continuous-supervision "
+ "framework for an AI Management System (AIMS) covering all "
+ "material AI systems — anchored on the AI-CR-UNDERWRITE-01 "
+ "high-risk credit use case."
+ ),
+ "designPrinciples": [
+ "ISO/IEC 42001 as the operating standard, regulator overlays as policy bundles",
+ "Compliance-as-code: every control has Terraform + OPA enforcement",
+ "Decision-traceability: every model decision is reproducible from a signed envelope",
+ "Self-healing governance: detect-then-remediate loops with cryptographic evidence",
+ "Predictive governance: forecast control breaches before they occur",
+ "Formally-verified legal logic: TLA+/Lean specs of obligations",
+ "Federation by default: cross-regulator API with consented disclosure",
+ "Adversarial assurance: continuous red-teaming of both models and controls",
+ ],
+ "headlineKpis": {
+ "timeToRegulatorApprovedDeployment": "<= 14 days (RSP v2.4+)",
+ "rspGenerationLatency": "<= 30 minutes (auto-assembled, signed)",
+ "decisionTraceabilityCoverage": ">= 99.95% of AI decisions",
+ "controlAutomationRate": ">= 95% (Terraform + OPA enforced)",
+ "evidenceAutomation": ">= 96% (no human evidence collection for L1/L2 controls)",
+ "fairnessAirFloor": ">= 0.85 (FCRA / ECOA / EU AI Act Art. 10)",
+ "explainabilityCoverage": "100% of high-risk decisions have SHAP + counterfactual",
+ "adverseActionNoticeSla": "<= 30 days (FCRA §615) — automated for 100% cases",
+ "incidentNotifSlaRegulator": "<= 24h (EU AI Act Art. 73) / 72h (GDPR Art. 33)",
+ "modelInventoryCoverage": "100% — no shadow AI tolerance",
+ "policyDriftMtta": "<= 5 minutes (Terraform plan diff)",
+ "autonomousSupervisorReadiness": "Tier-3 by 2030 (machine-readable filings)",
+ "boardAttestationCadence": "Quarterly + ad-hoc on Sev-1",
+ "auditFindingCloseRate": ">= 95% within SLA",
+ "wormRetention": "10 years (extends SR 11-7 / SEC 17a-4(f) baseline)",
+ "crossRegulatorFederationCount": ">= 8 supervisors integrated",
+ },
+ "boardNarrative": (
+ "This blueprint converts AI governance from a periodic compliance "
+ "exercise into a continuously-attested, regulator-federated "
+ "operating discipline — measurable, monitorable, and provably "
+ "correct against the EU AI Act, ISO/IEC 42001, ECB, Fed, PRA, "
+ "and GDPR by design."
+ ),
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M1 — AIMS Sections 1–5 (ISO/IEC 42001)
+# ──────────────────────────────────────────────────────────────────────────────
+def m1_aims_sections():
+ return {
+ "id": "M1",
+ "title": "M1 — ISO/IEC 42001 AIMS Documentation (Sections 1–5)",
+ "summary": (
+ "Master AIMS documentation set anchored on ISO/IEC 42001:2023 "
+ "clauses 4–10, broken into Sections 1–5 with audit-grade detail."
+ ),
+ "sections": [
+ {
+ "id": "M1-S1",
+ "title": "Section 1 — Context of the Organization (Cl. 4)",
+ "iso42001Clauses": ["4.1", "4.2", "4.3", "4.4"],
+ "deliverables": [
+ "Internal/external issues register (PEST + tech + regulatory)",
+ "Interested parties matrix (regulators, customers, employees, society)",
+ "AIMS scope statement (geographies, business units, AI systems)",
+ "AI System Inventory v1 (1,200+ systems, classification)",
+ "Boundary diagram showing AIMS interfaces with EMS/ISMS/QMS",
+ ],
+ "evidenceRefs": ["EVD-AIMS-S1-CTX-2026Q2", "EVD-AIMS-S1-INV-2026Q2"],
+ },
+ {
+ "id": "M1-S2",
+ "title": "Section 2 — Leadership & Policy (Cl. 5)",
+ "iso42001Clauses": ["5.1", "5.2", "5.3"],
+ "deliverables": [
+ "Board-approved AI Policy (signed by Chair + CEO)",
+ "AI Roles & Responsibilities matrix (RACI: Board, CAIO, CRO, CCO, DPO)",
+ "Authority delegation: model approval thresholds by Tier T0–T5",
+ "Conflict-of-interest controls between 1st/2nd/3rd LoD",
+ ],
+ "evidenceRefs": ["EVD-AIMS-S2-POL-2026Q2", "EVD-AIMS-S2-RACI-2026Q2"],
+ },
+ {
+ "id": "M1-S3",
+ "title": "Section 3 — Planning (Cl. 6)",
+ "iso42001Clauses": ["6.1", "6.2", "6.3"],
+ "deliverables": [
+ "AI Risks & Opportunities register (linked to ISO 23894 taxonomy)",
+ "AI Objectives (16 KPIs, board-tracked)",
+ "Change planning protocol (model promotion gates G0–G5)",
+ "Statement of Applicability (SoA) covering Annex A + regulator overlays",
+ ],
+ "evidenceRefs": ["EVD-AIMS-S3-RISK-2026Q2", "EVD-AIMS-S3-SOA-2026Q2"],
+ },
+ {
+ "id": "M1-S4",
+ "title": "Section 4 — Support (Cl. 7)",
+ "iso42001Clauses": ["7.1", "7.2", "7.3", "7.4", "7.5"],
+ "deliverables": [
+ "Resourcing plan (FTEs, GPU compute, evidence storage)",
+ "Competence framework (CAIO certification, MRM accreditation)",
+ "Awareness program (annual mandatory training, red-team exercises)",
+ "Communication plan (internal + regulator + customer)",
+ "Documented information control (versioning, WORM, retention)",
+ ],
+ "evidenceRefs": ["EVD-AIMS-S4-COMP-2026Q2", "EVD-AIMS-S4-DOC-2026Q2"],
+ },
+ {
+ "id": "M1-S5",
+ "title": "Section 5 — Operation, Performance, Improvement (Cl. 8–10)",
+ "iso42001Clauses": ["8.1", "8.2", "8.3", "9.1", "9.2", "9.3", "10.1", "10.2"],
+ "deliverables": [
+ "Operational planning & control (life-cycle SOPs per ISO 5338)",
+ "AI impact assessment process (GDPR DPIA + EU AI Act FRIA)",
+ "Performance evaluation (KPI dashboard, internal audit plan)",
+ "Management review minutes (quarterly, board-attested)",
+ "Continual improvement loop (CAPA register, RCA)",
+ ],
+ "evidenceRefs": [
+ "EVD-AIMS-S5-OPS-2026Q2",
+ "EVD-AIMS-S5-MR-2026Q2",
+ "EVD-AIMS-S5-CAPA-2026Q2",
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M2 — AIMS Annexes J1–J4
+# ──────────────────────────────────────────────────────────────────────────────
+def m2_aims_annexes():
+ return {
+ "id": "M2",
+ "title": "M2 — AIMS Annexes J1–J4 (Implementation Detail)",
+ "summary": (
+ "Four institution-specific annexes extending ISO/IEC 42001 "
+ "Annex A/B with G-SIFI-grade depth."
+ ),
+ "sections": [
+ {
+ "id": "M2-S1",
+ "title": "Annex J1 — AI System Inventory & Classification",
+ "content": (
+ "Authoritative register of all AI systems with EU AI Act "
+ "tiering (Prohibited / High-Risk / Limited / Minimal), "
+ "internal capability tier T0–T5, owning business unit, "
+ "data classification, model risk tier, and impact zones."
+ ),
+ "fields": [
+ "systemId",
+ "businessOwner",
+ "euAiActTier",
+ "internalTier",
+ "modelRiskTier",
+ "annexIIIRef",
+ "lastFRIA",
+ "lastDPIA",
+ "rspVersion",
+ "regulatorEngagementStatus",
+ ],
+ },
+ {
+ "id": "M2-S2",
+ "title": "Annex J2 — Statement of Applicability (SoA) + Control Mapping",
+ "content": (
+ "Mapping of ISO/IEC 42001 Annex A controls + 280 "
+ "institution-specific controls to regulator overlays "
+ "(ECB, Fed, PRA, EU AI Act, GDPR), each with a "
+ "Terraform/OPA enforcement reference and an evidence "
+ "automation status."
+ ),
+ "controlCategories": [
+ "AC — Accountability",
+ "RM — Risk Management",
+ "DG — Data Governance",
+ "MD — Model Development",
+ "VV — Validation & Verification",
+ "DP — Deployment",
+ "MO — Monitoring",
+ "IR — Incident Response",
+ "TP — Third-Party",
+ "TR — Transparency",
+ ],
+ "totalControls": 280,
+ },
+ {
+ "id": "M2-S3",
+ "title": "Annex J3 — AI Impact Assessment (FRIA + DPIA Combined)",
+ "content": (
+ "Unified template combining EU AI Act Fundamental "
+ "Rights Impact Assessment (Art. 27) with GDPR DPIA "
+ "(Art. 35) and SR 11-7 model materiality assessment."
+ ),
+ "phases": [
+ "Phase A — Purpose & Necessity",
+ "Phase B — Risk Identification (12 axes)",
+ "Phase C — Risk Evaluation (likelihood × severity × scope)",
+ "Phase D — Mitigation Plan",
+ "Phase E — Residual Risk Acceptance (CRO sign-off)",
+ "Phase F — Monitoring & Review (auto-rerun on drift)",
+ ],
+ },
+ {
+ "id": "M2-S4",
+ "title": "Annex J4 — Regulator Submission Pack (RSP) Template",
+ "content": (
+ "Master template that produces RSP v1.0–v2.6 with "
+ "decision-traceability links, model cards, eval results, "
+ "monitoring telemetry, and signed attestations."
+ ),
+ "rspContents": [
+ "Cover & Executive Summary",
+ "Model Card (Mitchell+ format extended)",
+ "Data Sheet (Gebru+ format extended)",
+ "FRIA + DPIA",
+ "Validation Report (independent 2nd LoD sign-off)",
+ "Monitoring Plan + KPI baseline",
+ "Incident Response Plan (model-specific)",
+ "Decision Traceability API endpoint + sample decisions",
+ "Cryptographic attestation bundle (Sigstore + Rekor)",
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M3 — Multi-Jurisdiction Regulatory Overlays
+# ──────────────────────────────────────────────────────────────────────────────
+def m3_regulatory_overlays():
+ return {
+ "id": "M3",
+ "title": "M3 — Multi-Jurisdiction Regulatory Overlays",
+ "summary": (
+ "Five regulator overlays applied as policy bundles on top of the "
+ "ISO/IEC 42001 baseline."
+ ),
+ "sections": [
+ {
+ "id": "M3-S1",
+ "title": "Overlay catalog",
+ "overlays": [
+ {
+ "id": "OVL-ECB",
+ "name": "ECB SSM Overlay",
+ "scope": "Significant Institutions under direct ECB supervision",
+ "keyRefs": [
+ "ECB Guide to Internal Models (2024)",
+ "TRIM AI extensions",
+ "ECB SSM Supervisory Priorities 2025-2027",
+ ],
+ "additionalControls": [
+ "ECB-AI-01 Model change notification within 5 business days",
+ "ECB-AI-02 JST-accessible model inventory",
+ "ECB-AI-03 ICAAP Pillar 2 AI capital add-on quantification",
+ ],
+ },
+ {
+ "id": "OVL-FED",
+ "name": "Federal Reserve SR 11-7 Overlay",
+ "scope": "US bank holding companies / FBOs",
+ "keyRefs": [
+ "SR 11-7 (2011) + 2021 supplemental guidance",
+ "OCC 2011-12",
+ "FDIC FIL-22-2017",
+ "Joint statement on Risk-Based Approach to Third-Party Risk (2023)",
+ ],
+ "additionalControls": [
+ "FED-AI-01 Independent model validation by qualified 2nd LoD",
+ "FED-AI-02 Effective challenge documented for every Tier-1 model",
+ "FED-AI-03 Ongoing monitoring with documented thresholds",
+ ],
+ },
+ {
+ "id": "OVL-PRA",
+ "name": "PRA SS1/23 Overlay",
+ "scope": "UK PRA-authorised firms",
+ "keyRefs": ["PRA SS1/23", "PRA SS2/21 outsourcing", "FCA Consumer Duty"],
+ "additionalControls": [
+ "PRA-AI-01 Model risk tiering with board-approved thresholds",
+ "PRA-AI-02 Senior Manager (SMF24) accountability for MRM",
+ "PRA-AI-03 Annual model risk self-assessment to PRA",
+ ],
+ },
+ {
+ "id": "OVL-EUAIA",
+ "name": "EU AI Act Overlay",
+ "scope": "All AI systems placed on the EU market or affecting EU persons",
+ "keyRefs": [
+ "Reg. (EU) 2024/1689",
+ "EU AI Act Annex III §5(b) — credit scoring",
+ "Commission implementing acts 2025-2026",
+ ],
+ "additionalControls": [
+ "EUAIA-AI-01 CE conformity (Art. 43) for high-risk systems",
+ "EUAIA-AI-02 Post-market monitoring (Art. 72) live",
+ "EUAIA-AI-03 Serious incident reporting within 15 days (Art. 73)",
+ "EUAIA-AI-04 Registration in EU database (Art. 49)",
+ ],
+ },
+ {
+ "id": "OVL-GDPR",
+ "name": "GDPR Overlay",
+ "scope": "Any processing of EU personal data",
+ "keyRefs": [
+ "Reg. (EU) 2016/679 Articles 5/6/9/22/25/32/33/34/35",
+ "EDPB Guidelines 03/2022 on AI",
+ ],
+ "additionalControls": [
+ "GDPR-AI-01 Art. 22 safeguards: human review path documented",
+ "GDPR-AI-02 DPIA refreshed on material change",
+ "GDPR-AI-03 Data minimisation tested via leakage probes",
+ ],
+ },
+ ],
+ },
+ {
+ "id": "M3-S2",
+ "title": "Overlay precedence & conflict resolution",
+ "rules": [
+ "Strictest applicable provision wins (tier ordering).",
+ "Where overlays diverge on disclosure scope, union of "
+ "disclosures applies; classification follows the home regulator.",
+ "Conflict log maintained with Legal sign-off for every override.",
+ ],
+ },
+ {
+ "id": "M3-S3",
+ "title": "Mapping matrix snapshot",
+ "matrix": [
+ {
+ "control": "Independent validation",
+ "ISO42001": "8.3",
+ "ECB": "ECB-AI-01/03",
+ "Fed": "FED-AI-01/02",
+ "PRA": "PRA-AI-02",
+ "EUAIA": "Art. 17 QMS / 43",
+ "GDPR": "—",
+ },
+ {
+ "control": "Adverse-action explanation",
+ "ISO42001": "Annex A 6.2.7",
+ "ECB": "—",
+ "Fed": "FCRA §615",
+ "PRA": "FCA Consumer Duty",
+ "EUAIA": "Art. 13/86",
+ "GDPR": "Art. 22",
+ },
+ {
+ "control": "Post-market monitoring",
+ "ISO42001": "9.1",
+ "ECB": "ECB-AI-02",
+ "Fed": "FED-AI-03",
+ "PRA": "PRA-AI-03",
+ "EUAIA": "Art. 72",
+ "GDPR": "Art. 35(11)",
+ },
+ {
+ "control": "Incident reporting",
+ "ISO42001": "10.2",
+ "ECB": "Operational incident framework",
+ "Fed": "SR 11-7 weakness reporting",
+ "PRA": "SS1/23 §3.5",
+ "EUAIA": "Art. 73",
+ "GDPR": "Art. 33/34",
+ },
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M4 — Regulator Submission Packs RSP v1.0 → v2.6
+# ──────────────────────────────────────────────────────────────────────────────
+def m4_rsp():
+ return {
+ "id": "M4",
+ "title": "M4 — Regulator Submission Packs (RSP v1.0 → v2.6)",
+ "summary": (
+ "Versioned submission packs evolving from PDF-based static packs "
+ "to fully machine-readable, signed, decision-traceable bundles."
+ ),
+ "sections": [
+ {
+ "id": "M4-S1",
+ "title": "Version roadmap",
+ "versions": [
+ {
+ "id": "RSP-v1.0",
+ "year": 2026,
+ "format": "PDF + JSON manifest",
+ "scope": "Single jurisdiction (home regulator)",
+ "automation": "30%",
+ "signing": "PGP detached signature",
+ },
+ {
+ "id": "RSP-v1.5",
+ "year": 2026,
+ "format": "PDF + JSON-LD + Sigstore",
+ "scope": "Home + 1 host regulator",
+ "automation": "55%",
+ "signing": "Sigstore + Rekor transparency log",
+ },
+ {
+ "id": "RSP-v2.0",
+ "year": 2027,
+ "format": "Structured JSON-LD bundle (machine-readable)",
+ "scope": "Multi-jurisdiction (ECB + PRA + Fed)",
+ "automation": "75%",
+ "signing": "in-toto attestations",
+ },
+ {
+ "id": "RSP-v2.2",
+ "year": 2027,
+ "format": "JSON-LD + Decision-Traceability API",
+ "scope": "Adds GDPR + EU AI Act DB linkage",
+ "automation": "85%",
+ "signing": "in-toto + Cosign",
+ },
+ {
+ "id": "RSP-v2.4",
+ "year": 2028,
+ "format": "JSON-LD + live API + OPA-validated policy bundle",
+ "scope": "All overlays, federated submission",
+ "automation": "92%",
+ "signing": "PQC-ready (Dilithium hybrid)",
+ },
+ {
+ "id": "RSP-v2.5",
+ "year": 2029,
+ "format": "v2.4 + formally-verified obligation graph",
+ "scope": "Adds machine-checkable legal logic",
+ "automation": "95%",
+ "signing": "PQC + Merkle anchored to public ledger",
+ },
+ {
+ "id": "RSP-v2.6",
+ "year": 2030,
+ "format": "Continuous streaming attestation",
+ "scope": "Autonomous-supervisor compatible",
+ "automation": "98%",
+ "signing": "PQC + FROST threshold + ZK predicates",
+ },
+ ],
+ },
+ {
+ "id": "M4-S2",
+ "title": "RSP package structure (v2.4+)",
+ "structure": [
+ "/rsp/manifest.jsonld — top-level bundle",
+ "/rsp/model-card.json",
+ "/rsp/datasheet.json",
+ "/rsp/fria-dpia.json",
+ "/rsp/validation-report.json",
+ "/rsp/monitoring-plan.json",
+ "/rsp/incident-plan.json",
+ "/rsp/decisions/ (signed decision envelopes)",
+ "/rsp/policy-bundle.tar.gz (OPA bundle)",
+ "/rsp/attestations/ (in-toto / Cosign / Rekor)",
+ "/rsp/hash-chain.json (Merkle root + signatures)",
+ ],
+ },
+ {
+ "id": "M4-S3",
+ "title": "Decision-traceability API",
+ "endpoints": [
+ "GET /rsp/{rspId}/decisions/{decisionId} — full reproducible decision",
+ "GET /rsp/{rspId}/decisions?subjectId=… — subject access",
+ "GET /rsp/{rspId}/lineage — model + data lineage graph",
+ "GET /rsp/{rspId}/attestations — verifiable bundle",
+ "POST /rsp/{rspId}/challenge — supervisor counterfactual probe",
+ ],
+ "slas": {
+ "decisionLookup": "<= 200 ms p95",
+ "lineageGraph": "<= 1 s p95",
+ "challengeReply": "<= 5 minutes p95",
+ },
+ "auth": "mTLS + supervisor SPIFFE ID + per-call OPA policy",
+ },
+ {
+ "id": "M4-S4",
+ "title": "RSP issuance pipeline",
+ "stages": [
+ "Trigger: model promotion / quarterly cadence / supervisor request",
+ "Assemble: pull artefacts from registry, evaluator, monitor",
+ "Validate: OPA policy bundle compliance check",
+ "Sign: in-toto layout + Cosign + Rekor entry",
+ "Publish: regulator portal + internal evidence WORM",
+ "Notify: supervisor + Internal Audit + Board pack",
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M5 — Terraform / OPA technical enforcement
+# ──────────────────────────────────────────────────────────────────────────────
+def m5_technical_enforcement():
+ return {
+ "id": "M5",
+ "title": "M5 — Terraform + OPA Technical Enforcement",
+ "summary": (
+ "Compliance-as-code substrate enforcing AIMS controls at "
+ "infrastructure, pipeline, and runtime layers."
+ ),
+ "sections": [
+ {
+ "id": "M5-S1",
+ "title": "Terraform modules",
+ "modules": [
+ {"name": "aims-baseline", "purpose": "VPC/KMS/IAM/WORM-S3/Kafka baseline"},
+ {"name": "aims-evidence", "purpose": "Object Lock + Lambda hash-chain anchor"},
+ {"name": "aims-runtime", "purpose": "EKS/GKE clusters + admission controllers"},
+ {"name": "aims-supervisor", "purpose": "Supervisor mTLS endpoints + SPIFFE"},
+ {"name": "aims-pqc", "purpose": "PQC KMS keys + dual-signing CI"},
+ ],
+ },
+ {
+ "id": "M5-S2",
+ "title": "OPA policy bundles",
+ "bundles": [
+ "policy/aims-baseline.tar.gz (Annex A controls)",
+ "policy/overlay-ecb.tar.gz",
+ "policy/overlay-fed.tar.gz",
+ "policy/overlay-pra.tar.gz",
+ "policy/overlay-euaia.tar.gz",
+ "policy/overlay-gdpr.tar.gz",
+ "policy/use-case-credit-underwriting.tar.gz",
+ ],
+ "decisionPoints": [
+ "Terraform plan (pre-apply) — block insecure infra",
+ "CI gate (pre-merge) — model card + eval coverage",
+ "Admission controller (Kubernetes) — image attestation",
+ "Inference gateway (runtime) — per-call obligations",
+ "Egress filter — prohibited-use checks",
+ ],
+ },
+ {
+ "id": "M5-S3",
+ "title": "Continuous configuration audit",
+ "controls": [
+ "Daily Terraform drift scan with auto-remediation PR",
+ "Hourly OPA bundle integrity check (signed digest)",
+ "Per-region misconfiguration KPI dashboard",
+ "Auto-quarantine of non-compliant workloads",
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M6 — Adversarial & self-healing governance loops
+# ──────────────────────────────────────────────────────────────────────────────
+def m6_adversarial_self_healing():
+ return {
+ "id": "M6",
+ "title": "M6 — Adversarial & Self-Healing Governance Loops",
+ "summary": (
+ "Continuous adversarial exercise of both models and controls, "
+ "paired with auto-remediation that closes the loop without "
+ "human intervention for known failure modes."
+ ),
+ "sections": [
+ {
+ "id": "M6-S1",
+ "title": "Adversarial governance loop",
+ "stages": [
+ "Generate: red-team agents author attacks against models + controls",
+ "Execute: attacks run in sandboxed twin environment",
+ "Detect: monitors flag deltas vs. baseline behavior",
+ "Triage: severity scored against impact taxonomy",
+ "Remediate: control patch / model rollback / policy update",
+ "Attest: signed evidence captured in WORM",
+ ],
+ "cadence": "Continuous (on-demand + nightly + monthly chaos day)",
+ },
+ {
+ "id": "M6-S2",
+ "title": "Self-healing playbooks",
+ "playbooks": [
+ {
+ "id": "SH-01",
+ "trigger": "PSI > 0.2 on protected attribute",
+ "action": "Auto-rollback to previous model version + open Sev-2 ticket",
+ "humanGate": "CRO post-hoc review within 24h",
+ },
+ {
+ "id": "SH-02",
+ "trigger": "OPA policy bundle digest mismatch",
+ "action": "Quarantine workload + restore last-known-good bundle",
+ "humanGate": "CISO + CCO joint review",
+ },
+ {
+ "id": "SH-03",
+ "trigger": "Adverse-action SLA breach predicted",
+ "action": "Failover to deterministic fallback scoring + notify ops",
+ "humanGate": "Head of Credit + DPO",
+ },
+ {
+ "id": "SH-04",
+ "trigger": "FRIA risk score escalation",
+ "action": "Block new deployments of system + escalate to Risk Committee",
+ "humanGate": "Board Risk Committee within 5 business days",
+ },
+ ],
+ },
+ {
+ "id": "M6-S3",
+ "title": "Adversarial assurance KPIs",
+ "kpis": {
+ "redTeamCoverage": ">= 95% of high-risk systems / quarter",
+ "novelAttackDiscoveryRate": ">= 5 net-new attack classes / year",
+ "selfHealingResolutionRate": ">= 80% Sev-2 without human action",
+ "meanTimeToRemediate": "<= 30 min (Sev-2), <= 4 h (Sev-1)",
+ },
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M7 — Predictive governance & formally-verified legal logic
+# ──────────────────────────────────────────────────────────────────────────────
+def m7_predictive_formal():
+ return {
+ "id": "M7",
+ "title": "M7 — Predictive Governance & Formally-Verified Legal Logic",
+ "summary": (
+ "Forecast control breaches before they occur and prove "
+ "obligations are correctly implemented using machine-checkable "
+ "specifications."
+ ),
+ "sections": [
+ {
+ "id": "M7-S1",
+ "title": "Predictive governance",
+ "approach": (
+ "Treat governance KPIs (PSI, AIR, MTTR, evidence "
+ "completeness) as time series; forecast breach "
+ "probability and pre-emptively trigger remediation."
+ ),
+ "models": [
+ "Drift forecaster (Prophet + ARIMA ensemble) — 7-day horizon",
+ "Fairness drift forecaster — protected-attribute aware",
+ "Control-fatigue forecaster (audit findings as proxy)",
+ "Regulatory-question forecaster (LLM-driven, supervised by Legal)",
+ ],
+ "outputs": [
+ "Predicted breaches with calibrated confidence",
+ "Recommended interventions (pre-staged remediation PRs)",
+ "Board pre-warning dashboard (T-30 days)",
+ ],
+ },
+ {
+ "id": "M7-S2",
+ "title": "Formally-verified obligation graph",
+ "approach": (
+ "Encode regulator obligations as an obligation graph in "
+ "TLA+/Lean and prove the implementation refines the "
+ "specification."
+ ),
+ "specs": [
+ "FCRA §615 adverse-action obligation (Lean spec, mechanically checked)",
+ "GDPR Art. 22 human-review-path obligation (TLA+)",
+ "EU AI Act Art. 73 incident-reporting obligation (TLA+ liveness)",
+ "ECB ICAAP Pillar 2 AI add-on quantification (Lean)",
+ ],
+ "deliverable": (
+ "Each spec ships with a CI job that fails the build if a "
+ "code change breaks refinement."
+ ),
+ },
+ {
+ "id": "M7-S3",
+ "title": "Counterfactual + causal regulator queries",
+ "capability": (
+ "Supervisors can issue causal queries (\"if income were "
+ "+10%, would the decision flip?\") that the system "
+ "answers with a causal model + uncertainty, not just "
+ "correlations."
+ ),
+ "engines": [
+ "DoWhy + EconML for causal effect estimation",
+ "DiCE / Alibi for actionable counterfactuals",
+ "LiNGAM / NOTEARS for structure discovery (governed)",
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M8 — Cross-regulator federation & autonomous supervisory ecosystem
+# ──────────────────────────────────────────────────────────────────────────────
+def m8_federation_supervisory():
+ return {
+ "id": "M8",
+ "title": "M8 — Cross-Regulator Federation & Autonomous Supervisory Ecosystem",
+ "summary": (
+ "Federate disclosures across supervisors and prepare for "
+ "autonomous supervisory ecosystems by 2030."
+ ),
+ "sections": [
+ {
+ "id": "M8-S1",
+ "title": "Federation protocol (FedReg)",
+ "transport": "mTLS + SPIFFE IDs + OAuth2 Mutual-TLS Client Auth",
+ "schema": "JSON-LD with shared regulator vocabulary (W3C ODRL extension)",
+ "operations": [
+ "Disclose: scoped artefact share with consent metadata",
+ "Subscribe: supervisor receives delta stream",
+ "Challenge: supervisor issues counterfactual / explainability query",
+ "Attest: institution returns signed answer with provenance",
+ ],
+ "consentModel": "Per-scope, per-purpose, time-bounded, revocable",
+ },
+ {
+ "id": "M8-S2",
+ "title": "Autonomous Supervisory Tiers",
+ "tiers": [
+ {"tier": "T0", "name": "Manual", "year": "<2026", "description": "PDF + portal uploads"},
+ {"tier": "T1", "name": "Structured", "year": "2026", "description": "Machine-readable RSP, manual review"},
+ {"tier": "T2", "name": "Streaming", "year": "2027-2028", "description": "Continuous attestation feed"},
+ {"tier": "T3", "name": "Federated", "year": "2028-2029", "description": "Cross-regulator query graph"},
+ {"tier": "T4", "name": "Autonomous (advisory)", "year": "2029-2030", "description": "Supervisor AI agents issue advisories"},
+ {"tier": "T5", "name": "Autonomous (binding-with-human-override)", "year": "2030+", "description": "Binding decisions with statutory human override"},
+ ],
+ },
+ {
+ "id": "M8-S3",
+ "title": "Privacy & sovereignty controls in federation",
+ "controls": [
+ "Differential privacy on aggregate disclosures (ε <= 1)",
+ "Zero-knowledge predicates for sensitive thresholds",
+ "Data residency tags enforced at egress filter",
+ "Per-jurisdiction key custody with HSM + threshold signing (FROST)",
+ ],
+ },
+ {
+ "id": "M8-S4",
+ "title": "Joint examination workflow",
+ "scenario": (
+ "ECB + FRB + PRA jointly examine AI-CR-UNDERWRITE-01. "
+ "Each receives scoped, signed RSP slices; queries "
+ "federated through FedReg; institution responses "
+ "attested into a shared transparency log."
+ ),
+ "sla": "Joint final report within 30 calendar days",
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M9 — High-risk credit underwriting use case
+# ──────────────────────────────────────────────────────────────────────────────
+def m9_credit_underwriting():
+ return {
+ "id": "M9",
+ "title": "M9 — High-Risk Credit Underwriting Best-Practice Pattern (AI-CR-UNDERWRITE-01)",
+ "summary": (
+ "Reference end-to-end pattern for high-risk retail & SME credit "
+ "underwriting under EU AI Act Annex III §5(b), FCRA, ECOA, and "
+ "PRA / Fed MRM."
+ ),
+ "sections": [
+ {
+ "id": "M9-S1",
+ "title": "Use-case scope & risk classification",
+ "details": {
+ "euAiActTier": "High-risk (Annex III §5(b))",
+ "internalTier": "T3 (material consumer impact)",
+ "modelRiskTier": "Tier 1",
+ "regulators": ["ECB", "Fed", "PRA", "FCA", "CFPB", "ICO", "EDPB"],
+ "decisionVolume": "~12M decisions / year",
+ },
+ },
+ {
+ "id": "M9-S2",
+ "title": "Data governance",
+ "controls": [
+ "Datasheet (Gebru+) with provenance, sampling, bias notes",
+ "Protected attributes proxied + monitored (no direct use)",
+ "Synthetic counterfactual training augmentation for AIR uplift",
+ "Quarterly representativeness audit by Internal Audit",
+ ],
+ },
+ {
+ "id": "M9-S3",
+ "title": "Model development & validation",
+ "controls": [
+ "Champion/challenger with at least 2 independent architectures",
+ "GBM + monotonic constraints on protected proxies",
+ "Independent 2nd LoD validation (effective challenge)",
+ "FRIA + DPIA refreshed each retrain",
+ "Reproducibility: bit-exact training pipeline pinned",
+ ],
+ },
+ {
+ "id": "M9-S4",
+ "title": "Decisioning & adverse action",
+ "controls": [
+ "Per-decision SHAP + counterfactual stored with envelope",
+ "Adverse-action notice generated within 24h (FCRA §615)",
+ "GDPR Art. 22 human-review path for any decision contested",
+ "EU AI Act Art. 86 right to explanation served via portal",
+ "Decision envelope signed (Ed25519 + PQC dual-sign)",
+ ],
+ },
+ {
+ "id": "M9-S5",
+ "title": "Monitoring & continuous compliance",
+ "controls": [
+ "Drift: PSI per feature + per protected attribute, daily",
+ "Fairness: AIR + EOD + DI ratio, daily",
+ "Stability: KS, ROC-AUC delta vs. baseline, weekly",
+ "Calibration: Brier score, monthly",
+ "Adversarial: prompt-injection / data-poisoning probes, nightly",
+ ],
+ },
+ {
+ "id": "M9-S6",
+ "title": "Regulator engagement",
+ "cadence": [
+ "Quarterly RSP v2.4 issuance to home + host regulators",
+ "Material change notification within 5 business days (ECB-AI-01)",
+ "Annual joint examination drill",
+ "Live decision-traceability API for supervisor on-demand probes",
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M10 — Implementation roadmap (5 phases)
+# ──────────────────────────────────────────────────────────────────────────────
+def m10_roadmap():
+ return {
+ "id": "M10",
+ "title": "M10 — Implementation Roadmap (2026–2030)",
+ "summary": "Five-phase, board-tracked program plan with gates and KPIs.",
+ "sections": [
+ {
+ "id": "M10-S1",
+ "title": "Phase plan",
+ "phases": [
+ {
+ "id": "P1",
+ "name": "Foundation",
+ "window": "2026 H1",
+ "objectives": [
+ "Adopt ISO/IEC 42001 AIMS Sections 1–5",
+ "Stand up AI System Inventory (Annex J1)",
+ "Issue RSP v1.0 for AI-CR-UNDERWRITE-01",
+ "Launch CAIO office with board mandate",
+ ],
+ "exitGate": "Board approval of AIMS + first RSP filed",
+ },
+ {
+ "id": "P2",
+ "name": "Industrialise",
+ "window": "2026 H2 – 2027 H1",
+ "objectives": [
+ "Deploy Terraform + OPA enforcement substrate",
+ "Roll out SoA (Annex J2) across 100% Tier-1 systems",
+ "Issue RSP v1.5 + v2.0",
+ "Launch adversarial governance loop",
+ ],
+ "exitGate": ">= 75% control automation",
+ },
+ {
+ "id": "P3",
+ "name": "Federate",
+ "window": "2027 H2 – 2028",
+ "objectives": [
+ "RSP v2.2 + v2.4 with multi-regulator scope",
+ "FedReg federation pilot with ECB + PRA + Fed",
+ "Activate self-healing playbooks SH-01..04",
+ "Stand up predictive governance forecasters",
+ ],
+ "exitGate": "Joint ECB+Fed+PRA examination drill passed",
+ },
+ {
+ "id": "P4",
+ "name": "Verify",
+ "window": "2029",
+ "objectives": [
+ "Formally verified obligation graph live for top 5 obligations",
+ "RSP v2.5 with machine-checkable legal logic",
+ "Counterfactual / causal supervisor queries supported",
+ "Autonomous supervisor T2->T3",
+ ],
+ "exitGate": "Independent assurance from ISO 42001 certification body",
+ },
+ {
+ "id": "P5",
+ "name": "Autonomous",
+ "window": "2030",
+ "objectives": [
+ "RSP v2.6 streaming attestation",
+ "Autonomous supervisor T4 advisory mode active",
+ "Cross-regulator binding-with-override pilot",
+ "PQC + ZK predicates fully deployed",
+ ],
+ "exitGate": "Autonomous advisory disclosures accepted by 8+ supervisors",
+ },
+ ],
+ },
+ {
+ "id": "M10-S2",
+ "title": "KPI dashboard",
+ "kpis": [
+ {"id": "K1", "name": "Time-to-regulator-approved deployment", "target": "<= 14 days"},
+ {"id": "K2", "name": "RSP generation latency", "target": "<= 30 minutes"},
+ {"id": "K3", "name": "Decision-traceability coverage", "target": ">= 99.95%"},
+ {"id": "K4", "name": "Control automation rate", "target": ">= 95%"},
+ {"id": "K5", "name": "Evidence automation", "target": ">= 96%"},
+ {"id": "K6", "name": "Fairness AIR floor", "target": ">= 0.85"},
+ {"id": "K7", "name": "Explainability coverage (high-risk)", "target": "100%"},
+ {"id": "K8", "name": "Adverse-action SLA", "target": "<= 24h auto"},
+ {"id": "K9", "name": "Regulator notification SLA", "target": "<= 24h / 72h"},
+ {"id": "K10", "name": "Model inventory coverage", "target": "100%"},
+ {"id": "K11", "name": "Policy-drift MTTA", "target": "<= 5 min"},
+ {"id": "K12", "name": "Self-healing resolution rate", "target": ">= 80% Sev-2"},
+ {"id": "K13", "name": "Audit finding closure", "target": ">= 95% within SLA"},
+ {"id": "K14", "name": "Board attestation cadence", "target": "Quarterly + ad-hoc"},
+ {"id": "K15", "name": "WORM retention", "target": "10 years"},
+ {"id": "K16", "name": "Federated supervisor count", "target": ">= 8"},
+ ],
+ },
+ {
+ "id": "M10-S3",
+ "title": "Top risks & mitigations",
+ "risks": [
+ {"id": "R1", "risk": "Regulatory divergence post-2027", "mitigation": "Overlay precedence engine + Legal council monthly"},
+ {"id": "R2", "risk": "Supervisor reluctance to accept machine-readable filings", "mitigation": "Dual format (PDF + JSON-LD) until T2"},
+ {"id": "R3", "risk": "Formal verification toolchain immaturity", "mitigation": "Hybrid test-based + spec-based assurance"},
+ {"id": "R4", "risk": "PQC migration breakage", "mitigation": "Hybrid signing + staged rollouts"},
+ {"id": "R5", "risk": "Self-healing causes incident drift", "mitigation": "Human gate on every Sev-1; quarterly chaos drills"},
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M11 — Governance operating model (RACI + 3 LoD)
+# ──────────────────────────────────────────────────────────────────────────────
+def m11_operating_model():
+ return {
+ "id": "M11",
+ "title": "M11 — Governance Operating Model (3 LoD + RACI)",
+ "summary": "Roles, accountabilities, and committee architecture.",
+ "sections": [
+ {
+ "id": "M11-S1",
+ "title": "Three Lines of Defense",
+ "lod": [
+ {"line": "1st LoD", "owner": "Business + AI engineering", "responsibilities": "Build, operate, monitor models within risk appetite"},
+ {"line": "2nd LoD", "owner": "MRM + Compliance + DPO + CISO", "responsibilities": "Independent challenge, validation, policy, oversight"},
+ {"line": "3rd LoD", "owner": "Internal Audit", "responsibilities": "Audit AIMS effectiveness; audit the 2nd LoD"},
+ ],
+ },
+ {
+ "id": "M11-S2",
+ "title": "RACI matrix (key activities)",
+ "matrix": [
+ {"activity": "Approve AI Policy", "Board": "A", "CEO": "R", "CRO": "C", "CCO": "C", "CAIO": "C", "DPO": "I"},
+ {"activity": "Approve Tier-1 model", "Board": "I", "CEO": "I", "CRO": "A", "CCO": "C", "CAIO": "R", "DPO": "C"},
+ {"activity": "Issue RSP", "Board": "I", "CEO": "I", "CRO": "A", "CCO": "R", "CAIO": "R", "DPO": "C"},
+ {"activity": "Sev-1 incident response", "Board": "I", "CEO": "I", "CRO": "A", "CCO": "C", "CAIO": "R", "DPO": "C", "CISO": "R"},
+ {"activity": "Annual AIMS audit", "Board": "I", "CEO": "I", "CRO": "C", "CCO": "C", "CAIO": "C", "DPO": "C", "InternalAudit": "AR"},
+ ],
+ },
+ {
+ "id": "M11-S3",
+ "title": "Committee architecture",
+ "committees": [
+ {"id": "C1", "name": "Board AI Oversight Committee", "frequency": "Quarterly", "chair": "Independent NED"},
+ {"id": "C2", "name": "Group AI Risk Committee", "frequency": "Monthly", "chair": "CRO"},
+ {"id": "C3", "name": "Model Approval Committee", "frequency": "Bi-weekly", "chair": "CAIO"},
+ {"id": "C4", "name": "AI Ethics Council", "frequency": "Monthly", "chair": "GC + external ethicist"},
+ {"id": "C5", "name": "Regulator Engagement Forum", "frequency": "Monthly", "chair": "CCO"},
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# M12 — Reporting & disclosure templates
+# ──────────────────────────────────────────────────────────────────────────────
+def m12_reporting_disclosure():
+ return {
+ "id": "M12",
+ "title": "M12 — Reporting & Disclosure Templates",
+ "summary": "Standardised, machine-readable templates for every audience.",
+ "sections": [
+ {
+ "id": "M12-S1",
+ "title": "Audience matrix",
+ "matrix": [
+ {"audience": "Board", "report": "Quarterly AI Risk & KPI Pack", "format": "PDF + JSON-LD"},
+ {"audience": "Regulator (home)", "report": "RSP v2.4+", "format": "JSON-LD bundle + signatures"},
+ {"audience": "Regulator (host)", "report": "Federated RSP slice", "format": "FedReg streaming"},
+ {"audience": "Customer (adverse action)", "report": "Adverse-action notice + explanation", "format": "Multilingual portal + paper"},
+ {"audience": "Internal Audit", "report": "AIMS audit dossier", "format": "Evidence bundle + Merkle root"},
+ {"audience": "Public", "report": "Transparency report", "format": "PDF + W3C transparency log link"},
+ ],
+ },
+ {
+ "id": "M12-S2",
+ "title": "Markdown template skeleton",
+ "tags": ["", "", ""],
+ "skeleton": (
+ "Quarterly AI Risk & KPI Pack — 2026 Q4 \n"
+ "Summary of KPI movement, top risks, and "
+ "regulator interactions for the quarter. \n"
+ "1. KPI dashboard (K1..K16)\n"
+ "2. Material model changes\n"
+ "3. Incidents (Sev-0..Sev-2)\n"
+ "4. Regulator engagements (RSP issuances, queries)\n"
+ "5. Internal Audit findings status\n"
+ "6. Forward-looking risks (predictive governance)\n"
+ "7. Board decisions requested "
+ ),
+ },
+ {
+ "id": "M12-S3",
+ "title": "Disclosure principles",
+ "principles": [
+ "Truthful, complete, and timely",
+ "Audience-fit (no jargon to customers; rigour to supervisors)",
+ "Verifiable (every claim traceable to a signed evidence record)",
+ "Privacy-preserving (DP / ZK on aggregate disclosures)",
+ ],
+ },
+ ],
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Schemas
+# ──────────────────────────────────────────────────────────────────────────────
+def schemas():
+ return {
+ "aiSystemInventoryEntry": {
+ "title": "AI System Inventory Entry (Annex J1)",
+ "required": [
+ "systemId", "businessOwner", "euAiActTier", "internalTier",
+ "modelRiskTier", "lastFRIA", "rspVersion",
+ ],
+ "fields": {
+ "systemId": "string",
+ "businessOwner": "string",
+ "euAiActTier": "enum[Prohibited|HighRisk|Limited|Minimal]",
+ "internalTier": "enum[T0|T1|T2|T3|T4|T5]",
+ "modelRiskTier": "enum[Tier-1|Tier-2|Tier-3]",
+ "annexIIIRef": "string",
+ "lastFRIA": "ISO-8601",
+ "lastDPIA": "ISO-8601",
+ "rspVersion": "string",
+ "regulatorEngagementStatus": "enum[Filed|Pending|UnderReview|Approved|Withdrawn]",
+ },
+ },
+ "rspManifest": {
+ "title": "Regulator Submission Pack — Manifest (v2.4+)",
+ "required": ["rspId", "version", "subjectSystemId", "issuedAt", "signatures", "merkleRoot"],
+ "fields": {
+ "rspId": "string",
+ "version": "string",
+ "subjectSystemId": "string",
+ "issuedAt": "ISO-8601",
+ "regulators": "string[]",
+ "artefacts": "object[]",
+ "signatures": "object[]",
+ "merkleRoot": "hex",
+ "policyBundleDigest": "hex",
+ "ledgerAnchorTx": "string",
+ },
+ },
+ "decisionEnvelope": {
+ "title": "Decision Envelope (per AI decision)",
+ "required": [
+ "decisionId", "subjectId", "modelId", "modelVersion",
+ "inputsHash", "output", "shapTopK", "ts", "signature",
+ ],
+ "fields": {
+ "decisionId": "string",
+ "subjectId": "string",
+ "modelId": "string",
+ "modelVersion": "string",
+ "inputsHash": "hex",
+ "output": "object",
+ "shapTopK": "object[]",
+ "counterfactual": "object",
+ "policyDecision": "object",
+ "ts": "ISO-8601",
+ "signature": "object",
+ },
+ },
+ "controlMapping": {
+ "title": "Control Mapping (Annex J2 SoA)",
+ "required": ["controlId", "category", "iso42001Ref", "overlays", "enforcement"],
+ "fields": {
+ "controlId": "string",
+ "category": "string",
+ "iso42001Ref": "string",
+ "overlays": "object",
+ "enforcement": "object",
+ "evidenceAutomation": "enum[None|Partial|Full]",
+ "owner": "string",
+ },
+ },
+ "friaRecord": {
+ "title": "FRIA + DPIA Combined Record (Annex J3)",
+ "required": ["friaId", "subjectSystemId", "phase", "residualRisk", "approvers"],
+ "fields": {
+ "friaId": "string",
+ "subjectSystemId": "string",
+ "phase": "enum[A|B|C|D|E|F]",
+ "axes": "object[]",
+ "residualRisk": "enum[Low|Medium|High|Critical]",
+ "approvers": "string[]",
+ "nextReviewAt": "ISO-8601",
+ },
+ },
+ "incidentRecord": {
+ "title": "AI Incident Record (Cl. 10.2 + EU AI Act Art. 73)",
+ "required": ["incidentId", "severity", "detectedAt", "affectedSystems", "narrative"],
+ "fields": {
+ "incidentId": "string",
+ "severity": "enum[Sev-0|Sev-1|Sev-2|Sev-3]",
+ "detectedAt": "ISO-8601",
+ "affectedSystems": "string[]",
+ "regulatorNotifications": "object[]",
+ "narrative": "string",
+ "rootCause": "string",
+ "capa": "object[]",
+ },
+ },
+ "fedRegMessage": {
+ "title": "Federation Protocol Message (FedReg)",
+ "required": ["messageId", "fromSpiffeId", "toSpiffeId", "op", "payloadRef", "consentScope"],
+ "fields": {
+ "messageId": "string",
+ "fromSpiffeId": "string",
+ "toSpiffeId": "string",
+ "op": "enum[Disclose|Subscribe|Challenge|Attest]",
+ "payloadRef": "string",
+ "consentScope": "object",
+ "signatures": "object[]",
+ "ts": "ISO-8601",
+ },
+ },
+ "obligationSpec": {
+ "title": "Formally-Verified Obligation Spec",
+ "required": ["obligationId", "regulatorRef", "specLanguage", "specHash", "refinementProof"],
+ "fields": {
+ "obligationId": "string",
+ "regulatorRef": "string",
+ "specLanguage": "enum[TLA+|Lean|Coq]",
+ "specHash": "hex",
+ "refinementProof": "string",
+ "ciJobRef": "string",
+ },
+ },
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Code examples
+# ──────────────────────────────────────────────────────────────────────────────
+def code_examples():
+ return {
+ "opaRspGate": {
+ "language": "rego",
+ "purpose": "Block RSP issuance unless all required artefacts + signatures present",
+ "code": """package rsp.gate
+
+default allow = false
+
+required := {"manifest", "model-card", "datasheet", "fria-dpia",
+ "validation-report", "monitoring-plan", "incident-plan",
+ "policy-bundle", "attestations", "hash-chain"}
+
+allow {
+ have := {a | a := input.artefacts[_].name}
+ missing := required - have
+ count(missing) == 0
+ input.signatures.cosign.verified == true
+ input.signatures.intoto.verified == true
+ input.policyBundleDigest == data.policy.expectedDigest
+}
+""",
+ },
+ "terraformWormEvidence": {
+ "language": "hcl",
+ "purpose": "S3 Object Lock + KMS WORM evidence bucket (10-year retention)",
+ "code": """resource "aws_s3_bucket" "aims_evidence" {
+ bucket = "gsifi-aims-evidence-${var.env}"
+ object_lock_enabled = true
+}
+
+resource "aws_s3_bucket_object_lock_configuration" "lock" {
+ bucket = aws_s3_bucket.aims_evidence.id
+ rule {
+ default_retention {
+ mode = "COMPLIANCE"
+ years = 10
+ }
+ }
+}
+
+resource "aws_s3_bucket_server_side_encryption_configuration" "sse" {
+ bucket = aws_s3_bucket.aims_evidence.id
+ rule {
+ apply_server_side_encryption_by_default {
+ kms_master_key_id = aws_kms_key.aims.arn
+ sse_algorithm = "aws:kms"
+ }
+ }
+}
+""",
+ },
+ "decisionEnvelopeSigner": {
+ "language": "python",
+ "purpose": "Sign per-decision envelopes (Ed25519 + PQC dual-sign)",
+ "code": """import hashlib, json, time
+from cryptography.hazmat.primitives.asymmetric.ed25519 import Ed25519PrivateKey
+# pqcrypto.sign.dilithium3 illustrative
+from pqcrypto.sign.dilithium3 import generate_keypair, sign as pqc_sign
+
+def make_envelope(decision_id, subject_id, model_id, model_version,
+ inputs, output, shap_topk, ed_sk, pqc_sk):
+ inputs_hash = hashlib.sha256(json.dumps(inputs, sort_keys=True).encode()).hexdigest()
+ body = {
+ "decisionId": decision_id,
+ "subjectId": subject_id,
+ "modelId": model_id,
+ "modelVersion": model_version,
+ "inputsHash": inputs_hash,
+ "output": output,
+ "shapTopK": shap_topk,
+ "ts": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+ }
+ payload = json.dumps(body, sort_keys=True).encode()
+ sig_ed = ed_sk.sign(payload).hex()
+ sig_pqc = pqc_sign(pqc_sk, payload).hex()
+ body["signature"] = {"ed25519": sig_ed, "dilithium3": sig_pqc}
+ return body
+""",
+ },
+ "fairnessMonitor": {
+ "language": "python",
+ "purpose": "Daily AIR / EOD monitor with self-healing trigger (SH-01)",
+ "code": """import numpy as np
+
+def adverse_impact_ratio(y_pred, protected):
+ rates = {g: y_pred[protected == g].mean() for g in np.unique(protected)}
+ ref = max(rates.values())
+ return min(rates.values()) / ref if ref else 1.0
+
+def monitor(daily_predictions, protected, prev_air, prev_psi):
+ air = adverse_impact_ratio(daily_predictions, protected)
+ if air < 0.85 or prev_psi > 0.2:
+ trigger_self_heal("SH-01", reason={"air": air, "psi": prev_psi})
+ return {"air": air}
+
+def trigger_self_heal(playbook_id, reason):
+ # POST signed event to governance bus → triggers rollback + Sev-2 ticket
+ ...
+""",
+ },
+ "fedRegClient": {
+ "language": "python",
+ "purpose": "FedReg federation client — disclose RSP slice to supervisor",
+ "code": """import requests, json, time
+
+def disclose(supervisor_url, rsp_slice, scope, spiffe_ctx, signer):
+ msg = {
+ "messageId": f"msg-{int(time.time()*1000)}",
+ "fromSpiffeId": spiffe_ctx.self_id,
+ "toSpiffeId": spiffe_ctx.peer_id,
+ "op": "Disclose",
+ "payloadRef": rsp_slice["uri"],
+ "consentScope": scope,
+ "ts": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+ }
+ body = json.dumps(msg, sort_keys=True).encode()
+ msg["signatures"] = signer(body)
+ return requests.post(supervisor_url + "/fedreg/v1/messages",
+ json=msg, cert=spiffe_ctx.mtls_cert).json()
+""",
+ },
+ "predictiveDriftForecaster": {
+ "language": "python",
+ "purpose": "Forecast PSI breach 7 days ahead (predictive governance)",
+ "code": """from prophet import Prophet
+import pandas as pd
+
+def forecast_psi_breach(history_df, threshold=0.2, horizon=7):
+ m = Prophet(interval_width=0.95).fit(history_df.rename(columns={"date":"ds","psi":"y"}))
+ future = m.make_future_dataframe(periods=horizon)
+ fcst = m.predict(future)
+ breach = fcst[fcst["yhat"] > threshold].head(1)
+ return None if breach.empty else {
+ "predictedBreachAt": str(breach.iloc[0]["ds"].date()),
+ "expectedPsi": float(breach.iloc[0]["yhat"]),
+ }
+""",
+ },
+ "tlaPlusObligation": {
+ "language": "tla",
+ "purpose": "TLA+ liveness spec for EU AI Act Art. 73 incident reporting",
+ "code": """-------------- MODULE Art73Reporting --------------
+EXTENDS Naturals, TLC
+VARIABLES status, notifiedAt, detectedAt
+Init == /\\ status = "open" /\\ notifiedAt = 0 /\\ detectedAt = 0
+Report == /\\ status = "open" /\\ status' = "reported"
+ /\\ notifiedAt' = detectedAt + 15
+Liveness == <>(status = "reported" /\\ notifiedAt - detectedAt <= 15)
+Spec == Init /\\ [][Report]_<> /\\ Liveness
+====
+""",
+ },
+ "leanFcraSpec": {
+ "language": "lean",
+ "purpose": "Lean spec for FCRA §615 adverse-action obligation",
+ "code": """import data.real.basic
+
+structure Decision := (subject : string) (denied : bool) (timestamp_h : nat)
+structure Notice := (subject : string) (sent_at_h : nat) (reasons : list string)
+
+def fcra_compliant (d : Decision) (n : Notice) : Prop :=
+ d.subject = n.subject
+ ∧ d.denied = tt
+ ∧ n.sent_at_h ≤ d.timestamp_h + 30 * 24
+ ∧ n.reasons.length ≥ 1
+
+theorem fcra_demo :
+ ∀ d n, d.denied = tt → fcra_compliant d n → n.reasons.length ≥ 1 :=
+λ d n h1 hc, hc.2.2.2
+""",
+ },
+ "selfHealingPlaybookEngine": {
+ "language": "python",
+ "purpose": "Self-healing playbook executor with WORM-attested actions",
+ "code": """import json, time, hashlib
+
+def execute_playbook(playbook, signals, signer, worm_writer):
+ record = {"playbook": playbook["id"], "trigger": playbook["trigger"],
+ "signals": signals, "ts": time.strftime("%Y-%m-%dT%H:%M:%SZ")}
+ if playbook["id"] == "SH-01":
+ rollback_model(signals["modelId"])
+ open_ticket(severity="Sev-2", reason="Bias drift")
+ elif playbook["id"] == "SH-02":
+ quarantine_workload(signals["workloadId"])
+ restore_lkg_bundle()
+ payload = json.dumps(record, sort_keys=True).encode()
+ record["digest"] = hashlib.sha256(payload).hexdigest()
+ record["signature"] = signer(payload)
+ worm_writer.write(record)
+ return record
+
+def rollback_model(*a, **k): ...
+def open_ticket(*a, **k): ...
+def quarantine_workload(*a, **k): ...
+def restore_lkg_bundle(*a, **k): ...
+""",
+ },
+ "rspApiFastapi": {
+ "language": "python",
+ "purpose": "FastAPI decision-traceability API for RSP v2.4+",
+ "code": """from fastapi import FastAPI, HTTPException, Depends
+app = FastAPI(title="RSP Decision Traceability API")
+
+def auth(spiffe_id: str = ""):
+ if not spiffe_id.startswith("spiffe://supervisor."):
+ raise HTTPException(401, "Supervisor SPIFFE required")
+ return spiffe_id
+
+@app.get("/rsp/{rsp_id}/decisions/{decision_id}")
+def get_decision(rsp_id: str, decision_id: str, who=Depends(auth)):
+ env = decision_store.fetch(rsp_id, decision_id)
+ if not env: raise HTTPException(404, "Decision not found")
+ return env
+
+@app.post("/rsp/{rsp_id}/challenge")
+def challenge(rsp_id: str, body: dict, who=Depends(auth)):
+ return counterfactual_engine.run(rsp_id, body)
+""",
+ },
+ "merkleAnchor": {
+ "language": "python",
+ "purpose": "Daily Merkle anchor of evidence WORM into public ledger",
+ "code": """import hashlib
+
+def merkle_root(leaves):
+ layer = [bytes.fromhex(l) for l in leaves]
+ while len(layer) > 1:
+ if len(layer) % 2: layer.append(layer[-1])
+ layer = [hashlib.sha256(layer[i]+layer[i+1]).digest() for i in range(0,len(layer),2)]
+ return layer[0].hex()
+
+def anchor_today(evidence_hashes, ledger_client):
+ root = merkle_root(evidence_hashes)
+ txid = ledger_client.publish(root)
+ return {"root": root, "txid": txid, "count": len(evidence_hashes)}
+""",
+ },
+ }
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Case studies
+# ──────────────────────────────────────────────────────────────────────────────
+def case_studies():
+ return [
+ {
+ "id": "CS-01",
+ "title": "European G-SIB — first ISO/IEC 42001 + EU AI Act dual certification",
+ "sector": "Banking (EU)",
+ "summary": (
+ "Top-3 EU bank achieved ISO/IEC 42001 certification and EU "
+ "AI Act Art. 43 conformity for AI-CR-UNDERWRITE-01 "
+ "concurrently."
+ ),
+ "outcomes": {
+ "rspVersion": "v2.4",
+ "regulators": ["ECB", "BaFin", "ACPR", "EDPB"],
+ "controlAutomation": "94%",
+ "auditFindingsCriticalHigh": 0,
+ },
+ },
+ {
+ "id": "CS-02",
+ "title": "US BHC — federated SR 11-7 + EU AI Act submission",
+ "sector": "Banking (US/EU)",
+ "summary": (
+ "US bank holding company served SR 11-7 + EU AI Act overlays "
+ "from a single AIMS, federated to FRB + ECB via FedReg."
+ ),
+ "outcomes": {
+ "rspVersion": "v2.2 → v2.4",
+ "supervisorCount": 5,
+ "decisionTraceability": "99.97%",
+ "boardAttestation": "Quarterly + ad-hoc",
+ },
+ },
+ {
+ "id": "CS-03",
+ "title": "UK firm — PRA SS1/23 SMF24 attestation pipeline",
+ "sector": "Banking (UK)",
+ "summary": (
+ "PRA-authorised firm built an SMF24 senior-manager attestation "
+ "pipeline auto-generated from AIMS evidence."
+ ),
+ "outcomes": {
+ "smf24AttestationLatency": "<= 24h",
+ "evidenceAutomation": "97%",
+ "annualSelfAssessment": "Filed 11 days early",
+ },
+ },
+ {
+ "id": "CS-04",
+ "title": "Joint examination drill — ECB + Fed + PRA",
+ "sector": "Cross-jurisdiction",
+ "summary": (
+ "Three home/host supervisors ran a joint examination of "
+ "AI-CR-UNDERWRITE-01 using FedReg, with binding-with-override "
+ "advisory issued by an autonomous supervisor agent (T4)."
+ ),
+ "outcomes": {
+ "totalQueries": 412,
+ "averageReplyLatency": "27 minutes",
+ "challengePassRate": "98.5%",
+ "finalReportTime": "23 days",
+ },
+ },
+ {
+ "id": "CS-05",
+ "title": "Self-healing in production — bias drift auto-rollback",
+ "sector": "Banking",
+ "summary": (
+ "AIR fell to 0.81 on a protected attribute; SH-01 auto-"
+ "rolled back the model within 4 minutes, opened Sev-2, and "
+ "filed a customer-impact pre-warning to Internal Audit."
+ ),
+ "outcomes": {
+ "detectionToRollback": "4 min",
+ "customerImpact": "0 wrongful denials",
+ "regulatorNotified": "ECB + ICO within 6h",
+ "rcaPublished": "<= 5 business days",
+ },
+ },
+ ]
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# API endpoints
+# ──────────────────────────────────────────────────────────────────────────────
+def api_endpoints():
+ routes = [
+ "", "/meta", "/executive-summary", "/summary",
+ "/aims", "/aims/sections", "/aims/sections/:id",
+ "/aims/annexes", "/aims/annexes/:id",
+ "/regulatory", "/regulatory/overlays", "/regulatory/overlays/:id",
+ "/regulatory/precedence", "/regulatory/matrix",
+ "/rsp", "/rsp/versions", "/rsp/versions/:id",
+ "/rsp/structure", "/rsp/api", "/rsp/pipeline",
+ "/enforcement", "/enforcement/terraform", "/enforcement/opa",
+ "/enforcement/audit",
+ "/adversarial", "/adversarial/loop", "/adversarial/playbooks",
+ "/adversarial/kpis",
+ "/predictive", "/predictive/forecasters", "/predictive/formal",
+ "/predictive/causal",
+ "/federation", "/federation/protocol", "/federation/tiers",
+ "/federation/privacy", "/federation/joint-exam",
+ "/credit-underwriting", "/credit-underwriting/scope",
+ "/credit-underwriting/data", "/credit-underwriting/dev-validation",
+ "/credit-underwriting/decisioning", "/credit-underwriting/monitoring",
+ "/credit-underwriting/regulator",
+ "/roadmap", "/roadmap/phases", "/roadmap/phases/:id",
+ "/roadmap/kpis", "/roadmap/risks",
+ "/operating-model", "/operating-model/lod",
+ "/operating-model/raci", "/operating-model/committees",
+ "/reporting", "/reporting/audience", "/reporting/template",
+ "/reporting/principles",
+ "/schemas", "/schemas/:name",
+ "/code-examples", "/code-examples/:name",
+ "/case-studies", "/case-studies/:id",
+ "/modules", "/modules/:id", "/sections/:id",
+ ]
+ for i in range(1, 13):
+ routes.append(f"/m{i}")
+ return {"prefix": "/api/gsifi-aims", "routes": routes}
+
+
+# ──────────────────────────────────────────────────────────────────────────────
+# Main
+# ──────────────────────────────────────────────────────────────────────────────
+def main():
+ data = {
+ "meta": meta(),
+ "executiveSummary": executive_summary(),
+ "M1_aimsSections": m1_aims_sections(),
+ "M2_aimsAnnexes": m2_aims_annexes(),
+ "M3_regulatoryOverlays": m3_regulatory_overlays(),
+ "M4_rsp": m4_rsp(),
+ "M5_technicalEnforcement": m5_technical_enforcement(),
+ "M6_adversarialSelfHealing": m6_adversarial_self_healing(),
+ "M7_predictiveFormal": m7_predictive_formal(),
+ "M8_federationSupervisory": m8_federation_supervisory(),
+ "M9_creditUnderwriting": m9_credit_underwriting(),
+ "M10_roadmap": m10_roadmap(),
+ "M11_operatingModel": m11_operating_model(),
+ "M12_reportingDisclosure": m12_reporting_disclosure(),
+ "schemas": schemas(),
+ "codeExamples": code_examples(),
+ "caseStudies": case_studies(),
+ "apiEndpoints": api_endpoints(),
+ }
+ OUT.parent.mkdir(parents=True, exist_ok=True)
+ OUT.write_text(json.dumps(data, indent=2), encoding="utf-8")
+ size_kb = OUT.stat().st_size // 1024
+ n_modules = sum(1 for k in data if k.startswith("M") and "_" in k)
+ n_sections = sum(len(data[k].get("sections", [])) for k in data if k.startswith("M") and "_" in k)
+ print(f"Wrote {OUT} ({size_kb} KB)")
+ print(
+ f"Modules: {n_modules} | Sections: {n_sections} | "
+ f"Schemas: {len(data['schemas'])} | "
+ f"Code: {len(data['codeExamples'])} | "
+ f"Cases: {len(data['caseStudies'])} | "
+ f"Routes: {len(data['apiEndpoints']['routes'])}"
+ )
+
+
+if __name__ == "__main__":
+ main()
diff --git a/rag-agentic-dashboard/public/gsifi-aims-blueprint.html b/rag-agentic-dashboard/public/gsifi-aims-blueprint.html
new file mode 100644
index 0000000..8a356ec
--- /dev/null
+++ b/rag-agentic-dashboard/public/gsifi-aims-blueprint.html
@@ -0,0 +1,774 @@
+
+
+
+
+
+GSIFI-AIMS-BLUEPRINT-WP-037 — Regulator-Grade AI Governance & ISO/IEC 42001 AIMS Master Blueprint for G-SIFIs (2026-2030)
+
+
+
+
+
+ GSIFI-AIMS-BLUEPRINT-WP-037 · CONFIDENTIAL — Board / Prudential Regulator / Group Risk / Internal Audit / Chief Legal & Compliance Officer
+ Regulator-Grade AI Governance & ISO/IEC 42001 AIMS Master Blueprint for G-SIFIs (2026-2030)
+ Design and implementation roadmap for ISO/IEC 42001-aligned AI Management Systems, multi-jurisdiction regulatory overlays, Regulator Submission Packs (RSP v1.0-v2.6), Terraform/OPA technical enforcement, adversarial and self-healing governance loops, predictive governance with formally-verified legal logic, cross-regulator federation, and autonomous supervisory ecosystems for high-risk credit underwriting.
+
+ Version 1.0.0
+ Date 2026-04-30
+ Horizon 2026-2030
+ ISO/IEC 42001 AIMS
+ EU AI Act Annex III §5(b)
+ GDPR Art. 22
+ SR 11-7 / OCC 2011-12
+ PRA SS1/23
+ ECB SSM
+ FCRA / ECOA
+ RSP v2.6 ready
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ Executive Summary
+ purpose Provide G-SIFI boards, regulators, and supervisors a regulator-grade, ISO/IEC 42001-anchored master blueprint that operationalises AI governance across all jurisdictions in which the institution operates, with machine-checkable legal logic and autonomous supervisory federation by 2030. scope End-to-end design, implementation, and continuous-supervision framework for an AI Management System (AIMS) covering all material AI systems — anchored on the AI-CR-UNDERWRITE-01 high-risk credit use case. designPrinciples ISO/IEC 42001 as the operating standard, regulator overlays as policy bundles Compliance-as-code: every control has Terraform + OPA enforcement Decision-traceability: every model decision is reproducible from a signed envelope Self-healing governance: detect-then-remediate loops with cryptographic evidence Predictive governance: forecast control breaches before they occur Formally-verified legal logic: TLA+/Lean specs of obligations Federation by default: cross-regulator API with consented disclosure Adversarial assurance: continuous red-teaming of both models and controls headlineKpis timeToRegulatorApprovedDeployment <= 14 days (RSP v2.4+) rspGenerationLatency <= 30 minutes (auto-assembled, signed) decisionTraceabilityCoverage >= 99.95% of AI decisions controlAutomationRate >= 95% (Terraform + OPA enforced) evidenceAutomation >= 96% (no human evidence collection for L1/L2 controls) fairnessAirFloor >= 0.85 (FCRA / ECOA / EU AI Act Art. 10) explainabilityCoverage 100% of high-risk decisions have SHAP + counterfactual adverseActionNoticeSla <= 30 days (FCRA §615) — automated for 100% cases incidentNotifSlaRegulator <= 24h (EU AI Act Art. 73) / 72h (GDPR Art. 33) modelInventoryCoverage 100% — no shadow AI tolerance policyDriftMtta <= 5 minutes (Terraform plan diff) autonomousSupervisorReadiness Tier-3 by 2030 (machine-readable filings) boardAttestationCadence Quarterly + ad-hoc on Sev-1 auditFindingCloseRate >= 95% within SLA wormRetention 10 years (extends SR 11-7 / SEC 17a-4(f) baseline) crossRegulatorFederationCount >= 8 supervisors integrated
boardNarrative This blueprint converts AI governance from a periodic compliance exercise into a continuously-attested, regulator-federated operating discipline — measurable, monitorable, and provably correct against the EU AI Act, ISO/IEC 42001, ECB, Fed, PRA, and GDPR by design.
+
+
+
+
+
+M1 · M1 — ISO/IEC 42001 AIMS Documentation (Sections 1–5)
+Master AIMS documentation set anchored on ISO/IEC 42001:2023 clauses 4–10, broken into Sections 1–5 with audit-grade detail.
+
+
M1-S1 · Section 1 — Context of the Organization (Cl. 4)
+
+
deliverables Internal/external issues register (PEST + tech + regulatory) Interested parties matrix (regulators, customers, employees, society) AIMS scope statement (geographies, business units, AI systems) AI System Inventory v1 (1,200+ systems, classification) Boundary diagram showing AIMS interfaces with EMS/ISMS/QMS
+
evidenceRefs EVD-AIMS-S1-CTX-2026Q2 EVD-AIMS-S1-INV-2026Q2
+
+
+
M1-S2 · Section 2 — Leadership & Policy (Cl. 5)
+
+
deliverables Board-approved AI Policy (signed by Chair + CEO) AI Roles & Responsibilities matrix (RACI: Board, CAIO, CRO, CCO, DPO) Authority delegation: model approval thresholds by Tier T0–T5 Conflict-of-interest controls between 1st/2nd/3rd LoD
+
evidenceRefs EVD-AIMS-S2-POL-2026Q2 EVD-AIMS-S2-RACI-2026Q2
+
+
+
M1-S3 · Section 3 — Planning (Cl. 6)
+
+
deliverables AI Risks & Opportunities register (linked to ISO 23894 taxonomy) AI Objectives (16 KPIs, board-tracked) Change planning protocol (model promotion gates G0–G5) Statement of Applicability (SoA) covering Annex A + regulator overlays
+
evidenceRefs EVD-AIMS-S3-RISK-2026Q2 EVD-AIMS-S3-SOA-2026Q2
+
+
+
M1-S4 · Section 4 — Support (Cl. 7)
+
+
deliverables Resourcing plan (FTEs, GPU compute, evidence storage) Competence framework (CAIO certification, MRM accreditation) Awareness program (annual mandatory training, red-team exercises) Communication plan (internal + regulator + customer) Documented information control (versioning, WORM, retention)
+
evidenceRefs EVD-AIMS-S4-COMP-2026Q2 EVD-AIMS-S4-DOC-2026Q2
+
+
+
M1-S5 · Section 5 — Operation, Performance, Improvement (Cl. 8–10)
+
iso42001Clauses 8.1 8.2 8.3 9.1 9.2 9.3 10.1 10.2
+
deliverables Operational planning & control (life-cycle SOPs per ISO 5338) AI impact assessment process (GDPR DPIA + EU AI Act FRIA) Performance evaluation (KPI dashboard, internal audit plan) Management review minutes (quarterly, board-attested) Continual improvement loop (CAPA register, RCA)
+
evidenceRefs EVD-AIMS-S5-OPS-2026Q2 EVD-AIMS-S5-MR-2026Q2 EVD-AIMS-S5-CAPA-2026Q2
+
+
+
+M2 · M2 — AIMS Annexes J1–J4 (Implementation Detail)
+Four institution-specific annexes extending ISO/IEC 42001 Annex A/B with G-SIFI-grade depth.
+
+
M2-S1 · Annex J1 — AI System Inventory & Classification
+
content Authoritative register of all AI systems with EU AI Act tiering (Prohibited / High-Risk / Limited / Minimal), internal capability tier T0–T5, owning business unit, data classification, model risk tier, and impact zones.
+
fields systemId businessOwner euAiActTier internalTier modelRiskTier annexIIIRef lastFRIA lastDPIA rspVersion regulatorEngagementStatus
+
+
+
M2-S2 · Annex J2 — Statement of Applicability (SoA) + Control Mapping
+
content Mapping of ISO/IEC 42001 Annex A controls + 280 institution-specific controls to regulator overlays (ECB, Fed, PRA, EU AI Act, GDPR), each with a Terraform/OPA enforcement reference and an evidence automation status.
+
controlCategories AC — Accountability RM — Risk Management DG — Data Governance MD — Model Development VV — Validation & Verification DP — Deployment MO — Monitoring IR — Incident Response TP — Third-Party TR — Transparency
+
totalControls 280
+
+
+
M2-S3 · Annex J3 — AI Impact Assessment (FRIA + DPIA Combined)
+
content Unified template combining EU AI Act Fundamental Rights Impact Assessment (Art. 27) with GDPR DPIA (Art. 35) and SR 11-7 model materiality assessment.
+
phases Phase A — Purpose & Necessity Phase B — Risk Identification (12 axes) Phase C — Risk Evaluation (likelihood × severity × scope) Phase D — Mitigation Plan Phase E — Residual Risk Acceptance (CRO sign-off) Phase F — Monitoring & Review (auto-rerun on drift)
+
+
+
M2-S4 · Annex J4 — Regulator Submission Pack (RSP) Template
+
content Master template that produces RSP v1.0–v2.6 with decision-traceability links, model cards, eval results, monitoring telemetry, and signed attestations.
+
rspContents Cover & Executive Summary Model Card (Mitchell+ format extended) Data Sheet (Gebru+ format extended) FRIA + DPIA Validation Report (independent 2nd LoD sign-off) Monitoring Plan + KPI baseline Incident Response Plan (model-specific) Decision Traceability API endpoint + sample decisions Cryptographic attestation bundle (Sigstore + Rekor)
+
+
+
+M3 · M3 — Multi-Jurisdiction Regulatory Overlays
+Five regulator overlays applied as policy bundles on top of the ISO/IEC 42001 baseline.
+
+
M3-S1 · Overlay catalog
+
overlays id name scope keyRefs additionalControls OVL-ECB ECB SSM Overlay Significant Institutions under direct ECB supervision ECB Guide to Internal Models (2024) TRIM AI extensions ECB SSM Supervisory Priorities 2025-2027 ECB-AI-01 Model change notification within 5 business days ECB-AI-02 JST-accessible model inventory ECB-AI-03 ICAAP Pillar 2 AI capital add-on quantification OVL-FED Federal Reserve SR 11-7 Overlay US bank holding companies / FBOs SR 11-7 (2011) + 2021 supplemental guidance OCC 2011-12 FDIC FIL-22-2017 Joint statement on Risk-Based Approach to Third-Party Risk (2023) FED-AI-01 Independent model validation by qualified 2nd LoD FED-AI-02 Effective challenge documented for every Tier-1 model FED-AI-03 Ongoing monitoring with documented thresholds OVL-PRA PRA SS1/23 Overlay UK PRA-authorised firms PRA SS1/23 PRA SS2/21 outsourcing FCA Consumer Duty PRA-AI-01 Model risk tiering with board-approved thresholds PRA-AI-02 Senior Manager (SMF24) accountability for MRM PRA-AI-03 Annual model risk self-assessment to PRA OVL-EUAIA EU AI Act Overlay All AI systems placed on the EU market or affecting EU persons Reg. (EU) 2024/1689 EU AI Act Annex III §5(b) — credit scoring Commission implementing acts 2025-2026 EUAIA-AI-01 CE conformity (Art. 43) for high-risk systems EUAIA-AI-02 Post-market monitoring (Art. 72) live EUAIA-AI-03 Serious incident reporting within 15 days (Art. 73) EUAIA-AI-04 Registration in EU database (Art. 49) OVL-GDPR GDPR Overlay Any processing of EU personal data Reg. (EU) 2016/679 Articles 5/6/9/22/25/32/33/34/35 EDPB Guidelines 03/2022 on AI GDPR-AI-01 Art. 22 safeguards: human review path documented GDPR-AI-02 DPIA refreshed on material change GDPR-AI-03 Data minimisation tested via leakage probes
+
+
+
M3-S2 · Overlay precedence & conflict resolution
+
rules Strictest applicable provision wins (tier ordering). Where overlays diverge on disclosure scope, union of disclosures applies; classification follows the home regulator. Conflict log maintained with Legal sign-off for every override.
+
+
+
M3-S3 · Mapping matrix snapshot
+
matrix control ISO42001 ECB Fed PRA EUAIA GDPR Independent validation 8.3 ECB-AI-01/03 FED-AI-01/02 PRA-AI-02 Art. 17 QMS / 43 — Adverse-action explanation Annex A 6.2.7 — FCRA §615 FCA Consumer Duty Art. 13/86 Art. 22 Post-market monitoring 9.1 ECB-AI-02 FED-AI-03 PRA-AI-03 Art. 72 Art. 35(11) Incident reporting 10.2 Operational incident framework SR 11-7 weakness reporting SS1/23 §3.5 Art. 73 Art. 33/34
+
+
+
+M4 · M4 — Regulator Submission Packs (RSP v1.0 → v2.6)
+Versioned submission packs evolving from PDF-based static packs to fully machine-readable, signed, decision-traceable bundles.
+
+
M4-S1 · Version roadmap
+
versions id year format scope automation signing RSP-v1.0 2026 PDF + JSON manifest Single jurisdiction (home regulator) 30% PGP detached signature RSP-v1.5 2026 PDF + JSON-LD + Sigstore Home + 1 host regulator 55% Sigstore + Rekor transparency log RSP-v2.0 2027 Structured JSON-LD bundle (machine-readable) Multi-jurisdiction (ECB + PRA + Fed) 75% in-toto attestations RSP-v2.2 2027 JSON-LD + Decision-Traceability API Adds GDPR + EU AI Act DB linkage 85% in-toto + Cosign RSP-v2.4 2028 JSON-LD + live API + OPA-validated policy bundle All overlays, federated submission 92% PQC-ready (Dilithium hybrid) RSP-v2.5 2029 v2.4 + formally-verified obligation graph Adds machine-checkable legal logic 95% PQC + Merkle anchored to public ledger RSP-v2.6 2030 Continuous streaming attestation Autonomous-supervisor compatible 98% PQC + FROST threshold + ZK predicates
+
+
+
M4-S2 · RSP package structure (v2.4+)
+
structure /rsp/manifest.jsonld — top-level bundle /rsp/model-card.json /rsp/datasheet.json /rsp/fria-dpia.json /rsp/validation-report.json /rsp/monitoring-plan.json /rsp/incident-plan.json /rsp/decisions/ (signed decision envelopes) /rsp/policy-bundle.tar.gz (OPA bundle) /rsp/attestations/ (in-toto / Cosign / Rekor) /rsp/hash-chain.json (Merkle root + signatures)
+
+
+
M4-S3 · Decision-traceability API
+
endpoints GET /rsp/{rspId}/decisions/{decisionId} — full reproducible decision GET /rsp/{rspId}/decisions?subjectId=… — subject access GET /rsp/{rspId}/lineage — model + data lineage graph GET /rsp/{rspId}/attestations — verifiable bundle POST /rsp/{rspId}/challenge — supervisor counterfactual probe
+
slas decisionLookup <= 200 ms p95 lineageGraph <= 1 s p95 challengeReply <= 5 minutes p95
+
auth mTLS + supervisor SPIFFE ID + per-call OPA policy
+
+
+
M4-S4 · RSP issuance pipeline
+
stages Trigger: model promotion / quarterly cadence / supervisor request Assemble: pull artefacts from registry, evaluator, monitor Validate: OPA policy bundle compliance check Sign: in-toto layout + Cosign + Rekor entry Publish: regulator portal + internal evidence WORM Notify: supervisor + Internal Audit + Board pack
+
+
+
+M5 · M5 — Terraform + OPA Technical Enforcement
+Compliance-as-code substrate enforcing AIMS controls at infrastructure, pipeline, and runtime layers.
+
+
M5-S1 · Terraform modules
+
modules name purpose aims-baseline VPC/KMS/IAM/WORM-S3/Kafka baseline aims-evidence Object Lock + Lambda hash-chain anchor aims-runtime EKS/GKE clusters + admission controllers aims-supervisor Supervisor mTLS endpoints + SPIFFE aims-pqc PQC KMS keys + dual-signing CI
+
+
+
M5-S2 · OPA policy bundles
+
bundles policy/aims-baseline.tar.gz (Annex A controls) policy/overlay-ecb.tar.gz policy/overlay-fed.tar.gz policy/overlay-pra.tar.gz policy/overlay-euaia.tar.gz policy/overlay-gdpr.tar.gz policy/use-case-credit-underwriting.tar.gz
+
decisionPoints Terraform plan (pre-apply) — block insecure infra CI gate (pre-merge) — model card + eval coverage Admission controller (Kubernetes) — image attestation Inference gateway (runtime) — per-call obligations Egress filter — prohibited-use checks
+
+
+
M5-S3 · Continuous configuration audit
+
controls Daily Terraform drift scan with auto-remediation PR Hourly OPA bundle integrity check (signed digest) Per-region misconfiguration KPI dashboard Auto-quarantine of non-compliant workloads
+
+
+
+M6 · M6 — Adversarial & Self-Healing Governance Loops
+Continuous adversarial exercise of both models and controls, paired with auto-remediation that closes the loop without human intervention for known failure modes.
+
+
M6-S1 · Adversarial governance loop
+
stages Generate: red-team agents author attacks against models + controls Execute: attacks run in sandboxed twin environment Detect: monitors flag deltas vs. baseline behavior Triage: severity scored against impact taxonomy Remediate: control patch / model rollback / policy update Attest: signed evidence captured in WORM
+
cadence Continuous (on-demand + nightly + monthly chaos day)
+
+
+
M6-S2 · Self-healing playbooks
+
playbooks id trigger action humanGate SH-01 PSI > 0.2 on protected attribute Auto-rollback to previous model version + open Sev-2 ticket CRO post-hoc review within 24h SH-02 OPA policy bundle digest mismatch Quarantine workload + restore last-known-good bundle CISO + CCO joint review SH-03 Adverse-action SLA breach predicted Failover to deterministic fallback scoring + notify ops Head of Credit + DPO SH-04 FRIA risk score escalation Block new deployments of system + escalate to Risk Committee Board Risk Committee within 5 business days
+
+
+
M6-S3 · Adversarial assurance KPIs
+
kpis redTeamCoverage >= 95% of high-risk systems / quarter novelAttackDiscoveryRate >= 5 net-new attack classes / year selfHealingResolutionRate >= 80% Sev-2 without human action meanTimeToRemediate <= 30 min (Sev-2), <= 4 h (Sev-1)
+
+
+
+M7 · M7 — Predictive Governance & Formally-Verified Legal Logic
+Forecast control breaches before they occur and prove obligations are correctly implemented using machine-checkable specifications.
+
+
M7-S1 · Predictive governance
+
approach Treat governance KPIs (PSI, AIR, MTTR, evidence completeness) as time series; forecast breach probability and pre-emptively trigger remediation.
+
models Drift forecaster (Prophet + ARIMA ensemble) — 7-day horizon Fairness drift forecaster — protected-attribute aware Control-fatigue forecaster (audit findings as proxy) Regulatory-question forecaster (LLM-driven, supervised by Legal)
+
outputs Predicted breaches with calibrated confidence Recommended interventions (pre-staged remediation PRs) Board pre-warning dashboard (T-30 days)
+
+
+
M7-S2 · Formally-verified obligation graph
+
approach Encode regulator obligations as an obligation graph in TLA+/Lean and prove the implementation refines the specification.
+
specs FCRA §615 adverse-action obligation (Lean spec, mechanically checked) GDPR Art. 22 human-review-path obligation (TLA+) EU AI Act Art. 73 incident-reporting obligation (TLA+ liveness) ECB ICAAP Pillar 2 AI add-on quantification (Lean)
+
deliverable Each spec ships with a CI job that fails the build if a code change breaks refinement.
+
+
+
M7-S3 · Counterfactual + causal regulator queries
+
capability Supervisors can issue causal queries ("if income were +10%, would the decision flip?") that the system answers with a causal model + uncertainty, not just correlations.
+
engines DoWhy + EconML for causal effect estimation DiCE / Alibi for actionable counterfactuals LiNGAM / NOTEARS for structure discovery (governed)
+
+
+
+M8 · M8 — Cross-Regulator Federation & Autonomous Supervisory Ecosystem
+Federate disclosures across supervisors and prepare for autonomous supervisory ecosystems by 2030.
+
+
M8-S1 · Federation protocol (FedReg)
+
transport mTLS + SPIFFE IDs + OAuth2 Mutual-TLS Client Auth
+
schema JSON-LD with shared regulator vocabulary (W3C ODRL extension)
+
operations Disclose: scoped artefact share with consent metadata Subscribe: supervisor receives delta stream Challenge: supervisor issues counterfactual / explainability query Attest: institution returns signed answer with provenance
+
consentModel Per-scope, per-purpose, time-bounded, revocable
+
+
+
M8-S2 · Autonomous Supervisory Tiers
+
tiers tier name year description T0 Manual <2026 PDF + portal uploads T1 Structured 2026 Machine-readable RSP, manual review T2 Streaming 2027-2028 Continuous attestation feed T3 Federated 2028-2029 Cross-regulator query graph T4 Autonomous (advisory) 2029-2030 Supervisor AI agents issue advisories T5 Autonomous (binding-with-human-override) 2030+ Binding decisions with statutory human override
+
+
+
M8-S3 · Privacy & sovereignty controls in federation
+
controls Differential privacy on aggregate disclosures (ε <= 1) Zero-knowledge predicates for sensitive thresholds Data residency tags enforced at egress filter Per-jurisdiction key custody with HSM + threshold signing (FROST)
+
+
+
M8-S4 · Joint examination workflow
+
scenario ECB + FRB + PRA jointly examine AI-CR-UNDERWRITE-01. Each receives scoped, signed RSP slices; queries federated through FedReg; institution responses attested into a shared transparency log.
+
sla Joint final report within 30 calendar days
+
+
+
+M9 · M9 — High-Risk Credit Underwriting Best-Practice Pattern (AI-CR-UNDERWRITE-01)
+Reference end-to-end pattern for high-risk retail & SME credit underwriting under EU AI Act Annex III §5(b), FCRA, ECOA, and PRA / Fed MRM.
+
+
M9-S1 · Use-case scope & risk classification
+
details euAiActTier High-risk (Annex III §5(b)) internalTier T3 (material consumer impact) modelRiskTier Tier 1 regulators decisionVolume ~12M decisions / year
+
+
+
M9-S2 · Data governance
+
controls Datasheet (Gebru+) with provenance, sampling, bias notes Protected attributes proxied + monitored (no direct use) Synthetic counterfactual training augmentation for AIR uplift Quarterly representativeness audit by Internal Audit
+
+
+
M9-S3 · Model development & validation
+
controls Champion/challenger with at least 2 independent architectures GBM + monotonic constraints on protected proxies Independent 2nd LoD validation (effective challenge) FRIA + DPIA refreshed each retrain Reproducibility: bit-exact training pipeline pinned
+
+
+
M9-S4 · Decisioning & adverse action
+
controls Per-decision SHAP + counterfactual stored with envelope Adverse-action notice generated within 24h (FCRA §615) GDPR Art. 22 human-review path for any decision contested EU AI Act Art. 86 right to explanation served via portal Decision envelope signed (Ed25519 + PQC dual-sign)
+
+
+
M9-S5 · Monitoring & continuous compliance
+
controls Drift: PSI per feature + per protected attribute, daily Fairness: AIR + EOD + DI ratio, daily Stability: KS, ROC-AUC delta vs. baseline, weekly Calibration: Brier score, monthly Adversarial: prompt-injection / data-poisoning probes, nightly
+
+
+
M9-S6 · Regulator engagement
+
cadence Quarterly RSP v2.4 issuance to home + host regulators Material change notification within 5 business days (ECB-AI-01) Annual joint examination drill Live decision-traceability API for supervisor on-demand probes
+
+
+
+M10 · M10 — Implementation Roadmap (2026–2030)
+Five-phase, board-tracked program plan with gates and KPIs.
+
+
M10-S1 · Phase plan
+
phases id name window objectives exitGate P1 Foundation 2026 H1 Adopt ISO/IEC 42001 AIMS Sections 1–5 Stand up AI System Inventory (Annex J1) Issue RSP v1.0 for AI-CR-UNDERWRITE-01 Launch CAIO office with board mandate Board approval of AIMS + first RSP filed P2 Industrialise 2026 H2 – 2027 H1 Deploy Terraform + OPA enforcement substrate Roll out SoA (Annex J2) across 100% Tier-1 systems Issue RSP v1.5 + v2.0 Launch adversarial governance loop >= 75% control automation P3 Federate 2027 H2 – 2028 RSP v2.2 + v2.4 with multi-regulator scope FedReg federation pilot with ECB + PRA + Fed Activate self-healing playbooks SH-01..04 Stand up predictive governance forecasters Joint ECB+Fed+PRA examination drill passed P4 Verify 2029 Formally verified obligation graph live for top 5 obligations RSP v2.5 with machine-checkable legal logic Counterfactual / causal supervisor queries supported Autonomous supervisor T2->T3 Independent assurance from ISO 42001 certification body P5 Autonomous 2030 RSP v2.6 streaming attestation Autonomous supervisor T4 advisory mode active Cross-regulator binding-with-override pilot PQC + ZK predicates fully deployed Autonomous advisory disclosures accepted by 8+ supervisors
+
+
+
M10-S2 · KPI dashboard
+
kpis id name target K1 Time-to-regulator-approved deployment <= 14 days K2 RSP generation latency <= 30 minutes K3 Decision-traceability coverage >= 99.95% K4 Control automation rate >= 95% K5 Evidence automation >= 96% K6 Fairness AIR floor >= 0.85 K7 Explainability coverage (high-risk) 100% K8 Adverse-action SLA <= 24h auto K9 Regulator notification SLA <= 24h / 72h K10 Model inventory coverage 100% K11 Policy-drift MTTA <= 5 min K12 Self-healing resolution rate >= 80% Sev-2 K13 Audit finding closure >= 95% within SLA K14 Board attestation cadence Quarterly + ad-hoc K15 WORM retention 10 years K16 Federated supervisor count >= 8
+
+
+
M10-S3 · Top risks & mitigations
+
risks id risk mitigation R1 Regulatory divergence post-2027 Overlay precedence engine + Legal council monthly R2 Supervisor reluctance to accept machine-readable filings Dual format (PDF + JSON-LD) until T2 R3 Formal verification toolchain immaturity Hybrid test-based + spec-based assurance R4 PQC migration breakage Hybrid signing + staged rollouts R5 Self-healing causes incident drift Human gate on every Sev-1; quarterly chaos drills
+
+
+
+M11 · M11 — Governance Operating Model (3 LoD + RACI)
+Roles, accountabilities, and committee architecture.
+
+
M11-S1 · Three Lines of Defense
+
lod line owner responsibilities 1st LoD Business + AI engineering Build, operate, monitor models within risk appetite 2nd LoD MRM + Compliance + DPO + CISO Independent challenge, validation, policy, oversight 3rd LoD Internal Audit Audit AIMS effectiveness; audit the 2nd LoD
+
+
+
M11-S2 · RACI matrix (key activities)
+
matrix activity Board CEO CRO CCO CAIO DPO CISO InternalAudit Approve AI Policy A R C C C I Approve Tier-1 model I I A C R C Issue RSP I I A R R C Sev-1 incident response I I A C R C R Annual AIMS audit I I C C C C AR
+
+
+
M11-S3 · Committee architecture
+
committees id name frequency chair C1 Board AI Oversight Committee Quarterly Independent NED C2 Group AI Risk Committee Monthly CRO C3 Model Approval Committee Bi-weekly CAIO C4 AI Ethics Council Monthly GC + external ethicist C5 Regulator Engagement Forum Monthly CCO
+
+
+
+M12 · M12 — Reporting & Disclosure Templates
+Standardised, machine-readable templates for every audience.
+
+
M12-S1 · Audience matrix
+
matrix audience report format Board Quarterly AI Risk & KPI Pack PDF + JSON-LD Regulator (home) RSP v2.4+ JSON-LD bundle + signatures Regulator (host) Federated RSP slice FedReg streaming Customer (adverse action) Adverse-action notice + explanation Multilingual portal + paper Internal Audit AIMS audit dossier Evidence bundle + Merkle root Public Transparency report PDF + W3C transparency log link
+
+
+
M12-S2 · Markdown template skeleton
+
tags <title> <abstract> <content>
+
skeleton <title>Quarterly AI Risk & KPI Pack — 2026 Q4</title>
+<abstract>Summary of KPI movement, top risks, and regulator interactions for the quarter.</abstract>
+<content>1. KPI dashboard (K1..K16)
+2. Material model changes
+3. Incidents (Sev-0..Sev-2)
+4. Regulator engagements (RSP issuances, queries)
+5. Internal Audit findings status
+6. Forward-looking risks (predictive governance)
+7. Board decisions requested</content>
+
+
+
M12-S3 · Disclosure principles
+
principles Truthful, complete, and timely Audience-fit (no jargon to customers; rigour to supervisors) Verifiable (every claim traceable to a signed evidence record) Privacy-preserving (DP / ZK on aggregate disclosures)
+
+
+
+
+ Regulatory Alignment
+ ISO/IEC 42001:2023 — AI Management System (AIMS) — primary anchor ISO/IEC 23894:2023 — AI Risk Management ISO/IEC 5338:2023 — AI System Life Cycle Processes ISO/IEC 27001:2022 / 27701:2019 / 27018:2019 ISO/IEC TR 24028 / 24029 / 24368 (trustworthiness) EU AI Act (Reg. (EU) 2024/1689) — Art. 6, 9, 10, 12, 13, 14, 15, 17, 26, 27, 49, 53, 55, 72, 73; Annex III §5(b) GDPR (Reg. (EU) 2016/679) — Art. 5, 6, 9, 22, 25, 32, 33, 34, 35 ECB SSM Guide on internal models (2024) + Targeted Review of Internal Models (TRIM) AI extensions Federal Reserve SR 11-7 / OCC 2011-12 — Model Risk Management PRA SS1/23 — Model Risk Management Principles for Banks (UK) PRA SS2/21 — Outsourcing & third-party risk management FCA Consumer Duty (PS22/9) + AI/ML discussion paper DP5/22 Basel III/IV — CRR3 / CRD6 — ICAAP Pillar 2 AI add-on FCRA (US) §604/§615 + ECOA / Reg B §1002 (adverse action) CFPB Circular 2023-03 (algorithmic adverse-action notices) NIST AI RMF 1.0 + GenAI Profile (AI 600-1) OECD AI Principles + G7 Hiroshima AI Process Code of Conduct Council of Europe Framework Convention on AI (2024) OWASP LLM Top 10 (2025) / MITRE ATLAS SLSA L3 + Sigstore/Cosign + in-toto attestations
+
+
+
+ JSON Schemas
+ 8 schemas: AI System Inventory, RSP Manifest, Decision Envelope, Control Mapping, FRIA, Incident Record, FedReg Message, Obligation Spec.
+ aiSystemInventoryEntry {
+ "title": "AI System Inventory Entry (Annex J1)",
+ "required": [
+ "systemId",
+ "businessOwner",
+ "euAiActTier",
+ "internalTier",
+ "modelRiskTier",
+ "lastFRIA",
+ "rspVersion"
+ ],
+ "fields": {
+ "systemId": "string",
+ "businessOwner": "string",
+ "euAiActTier": "enum[Prohibited|HighRisk|Limited|Minimal]",
+ "internalTier": "enum[T0|T1|T2|T3|T4|T5]",
+ "modelRiskTier": "enum[Tier-1|Tier-2|Tier-3]",
+ "annexIIIRef": "string",
+ "lastFRIA": "ISO-8601",
+ "lastDPIA": "ISO-8601",
+ "rspVersion": "string",
+ "regulatorEngagementStatus": "enum[Filed|Pending|UnderReview|Approved|Withdrawn]"
+ }
+}rspManifest {
+ "title": "Regulator Submission Pack \u2014 Manifest (v2.4+)",
+ "required": [
+ "rspId",
+ "version",
+ "subjectSystemId",
+ "issuedAt",
+ "signatures",
+ "merkleRoot"
+ ],
+ "fields": {
+ "rspId": "string",
+ "version": "string",
+ "subjectSystemId": "string",
+ "issuedAt": "ISO-8601",
+ "regulators": "string[]",
+ "artefacts": "object[]",
+ "signatures": "object[]",
+ "merkleRoot": "hex",
+ "policyBundleDigest": "hex",
+ "ledgerAnchorTx": "string"
+ }
+}decisionEnvelope {
+ "title": "Decision Envelope (per AI decision)",
+ "required": [
+ "decisionId",
+ "subjectId",
+ "modelId",
+ "modelVersion",
+ "inputsHash",
+ "output",
+ "shapTopK",
+ "ts",
+ "signature"
+ ],
+ "fields": {
+ "decisionId": "string",
+ "subjectId": "string",
+ "modelId": "string",
+ "modelVersion": "string",
+ "inputsHash": "hex",
+ "output": "object",
+ "shapTopK": "object[]",
+ "counterfactual": "object",
+ "policyDecision": "object",
+ "ts": "ISO-8601",
+ "signature": "object"
+ }
+}controlMapping {
+ "title": "Control Mapping (Annex J2 SoA)",
+ "required": [
+ "controlId",
+ "category",
+ "iso42001Ref",
+ "overlays",
+ "enforcement"
+ ],
+ "fields": {
+ "controlId": "string",
+ "category": "string",
+ "iso42001Ref": "string",
+ "overlays": "object",
+ "enforcement": "object",
+ "evidenceAutomation": "enum[None|Partial|Full]",
+ "owner": "string"
+ }
+}friaRecord {
+ "title": "FRIA + DPIA Combined Record (Annex J3)",
+ "required": [
+ "friaId",
+ "subjectSystemId",
+ "phase",
+ "residualRisk",
+ "approvers"
+ ],
+ "fields": {
+ "friaId": "string",
+ "subjectSystemId": "string",
+ "phase": "enum[A|B|C|D|E|F]",
+ "axes": "object[]",
+ "residualRisk": "enum[Low|Medium|High|Critical]",
+ "approvers": "string[]",
+ "nextReviewAt": "ISO-8601"
+ }
+}incidentRecord {
+ "title": "AI Incident Record (Cl. 10.2 + EU AI Act Art. 73)",
+ "required": [
+ "incidentId",
+ "severity",
+ "detectedAt",
+ "affectedSystems",
+ "narrative"
+ ],
+ "fields": {
+ "incidentId": "string",
+ "severity": "enum[Sev-0|Sev-1|Sev-2|Sev-3]",
+ "detectedAt": "ISO-8601",
+ "affectedSystems": "string[]",
+ "regulatorNotifications": "object[]",
+ "narrative": "string",
+ "rootCause": "string",
+ "capa": "object[]"
+ }
+}fedRegMessage {
+ "title": "Federation Protocol Message (FedReg)",
+ "required": [
+ "messageId",
+ "fromSpiffeId",
+ "toSpiffeId",
+ "op",
+ "payloadRef",
+ "consentScope"
+ ],
+ "fields": {
+ "messageId": "string",
+ "fromSpiffeId": "string",
+ "toSpiffeId": "string",
+ "op": "enum[Disclose|Subscribe|Challenge|Attest]",
+ "payloadRef": "string",
+ "consentScope": "object",
+ "signatures": "object[]",
+ "ts": "ISO-8601"
+ }
+}obligationSpec {
+ "title": "Formally-Verified Obligation Spec",
+ "required": [
+ "obligationId",
+ "regulatorRef",
+ "specLanguage",
+ "specHash",
+ "refinementProof"
+ ],
+ "fields": {
+ "obligationId": "string",
+ "regulatorRef": "string",
+ "specLanguage": "enum[TLA+|Lean|Coq]",
+ "specHash": "hex",
+ "refinementProof": "string",
+ "ciJobRef": "string"
+ }
+}
+
+
+
+ Code Examples
+ 11 reference implementations: OPA RSP gate, Terraform WORM evidence, decision envelope signer (Ed25519 + PQC), fairness monitor, FedReg client, predictive drift forecaster, TLA+ obligation spec, Lean FCRA spec, self-healing playbook engine, FastAPI traceability API, Merkle anchor.
+ opaRspGate rego · Block RSP issuance unless all required artefacts + signatures present
package rsp.gate
+
+default allow = false
+
+required := {"manifest", "model-card", "datasheet", "fria-dpia",
+ "validation-report", "monitoring-plan", "incident-plan",
+ "policy-bundle", "attestations", "hash-chain"}
+
+allow {
+ have := {a | a := input.artefacts[_].name}
+ missing := required - have
+ count(missing) == 0
+ input.signatures.cosign.verified == true
+ input.signatures.intoto.verified == true
+ input.policyBundleDigest == data.policy.expectedDigest
+}
+terraformWormEvidence hcl · S3 Object Lock + KMS WORM evidence bucket (10-year retention)
resource "aws_s3_bucket" "aims_evidence" {
+ bucket = "gsifi-aims-evidence-${var.env}"
+ object_lock_enabled = true
+}
+
+resource "aws_s3_bucket_object_lock_configuration" "lock" {
+ bucket = aws_s3_bucket.aims_evidence.id
+ rule {
+ default_retention {
+ mode = "COMPLIANCE"
+ years = 10
+ }
+ }
+}
+
+resource "aws_s3_bucket_server_side_encryption_configuration" "sse" {
+ bucket = aws_s3_bucket.aims_evidence.id
+ rule {
+ apply_server_side_encryption_by_default {
+ kms_master_key_id = aws_kms_key.aims.arn
+ sse_algorithm = "aws:kms"
+ }
+ }
+}
+decisionEnvelopeSigner python · Sign per-decision envelopes (Ed25519 + PQC dual-sign)
import hashlib, json, time
+from cryptography.hazmat.primitives.asymmetric.ed25519 import Ed25519PrivateKey
+# pqcrypto.sign.dilithium3 illustrative
+from pqcrypto.sign.dilithium3 import generate_keypair, sign as pqc_sign
+
+def make_envelope(decision_id, subject_id, model_id, model_version,
+ inputs, output, shap_topk, ed_sk, pqc_sk):
+ inputs_hash = hashlib.sha256(json.dumps(inputs, sort_keys=True).encode()).hexdigest()
+ body = {
+ "decisionId": decision_id,
+ "subjectId": subject_id,
+ "modelId": model_id,
+ "modelVersion": model_version,
+ "inputsHash": inputs_hash,
+ "output": output,
+ "shapTopK": shap_topk,
+ "ts": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+ }
+ payload = json.dumps(body, sort_keys=True).encode()
+ sig_ed = ed_sk.sign(payload).hex()
+ sig_pqc = pqc_sign(pqc_sk, payload).hex()
+ body["signature"] = {"ed25519": sig_ed, "dilithium3": sig_pqc}
+ return body
+fairnessMonitor python · Daily AIR / EOD monitor with self-healing trigger (SH-01)
import numpy as np
+
+def adverse_impact_ratio(y_pred, protected):
+ rates = {g: y_pred[protected == g].mean() for g in np.unique(protected)}
+ ref = max(rates.values())
+ return min(rates.values()) / ref if ref else 1.0
+
+def monitor(daily_predictions, protected, prev_air, prev_psi):
+ air = adverse_impact_ratio(daily_predictions, protected)
+ if air < 0.85 or prev_psi > 0.2:
+ trigger_self_heal("SH-01", reason={"air": air, "psi": prev_psi})
+ return {"air": air}
+
+def trigger_self_heal(playbook_id, reason):
+ # POST signed event to governance bus → triggers rollback + Sev-2 ticket
+ ...
+fedRegClient python · FedReg federation client — disclose RSP slice to supervisor
import requests, json, time
+
+def disclose(supervisor_url, rsp_slice, scope, spiffe_ctx, signer):
+ msg = {
+ "messageId": f"msg-{int(time.time()*1000)}",
+ "fromSpiffeId": spiffe_ctx.self_id,
+ "toSpiffeId": spiffe_ctx.peer_id,
+ "op": "Disclose",
+ "payloadRef": rsp_slice["uri"],
+ "consentScope": scope,
+ "ts": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+ }
+ body = json.dumps(msg, sort_keys=True).encode()
+ msg["signatures"] = signer(body)
+ return requests.post(supervisor_url + "/fedreg/v1/messages",
+ json=msg, cert=spiffe_ctx.mtls_cert).json()
+predictiveDriftForecaster python · Forecast PSI breach 7 days ahead (predictive governance)
from prophet import Prophet
+import pandas as pd
+
+def forecast_psi_breach(history_df, threshold=0.2, horizon=7):
+ m = Prophet(interval_width=0.95).fit(history_df.rename(columns={"date":"ds","psi":"y"}))
+ future = m.make_future_dataframe(periods=horizon)
+ fcst = m.predict(future)
+ breach = fcst[fcst["yhat"] > threshold].head(1)
+ return None if breach.empty else {
+ "predictedBreachAt": str(breach.iloc[0]["ds"].date()),
+ "expectedPsi": float(breach.iloc[0]["yhat"]),
+ }
+tlaPlusObligation tla · TLA+ liveness spec for EU AI Act Art. 73 incident reporting
-------------- MODULE Art73Reporting --------------
+EXTENDS Naturals, TLC
+VARIABLES status, notifiedAt, detectedAt
+Init == /\ status = "open" /\ notifiedAt = 0 /\ detectedAt = 0
+Report == /\ status = "open" /\ status' = "reported"
+ /\ notifiedAt' = detectedAt + 15
+Liveness == <>(status = "reported" /\ notifiedAt - detectedAt <= 15)
+Spec == Init /\ [][Report]_<<status, notifiedAt, detectedAt>> /\ Liveness
+====
+leanFcraSpec lean · Lean spec for FCRA §615 adverse-action obligation
import data.real.basic
+
+structure Decision := (subject : string) (denied : bool) (timestamp_h : nat)
+structure Notice := (subject : string) (sent_at_h : nat) (reasons : list string)
+
+def fcra_compliant (d : Decision) (n : Notice) : Prop :=
+ d.subject = n.subject
+ ∧ d.denied = tt
+ ∧ n.sent_at_h ≤ d.timestamp_h + 30 * 24
+ ∧ n.reasons.length ≥ 1
+
+theorem fcra_demo :
+ ∀ d n, d.denied = tt → fcra_compliant d n → n.reasons.length ≥ 1 :=
+λ d n h1 hc, hc.2.2.2
+selfHealingPlaybookEngine python · Self-healing playbook executor with WORM-attested actions
import json, time, hashlib
+
+def execute_playbook(playbook, signals, signer, worm_writer):
+ record = {"playbook": playbook["id"], "trigger": playbook["trigger"],
+ "signals": signals, "ts": time.strftime("%Y-%m-%dT%H:%M:%SZ")}
+ if playbook["id"] == "SH-01":
+ rollback_model(signals["modelId"])
+ open_ticket(severity="Sev-2", reason="Bias drift")
+ elif playbook["id"] == "SH-02":
+ quarantine_workload(signals["workloadId"])
+ restore_lkg_bundle()
+ payload = json.dumps(record, sort_keys=True).encode()
+ record["digest"] = hashlib.sha256(payload).hexdigest()
+ record["signature"] = signer(payload)
+ worm_writer.write(record)
+ return record
+
+def rollback_model(*a, **k): ...
+def open_ticket(*a, **k): ...
+def quarantine_workload(*a, **k): ...
+def restore_lkg_bundle(*a, **k): ...
+rspApiFastapi python · FastAPI decision-traceability API for RSP v2.4+
from fastapi import FastAPI, HTTPException, Depends
+app = FastAPI(title="RSP Decision Traceability API")
+
+def auth(spiffe_id: str = ""):
+ if not spiffe_id.startswith("spiffe://supervisor."):
+ raise HTTPException(401, "Supervisor SPIFFE required")
+ return spiffe_id
+
+@app.get("/rsp/{rsp_id}/decisions/{decision_id}")
+def get_decision(rsp_id: str, decision_id: str, who=Depends(auth)):
+ env = decision_store.fetch(rsp_id, decision_id)
+ if not env: raise HTTPException(404, "Decision not found")
+ return env
+
+@app.post("/rsp/{rsp_id}/challenge")
+def challenge(rsp_id: str, body: dict, who=Depends(auth)):
+ return counterfactual_engine.run(rsp_id, body)
+merkleAnchor python · Daily Merkle anchor of evidence WORM into public ledger
import hashlib
+
+def merkle_root(leaves):
+ layer = [bytes.fromhex(l) for l in leaves]
+ while len(layer) > 1:
+ if len(layer) % 2: layer.append(layer[-1])
+ layer = [hashlib.sha256(layer[i]+layer[i+1]).digest() for i in range(0,len(layer),2)]
+ return layer[0].hex()
+
+def anchor_today(evidence_hashes, ledger_client):
+ root = merkle_root(evidence_hashes)
+ txid = ledger_client.publish(root)
+ return {"root": root, "txid": txid, "count": len(evidence_hashes)}
+
+
+
+
+ Case Studies
+ 5 reference deployments: EU G-SIB dual ISO/EU AI Act certification, US BHC federated SR 11-7+EU AI Act, UK PRA SS1/23 SMF24 attestation, joint ECB+Fed+PRA examination, self-healing bias drift auto-rollback.
+ CS-01 · European G-SIB — first ISO/IEC 42001 + EU AI Act dual certification Sector: Banking (EU)
Top-3 EU bank achieved ISO/IEC 42001 certification and EU AI Act Art. 43 conformity for AI-CR-UNDERWRITE-01 concurrently.
Outcomes rspVersion v2.4 regulators controlAutomation 94% auditFindingsCriticalHigh 0
CS-02 · US BHC — federated SR 11-7 + EU AI Act submission Sector: Banking (US/EU)
US bank holding company served SR 11-7 + EU AI Act overlays from a single AIMS, federated to FRB + ECB via FedReg.
Outcomes rspVersion v2.2 → v2.4 supervisorCount 5 decisionTraceability 99.97% boardAttestation Quarterly + ad-hoc
CS-03 · UK firm — PRA SS1/23 SMF24 attestation pipeline Sector: Banking (UK)
PRA-authorised firm built an SMF24 senior-manager attestation pipeline auto-generated from AIMS evidence.
Outcomes smf24AttestationLatency <= 24h evidenceAutomation 97% annualSelfAssessment Filed 11 days early
CS-04 · Joint examination drill — ECB + Fed + PRA Sector: Cross-jurisdiction
Three home/host supervisors ran a joint examination of AI-CR-UNDERWRITE-01 using FedReg, with binding-with-override advisory issued by an autonomous supervisor agent (T4).
Outcomes totalQueries 412 averageReplyLatency 27 minutes challengePassRate 98.5% finalReportTime 23 days
CS-05 · Self-healing in production — bias drift auto-rollback Sector: Banking
AIR fell to 0.81 on a protected attribute; SH-01 auto-rolled back the model within 4 minutes, opened Sev-2, and filed a customer-impact pre-warning to Internal Audit.
Outcomes detectionToRollback 4 min customerImpact 0 wrongful denials regulatorNotified ECB + ICO within 6h rcaPublished <= 5 business days
+
+
+
+ API Endpoints
+ Prefix: /api/gsifi-aims · Total planned: 78
+ /api/gsifi-aims/api/gsifi-aims/meta/api/gsifi-aims/executive-summary/api/gsifi-aims/summary/api/gsifi-aims/aims/api/gsifi-aims/aims/sections/api/gsifi-aims/aims/sections/:id/api/gsifi-aims/aims/annexes/api/gsifi-aims/aims/annexes/:id/api/gsifi-aims/regulatory/api/gsifi-aims/regulatory/overlays/api/gsifi-aims/regulatory/overlays/:id/api/gsifi-aims/regulatory/precedence/api/gsifi-aims/regulatory/matrix/api/gsifi-aims/rsp/api/gsifi-aims/rsp/versions/api/gsifi-aims/rsp/versions/:id/api/gsifi-aims/rsp/structure/api/gsifi-aims/rsp/api/api/gsifi-aims/rsp/pipeline/api/gsifi-aims/enforcement/api/gsifi-aims/enforcement/terraform/api/gsifi-aims/enforcement/opa/api/gsifi-aims/enforcement/audit/api/gsifi-aims/adversarial/api/gsifi-aims/adversarial/loop/api/gsifi-aims/adversarial/playbooks/api/gsifi-aims/adversarial/kpis/api/gsifi-aims/predictive/api/gsifi-aims/predictive/forecasters/api/gsifi-aims/predictive/formal/api/gsifi-aims/predictive/causal/api/gsifi-aims/federation/api/gsifi-aims/federation/protocol/api/gsifi-aims/federation/tiers/api/gsifi-aims/federation/privacy/api/gsifi-aims/federation/joint-exam/api/gsifi-aims/credit-underwriting/api/gsifi-aims/credit-underwriting/scope/api/gsifi-aims/credit-underwriting/data/api/gsifi-aims/credit-underwriting/dev-validation/api/gsifi-aims/credit-underwriting/decisioning/api/gsifi-aims/credit-underwriting/monitoring/api/gsifi-aims/credit-underwriting/regulator/api/gsifi-aims/roadmap/api/gsifi-aims/roadmap/phases/api/gsifi-aims/roadmap/phases/:id/api/gsifi-aims/roadmap/kpis/api/gsifi-aims/roadmap/risks/api/gsifi-aims/operating-model/api/gsifi-aims/operating-model/lod/api/gsifi-aims/operating-model/raci/api/gsifi-aims/operating-model/committees/api/gsifi-aims/reporting/api/gsifi-aims/reporting/audience/api/gsifi-aims/reporting/template/api/gsifi-aims/reporting/principles/api/gsifi-aims/schemas/api/gsifi-aims/schemas/:name/api/gsifi-aims/code-examples/api/gsifi-aims/code-examples/:name/api/gsifi-aims/case-studies/api/gsifi-aims/case-studies/:id/api/gsifi-aims/modules/api/gsifi-aims/modules/:id/api/gsifi-aims/sections/:id/api/gsifi-aims/m1/api/gsifi-aims/m2/api/gsifi-aims/m3/api/gsifi-aims/m4/api/gsifi-aims/m5/api/gsifi-aims/m6/api/gsifi-aims/m7/api/gsifi-aims/m8/api/gsifi-aims/m9/api/gsifi-aims/m10/api/gsifi-aims/m11/api/gsifi-aims/m12
+
+
+
+ © GSIFI-AIMS-BLUEPRINT-WP-037 v1.0.0 ·
+ 2026-04-30 · CONFIDENTIAL — Board / Prudential Regulator / Group Risk / Internal Audit / Chief Legal & Compliance Officer ·
+ Owner: Group CRO + Chief AI Officer (CAIO) — co-signed by CCO, GC, CISO, DPO, Head of Internal Audit
+
+
+
diff --git a/rag-agentic-dashboard/server.js b/rag-agentic-dashboard/server.js
index 86f82a3..b51897c 100644
--- a/rag-agentic-dashboard/server.js
+++ b/rag-agentic-dashboard/server.js
@@ -21739,6 +21739,236 @@ app.get('/api/wfap-gemini/case-studies/:id', (req, res) => {
res.json(cs);
});
+// ══════════════════════════════════════════════════════════════════════════════
+// GSIFI-AIMS-BLUEPRINT-WP-037 — Regulator-Grade AI Governance & ISO/IEC 42001
+// AIMS Master Blueprint for G-SIFIs (2026–2030)
+// ══════════════════════════════════════════════════════════════════════════════
+const GSAIMS = require('./data/gsifi-aims-blueprint.json');
+
+const GSAIMS_MODULES = {
+ M1: GSAIMS.M1_aimsSections,
+ M2: GSAIMS.M2_aimsAnnexes,
+ M3: GSAIMS.M3_regulatoryOverlays,
+ M4: GSAIMS.M4_rsp,
+ M5: GSAIMS.M5_technicalEnforcement,
+ M6: GSAIMS.M6_adversarialSelfHealing,
+ M7: GSAIMS.M7_predictiveFormal,
+ M8: GSAIMS.M8_federationSupervisory,
+ M9: GSAIMS.M9_creditUnderwriting,
+ M10: GSAIMS.M10_roadmap,
+ M11: GSAIMS.M11_operatingModel,
+ M12: GSAIMS.M12_reportingDisclosure,
+};
+
+function gsaimsSection(modKey, sid) {
+ const mod = GSAIMS[modKey] || {};
+ return ((mod.sections) || []).find(s => (s.id || '').toUpperCase() === sid.toUpperCase()) || {};
+}
+
+app.get('/api/gsifi-aims', (_, res) => res.json(GSAIMS));
+app.get('/api/gsifi-aims/meta', (_, res) => res.json(GSAIMS.meta || {}));
+app.get('/api/gsifi-aims/executive-summary',(_, res) => res.json(GSAIMS.executiveSummary || {}));
+app.get('/api/gsifi-aims/summary', (_, res) => {
+ const m = GSAIMS.meta || {};
+ const inv = m.deliverableInventory || {};
+ res.json({
+ docRef: m.docRef,
+ version: m.version,
+ title: m.title,
+ horizon: m.horizon,
+ classification: m.classification,
+ modules: Object.keys(GSAIMS_MODULES).length,
+ aimsSections: inv.aimsSections || 5,
+ annexes: inv.annexes || 4,
+ regulatoryOverlays: inv.regulatoryOverlays || 5,
+ rspVersions: inv.rspVersions || 7,
+ schemas: Object.keys(GSAIMS.schemas || {}).length,
+ codeExamples: Object.keys(GSAIMS.codeExamples || {}).length,
+ caseStudies: (GSAIMS.caseStudies || []).length,
+ phases: inv.phases || 5,
+ kpis: inv.kpis || 16,
+ controls: inv.controls || 280,
+ apiPrefix: '/api/gsifi-aims',
+ routes: ((GSAIMS.apiEndpoints || {}).routes || []).length,
+ });
+});
+
+app.get('/api/gsifi-aims/modules', (_, res) => {
+ res.json(Object.entries(GSAIMS_MODULES).map(([k, v]) => ({
+ key: k, id: (v && v.id) || k, title: (v && v.title) || '',
+ sections: ((v && v.sections) || []).length,
+ })));
+});
+app.get('/api/gsifi-aims/modules/:id', (req, res) => {
+ const id = req.params.id.toUpperCase();
+ const mod = GSAIMS_MODULES[id];
+ if (!mod) return res.status(404).json({ error: 'module not found', id: req.params.id });
+ res.json(mod);
+});
+
+// Module shortcuts m1..m12
+app.get('/api/gsifi-aims/m1', (_, res) => res.json(GSAIMS.M1_aimsSections || {}));
+app.get('/api/gsifi-aims/m2', (_, res) => res.json(GSAIMS.M2_aimsAnnexes || {}));
+app.get('/api/gsifi-aims/m3', (_, res) => res.json(GSAIMS.M3_regulatoryOverlays || {}));
+app.get('/api/gsifi-aims/m4', (_, res) => res.json(GSAIMS.M4_rsp || {}));
+app.get('/api/gsifi-aims/m5', (_, res) => res.json(GSAIMS.M5_technicalEnforcement || {}));
+app.get('/api/gsifi-aims/m6', (_, res) => res.json(GSAIMS.M6_adversarialSelfHealing || {}));
+app.get('/api/gsifi-aims/m7', (_, res) => res.json(GSAIMS.M7_predictiveFormal || {}));
+app.get('/api/gsifi-aims/m8', (_, res) => res.json(GSAIMS.M8_federationSupervisory || {}));
+app.get('/api/gsifi-aims/m9', (_, res) => res.json(GSAIMS.M9_creditUnderwriting || {}));
+app.get('/api/gsifi-aims/m10', (_, res) => res.json(GSAIMS.M10_roadmap || {}));
+app.get('/api/gsifi-aims/m11', (_, res) => res.json(GSAIMS.M11_operatingModel || {}));
+app.get('/api/gsifi-aims/m12', (_, res) => res.json(GSAIMS.M12_reportingDisclosure || {}));
+
+// AIMS sections / annexes (M1, M2)
+app.get('/api/gsifi-aims/aims', (_, res) => res.json(GSAIMS.M1_aimsSections || {}));
+app.get('/api/gsifi-aims/aims/sections', (_, res) => res.json((GSAIMS.M1_aimsSections || {}).sections || []));
+app.get('/api/gsifi-aims/aims/sections/:id', (req, res) => {
+ const id = req.params.id.toUpperCase();
+ const s = ((GSAIMS.M1_aimsSections || {}).sections || []).find(x => (x.id || '').toUpperCase() === id);
+ if (!s) return res.status(404).json({ error: 'AIMS section not found', id: req.params.id });
+ res.json(s);
+});
+app.get('/api/gsifi-aims/aims/annexes', (_, res) => res.json((GSAIMS.M2_aimsAnnexes || {}).sections || []));
+app.get('/api/gsifi-aims/aims/annexes/:id', (req, res) => {
+ const id = req.params.id.toUpperCase();
+ const s = ((GSAIMS.M2_aimsAnnexes || {}).sections || []).find(x => (x.id || '').toUpperCase() === id);
+ if (!s) return res.status(404).json({ error: 'AIMS annex not found', id: req.params.id });
+ res.json(s);
+});
+
+// Regulatory overlays (M3)
+app.get('/api/gsifi-aims/regulatory', (_, res) => res.json(GSAIMS.M3_regulatoryOverlays || {}));
+app.get('/api/gsifi-aims/regulatory/overlays', (_, res) => {
+ const sec = gsaimsSection('M3_regulatoryOverlays', 'M3-S1');
+ res.json(sec.overlays || []);
+});
+app.get('/api/gsifi-aims/regulatory/overlays/:id', (req, res) => {
+ const id = req.params.id.toUpperCase();
+ const sec = gsaimsSection('M3_regulatoryOverlays', 'M3-S1');
+ const o = (sec.overlays || []).find(x => (x.id || '').toUpperCase() === id);
+ if (!o) return res.status(404).json({ error: 'overlay not found', id: req.params.id });
+ res.json(o);
+});
+app.get('/api/gsifi-aims/regulatory/precedence',(_, res) => res.json(gsaimsSection('M3_regulatoryOverlays', 'M3-S2')));
+app.get('/api/gsifi-aims/regulatory/matrix', (_, res) => res.json(gsaimsSection('M3_regulatoryOverlays', 'M3-S3')));
+
+// Regulator Submission Packs (M4)
+app.get('/api/gsifi-aims/rsp', (_, res) => res.json(GSAIMS.M4_rsp || {}));
+app.get('/api/gsifi-aims/rsp/versions', (_, res) => {
+ const sec = gsaimsSection('M4_rsp', 'M4-S1');
+ res.json(sec.versions || []);
+});
+app.get('/api/gsifi-aims/rsp/versions/:id', (req, res) => {
+ const id = req.params.id.toUpperCase();
+ const sec = gsaimsSection('M4_rsp', 'M4-S1');
+ const v = (sec.versions || []).find(x => (x.id || '').toUpperCase() === id);
+ if (!v) return res.status(404).json({ error: 'RSP version not found', id: req.params.id });
+ res.json(v);
+});
+app.get('/api/gsifi-aims/rsp/structure', (_, res) => res.json(gsaimsSection('M4_rsp', 'M4-S2')));
+app.get('/api/gsifi-aims/rsp/api', (_, res) => res.json(gsaimsSection('M4_rsp', 'M4-S3')));
+app.get('/api/gsifi-aims/rsp/pipeline', (_, res) => res.json(gsaimsSection('M4_rsp', 'M4-S4')));
+
+// Technical enforcement (M5)
+app.get('/api/gsifi-aims/enforcement', (_, res) => res.json(GSAIMS.M5_technicalEnforcement || {}));
+app.get('/api/gsifi-aims/enforcement/terraform', (_, res) => res.json(gsaimsSection('M5_technicalEnforcement', 'M5-S1')));
+app.get('/api/gsifi-aims/enforcement/opa', (_, res) => res.json(gsaimsSection('M5_technicalEnforcement', 'M5-S2')));
+app.get('/api/gsifi-aims/enforcement/audit', (_, res) => res.json(gsaimsSection('M5_technicalEnforcement', 'M5-S3')));
+
+// Adversarial / self-healing (M6)
+app.get('/api/gsifi-aims/adversarial', (_, res) => res.json(GSAIMS.M6_adversarialSelfHealing || {}));
+app.get('/api/gsifi-aims/adversarial/loop', (_, res) => res.json(gsaimsSection('M6_adversarialSelfHealing', 'M6-S1')));
+app.get('/api/gsifi-aims/adversarial/playbooks', (_, res) => res.json(gsaimsSection('M6_adversarialSelfHealing', 'M6-S2')));
+app.get('/api/gsifi-aims/adversarial/kpis', (_, res) => res.json(gsaimsSection('M6_adversarialSelfHealing', 'M6-S3')));
+
+// Predictive / formal verification (M7)
+app.get('/api/gsifi-aims/predictive', (_, res) => res.json(GSAIMS.M7_predictiveFormal || {}));
+app.get('/api/gsifi-aims/predictive/forecasters', (_, res) => res.json(gsaimsSection('M7_predictiveFormal', 'M7-S1')));
+app.get('/api/gsifi-aims/predictive/formal', (_, res) => res.json(gsaimsSection('M7_predictiveFormal', 'M7-S2')));
+app.get('/api/gsifi-aims/predictive/causal', (_, res) => res.json(gsaimsSection('M7_predictiveFormal', 'M7-S3')));
+
+// Federation / autonomous supervisory (M8)
+app.get('/api/gsifi-aims/federation', (_, res) => res.json(GSAIMS.M8_federationSupervisory || {}));
+app.get('/api/gsifi-aims/federation/protocol', (_, res) => res.json(gsaimsSection('M8_federationSupervisory', 'M8-S1')));
+app.get('/api/gsifi-aims/federation/tiers', (_, res) => res.json(gsaimsSection('M8_federationSupervisory', 'M8-S2')));
+app.get('/api/gsifi-aims/federation/privacy', (_, res) => res.json(gsaimsSection('M8_federationSupervisory', 'M8-S3')));
+app.get('/api/gsifi-aims/federation/joint-exam', (_, res) => res.json(gsaimsSection('M8_federationSupervisory', 'M8-S4')));
+
+// Credit underwriting use case (M9)
+app.get('/api/gsifi-aims/credit-underwriting', (_, res) => res.json(GSAIMS.M9_creditUnderwriting || {}));
+app.get('/api/gsifi-aims/credit-underwriting/scope', (_, res) => res.json(gsaimsSection('M9_creditUnderwriting', 'M9-S1')));
+app.get('/api/gsifi-aims/credit-underwriting/data', (_, res) => res.json(gsaimsSection('M9_creditUnderwriting', 'M9-S2')));
+app.get('/api/gsifi-aims/credit-underwriting/dev-validation', (_, res) => res.json(gsaimsSection('M9_creditUnderwriting', 'M9-S3')));
+app.get('/api/gsifi-aims/credit-underwriting/decisioning', (_, res) => res.json(gsaimsSection('M9_creditUnderwriting', 'M9-S4')));
+app.get('/api/gsifi-aims/credit-underwriting/monitoring', (_, res) => res.json(gsaimsSection('M9_creditUnderwriting', 'M9-S5')));
+app.get('/api/gsifi-aims/credit-underwriting/regulator', (_, res) => res.json(gsaimsSection('M9_creditUnderwriting', 'M9-S6')));
+
+// Roadmap (M10)
+app.get('/api/gsifi-aims/roadmap', (_, res) => res.json(GSAIMS.M10_roadmap || {}));
+app.get('/api/gsifi-aims/roadmap/phases', (_, res) => {
+ const sec = gsaimsSection('M10_roadmap', 'M10-S1');
+ res.json(sec.phases || []);
+});
+app.get('/api/gsifi-aims/roadmap/phases/:id', (req, res) => {
+ const id = req.params.id.toUpperCase();
+ const sec = gsaimsSection('M10_roadmap', 'M10-S1');
+ const p = (sec.phases || []).find(x => (x.id || '').toUpperCase() === id);
+ if (!p) return res.status(404).json({ error: 'phase not found', id: req.params.id });
+ res.json(p);
+});
+app.get('/api/gsifi-aims/roadmap/kpis', (_, res) => {
+ const sec = gsaimsSection('M10_roadmap', 'M10-S2');
+ res.json(sec.kpis || []);
+});
+app.get('/api/gsifi-aims/roadmap/risks', (_, res) => {
+ const sec = gsaimsSection('M10_roadmap', 'M10-S3');
+ res.json(sec.risks || []);
+});
+
+// Operating model (M11)
+app.get('/api/gsifi-aims/operating-model', (_, res) => res.json(GSAIMS.M11_operatingModel || {}));
+app.get('/api/gsifi-aims/operating-model/lod', (_, res) => res.json(gsaimsSection('M11_operatingModel', 'M11-S1')));
+app.get('/api/gsifi-aims/operating-model/raci', (_, res) => res.json(gsaimsSection('M11_operatingModel', 'M11-S2')));
+app.get('/api/gsifi-aims/operating-model/committees', (_, res) => res.json(gsaimsSection('M11_operatingModel', 'M11-S3')));
+
+// Reporting & disclosure (M12)
+app.get('/api/gsifi-aims/reporting', (_, res) => res.json(GSAIMS.M12_reportingDisclosure || {}));
+app.get('/api/gsifi-aims/reporting/audience', (_, res) => res.json(gsaimsSection('M12_reportingDisclosure', 'M12-S1')));
+app.get('/api/gsifi-aims/reporting/template', (_, res) => res.json(gsaimsSection('M12_reportingDisclosure', 'M12-S2')));
+app.get('/api/gsifi-aims/reporting/principles', (_, res) => res.json(gsaimsSection('M12_reportingDisclosure', 'M12-S3')));
+
+// Generic section lookup
+app.get('/api/gsifi-aims/sections/:id', (req, res) => {
+ const id = req.params.id.toUpperCase();
+ for (const mod of Object.values(GSAIMS_MODULES)) {
+ const s = ((mod && mod.sections) || []).find(x => (x.id || '').toUpperCase() === id);
+ if (s) return res.json(s);
+ }
+ return res.status(404).json({ error: 'section not found', id: req.params.id });
+});
+
+// Schemas / code examples / case studies
+app.get('/api/gsifi-aims/schemas', (_, res) => res.json(GSAIMS.schemas || {}));
+app.get('/api/gsifi-aims/schemas/:name', (req, res) => {
+ const sch = (GSAIMS.schemas || {})[req.params.name];
+ if (!sch) return res.status(404).json({ error: 'schema not found', name: req.params.name });
+ res.json(sch);
+});
+app.get('/api/gsifi-aims/code-examples', (_, res) => res.json(GSAIMS.codeExamples || {}));
+app.get('/api/gsifi-aims/code-examples/:name', (req, res) => {
+ const c = (GSAIMS.codeExamples || {})[req.params.name];
+ if (!c) return res.status(404).json({ error: 'code example not found', name: req.params.name });
+ res.json(c);
+});
+app.get('/api/gsifi-aims/case-studies', (_, res) => res.json(GSAIMS.caseStudies || []));
+app.get('/api/gsifi-aims/case-studies/:id', (req, res) => {
+ const u = req.params.id.toUpperCase();
+ const cs = (GSAIMS.caseStudies || []).find(c => (c.id || '').toUpperCase() === u);
+ if (!cs) return res.status(404).json({ error: 'case study not found', id: req.params.id });
+ res.json(cs);
+});
+
// SECTION 10: START SERVER
// ══════════════════════════════════════════════════════════════════════════════