← Back to interactive architecture

Security & Threat Model

Trust zones and crossing rules, the abuse-case threat model with residual risk stated, the identity fabric, the full asset-mapped control catalog, security operations, and AI-specific security.

1 · Security-by-design stance

Six rules govern every design in this package: least privilege (scoped identities, JIT elevation), deny by default (network, topics, data zones, tool grants), separation of duties (authorship ≠ deploy approval; backup admin realm ≠ production IdP), secure defaults (golden paths ship compliant), complete mediation (edge gateway for APIs, AI gateway for models, B2B gateway for partners — no side doors), and defense in depth (edge → gateway → service → data controls layered, each assuming the previous failed). Controls are mapped to components and flows — never listed in the abstract: see section 5 and data/controls.json.

Enforcement lives at the strongest available layer: admission control rather than convention, policy-as-code rather than documents, entitlement filters in the retrieval path rather than UI hiding. Where the source document names a discipline (S25, S34), this model names the enforcement point.

2 · Trust model

Eight (+1 logical) trust zones; every crossing is mediated, authenticated, encrypted and logged. Zone membership is a node field rendered on every view; crossings are visible as relationships with full security metadata (click any edge in the app).

ZoneMeaning
internetInternet (untrusted)
dmz-edgeEdge / DMZ (mediated ingress)
app-internalApplication zone (private)
data-internalData zone (private, restricted egress)
mgmt-planeManagement plane (privileged)
partner-boundaryPartner boundary (contracted external)
onprem-legacyOn-premises / legacy zone
saas-externalSaaS / external provider zone
n/aNot applicable (logical / organizational)
CrossingMediation rule
internet → dmz-edgeOnly via CDN/WAF/DDoS stack (sec-edge); TLS terminated and re-established; bot management active.
dmz-edge → app-internalOnly via the API gateway with validated tokens (C-APP-01); gateway-to-upstream mTLS.
app-internal → data-internalPrivate endpoints only; workload identity; classification-compatible access enforced by policy; no public data-plane endpoints (scan target: zero).
app-internal → saas-externalEgress via inspected paths; OAuth service principals with least scopes; vendor terms reviewed (procurement + security).
partner-boundary ↔ internalOnly via B2B gateway/MFT with per-partner identity; canonical validation before anything internal consumes partner input.
anything → mgmt-planePAM-brokered JIT human access with recording; workload paths via IaC/GitOps pipelines only.
onprem-legacy ↔ cloudPrivate connectivity only; legacy interfaces wrapped by the ACL; legacy service accounts vaulted and flagged as expiring exceptions.
AI runtime → anythingModels via the AI gateway only; agent tools via the tool registry only; retrieval entitlement-filtered (C-AI-01/03/04).

3 · Threat model

Method: assets and entry points from the canonical model; abuse cases below cover the highest-consequence classes. This is a living register — P1 discovery and each new capability extend it. Residual risk is stated, not hidden.

Abuse caseActorEntry pointsMitigationsControlsResidual risk & test focus
Credential theft / phishing (workforce)External attackerIdP login, MFA fatigue, session tokensPhishing-resistant MFA, conditional access, session revocation, sign-in analytics to SIEMC-ID-01, C-OPS-03Residual: token theft post-auth — short lifetimes + revocation drills
Credential stuffing / account takeover (customers)External attackerCIAM registration/login, password resetRisk-based MFA, bot protection on auth endpoints, breach-password screening, rate limitsC-ID-02, C-NET-01, C-APP-03Residual: low-and-slow attacks — detection tuning
API abuse & injectionExternal attacker / malicious partnerEdge gateway, partner APIs, webhooksWAF, schema validation, authN/Z on every call, rate limits/quotas, webhook signature verificationC-NET-01, C-APP-01/02/03Residual: business-logic abuse — abuse-case monitoring
Lateral movement after workload compromiseAttacker with a footholdEast-west network, service credentialsDeny-by-default segmentation, workload identity + mTLS, no static secrets, EDR on workloadsC-NET-02, C-ID-04, C-APP-04Residual: same-namespace movement — runtime detection depth
Supply-chain injection (dependency/pipeline)External attacker / insiderDependencies, CI runners, registryScan gates, SBOM, signing + provenance verified at admission, branch protection, separation of dutiesC-SC-01…04Residual: novel upstream compromise — provenance limits blast radius
Ransomware incl. backup destructionCriminal actor with admin theftBackup credentials, storage APIsImmutability windows, isolated cyber vault with separate credential realm, restore drills, PAM/JIT on adminC-RES-01/02, C-ID-03Residual: dwell-time encryption of live data — detection + tiered restore
Data exfiltration via analytics / AI retrievalInsider / over-permissioned consumer / prompt attackerLakehouse serving, BI, RAG retrievalClassification-driven access + masking, entitlement-filtered retrieval with provenance, DLP on user-to-SaaS, egress controlC-DP-01/03, C-AI-03, sec-sseResidual: authorized-user misuse — UEBA-style analytics maturity
Prompt injection & tool abuse (agents)External content author / malicious userModel inputs, retrieved content, tool outputsGuardrail filters, tool allowlists + transaction limits + sandbox, HITL above thresholds, kill switch, red-team suitesC-AI-02/04/05Residual: novel injection techniques — scheduled red-teaming, bounded permissions cap damage
Partner-boundary compromiseCompromised trading partnerB2B/EDI channels, MFT dropsPer-partner identities/certs, canonical validation before internal publish, reconciliation, anomaly alerts on document patternsC-INT-04/05, C-OPS-03Residual: valid-but-fraudulent documents — business-rule screening
Operator error / console driftWell-meaning insiderCloud consoles, cluster adminIaC-only with drift detection, JIT + session recording, change freezes, GitOps revert, PITRC-CLD-02, C-ID-03, RB-13/17Residual: bad-but-valid IaC — review + progressive rollout

Testing: control evidence per catalog entry; scheduled attack-path review joins posture data (sec-posture) with this register; pen-test scope per change class; AI red-team suites per C-AI-02. Evidence retention per C-GOV-03.

4 · Identity fabric

PopulationSystemModel
Workforce (employees/contractors)sec-idp-workforce + sec-igaHRIS-driven JML, SSO, phishing-resistant MFA, conditional access (the human PDP), access reviews.
Customerssec-ciamSelf-registration, risk-based MFA, consent capture, token issuance for channels/APIs.
Partners (humans)sec-ciam federationBrokered federation to partner IdPs where possible; partner-scoped claims.
Partners (systems)int-b2b / int-mftPer-partner certificates/keys under trading agreements; rotation via RB-08.
Privileged humanssec-pamVaulted, JIT, recorded; break-glass dual-controlled (RB-02).
Services / workloadssec-workload-id + sec-secretsPlatform-issued short-lived identities; mTLS on sensitive paths; static secrets are expiring exceptions.
PipelinesOIDC federationNo long-lived cloud keys in CI; environment-scoped deploy roles with separation of duties.
AI agentsai-tools grantsTool-scoped least-privilege identities; never broad service accounts; grants expire and are recertified (RB-12).

The zero-trust policy decision point (sec-pdp) is deliberately logical: one policy model realized across IdP conditional access (humans), gateway policies (APIs), platform admission (workloads) and retrieval filters (AI). The realization mapping lives on the node and prevents the classic gap where each PEP invents its own policy language.

5 · Control catalog — mapped to assets

49 controls across 12 categories. The canonical source is data/controls.json; the app's security view shows the same mapping interactively (click any control chip).

Identity controls

IDControlEnforcementApplies toEvidence
C-ID-01
preventive
Workforce SSO with phishing-resistant MFA
All workforce access via central IdP: SSO (OIDC/SAML), phishing-resistant MFA, conditional access evaluating device posture and context.
platform configurationWorkforce Identity Provider Employee Workspace Internal Developer Portal & Service Catalog Observability Platform (metrics / logs / traces) Collaboration & Productivity Suite ITSM & CMDB BI & Analytics ServiceIdP policy export; MFA coverage report; conditional-access rule review
C-ID-02
preventive
Customer & partner identity via CIAM
External identity handled by CIAM: registration, risk-based MFA, consent capture, token issuance; partner federation brokered - no local accounts in apps.
platform configurationCustomer & Partner Identity (CIAM) Customer Web Portal Customer Mobile App Partner PortalCIAM tenant policy export; token lifetime config; federation agreements
C-ID-03
preventive
Privileged access is JIT, vaulted and recorded
No standing admin rights: PAM-vaulted credentials, just-in-time elevation with approval, session recording, monitored break-glass.
platform configurationPrivileged Access Management Cloud Landing Zones & Account Structure On-Premises Data Center Container Platform (managed Kubernetes) Operational Databases (per-service)PAM access reports; break-glass drill evidence; standing-privilege scan results
C-ID-04
preventive
Workload identity replaces static service credentials
Service-to-service auth uses platform-issued short-lived workload identities; static API keys/secrets are exceptions with expiry.
platform configurationWorkload Identity & Service Identity (mTLS) Container Platform (managed Kubernetes) Order Management Service Billing & Invoicing Service Agent Tool Registry & Execution Sandbox CI/CD PipelinesSecret inventory trend; workload-identity adoption metric; exception register
C-ID-05
administrative
Joiner-mover-leaver automation with access reviews
HRIS-driven provisioning/deprovisioning via IGA; periodic access certification for privileged and sensitive-data roles.
process + toolingIdentity Governance & Administration HRIS (Workforce SoR) Workforce Identity ProviderJML latency metrics; completed review campaigns; orphan-account scans

Network controls

IDControlEnforcementApplies toEvidence
C-NET-01
preventive
Edge protection on all internet ingress
WAF managed rules + DDoS absorption + bot management + TLS policy in front of every channel and the API gateway; no bypass routes.
platform configurationWAF, DDoS & Bot Protection Customer Web Portal Customer Mobile App Partner Portal API Gateway (Edge)Edge policy export; bypass-route scan; blocked-attack dashboards
C-NET-02
preventive
Deny-by-default segmentation with private endpoints
East-west traffic denied unless declared; PaaS/data services reachable via private endpoints only; egress via inspected hub.
policy-as-codeNetwork Hub & Segmentation Container Platform (managed Kubernetes) Operational Databases (per-service) Lakehouse (raw / validated / curated / serving) Vector Store & IndexNetwork policy coverage report; public-endpoint scan (should be zero on data zone)
C-NET-03
preventive
Hybrid connectivity redundancy
Private circuit + independent VPN failover for on-prem coexistence; capacity and failover tested on schedule.
platform configurationHybrid Connectivity On-Premises Data CenterFailover test records; utilization/error monitoring

Application controls

IDControlEnforcementApplies toEvidence
C-APP-01
preventive
OAuth2/OIDC discipline on every API
Token validation (issuer, audience, expiry, signature), scopes/claims authorization, replay protection, key rotation; enforced at gateway and re-checked in services.
gateway + service configAPI Gateway (Edge) Experience API - Customer BFF Experience API - Partner BFF Order Management Service Billing & Invoicing Service AI Gateway & Model Access ControlGateway policy export; contract tests asserting 401/403 behavior; key-rotation log
C-APP-02
preventive
Schema validation and input handling at boundaries
Request/response schema validation, input sanitization, output encoding; contract tests in CI block drift.
service implementation + CIExperience API - Customer BFF Experience API - Partner BFF Order Management Service Customer Profile Service API Gateway (Edge)CI contract-test results; fuzzing findings trend
C-APP-03
preventive
Rate limiting, quotas and abuse monitoring
Per-client rate limits and quotas at the gateway; anomaly-based abuse detection feeding SIEM.
gateway configurationAPI Gateway (Edge) Experience API - Partner BFF AI Gateway & Model Access ControlLimit policy export; throttling dashboards; abuse-case runbook drills
C-APP-04
preventive
mTLS / service identity on sensitive internal paths
Mutual TLS or platform service identity required for service-to-service calls touching restricted data.
platform configurationWorkload Identity & Service Identity (mTLS) Billing & Invoicing Service Customer Profile Service RAG Retrieval Service (authorization-aware) Operational Databases (per-service)Mesh/identity policy export; plaintext-internal-traffic scan
C-APP-05
preventive
Secure session and token handling on channels
Short-lived access tokens, secure refresh flows, revocation on risk signals, no sensitive data in browser storage beyond policy.
channel implementationCustomer Web Portal Customer Mobile App Partner Portal Customer & Partner Identity (CIAM)Channel security review; token-lifetime config; revocation drill

Data-Protection controls

IDControlEnforcementApplies toEvidence
C-DP-01
administrative
Classification drives policy
Every dataset/data product registered and classified in the catalog; classification tags drive access, masking and retention automatically.
catalog + policy-as-codeData Catalog, Lineage & Metadata Lakehouse (raw / validated / curated / serving) Customer 360 Data Product Orders & Fulfillment Data Product Vector Store & Index Document & Content ServiceUnclassified-dataset count (target zero); tag-to-policy mapping tests
C-DP-02
preventive
Encryption at rest with owned keys
All stores encrypted at rest; keys in KMS (HSM-backed where justified) with documented ownership, rotation and revocation procedures.
platform configurationKMS / HSM, PKI & Certificate Lifecycle Operational Databases (per-service) Lakehouse (raw / validated / curated / serving) Vector Store & Index Backup & Immutable Backup PlatformKey inventory with owners; rotation logs; encryption coverage scan
C-DP-03
preventive
Masking / tokenization outside production need-to-know
Restricted data masked or tokenized in non-prod, analytics and AI pipelines unless a documented exception exists.
platform configurationLakehouse (raw / validated / curated / serving) Customer 360 Data Product Embedding & Chunking Pipeline Operational Databases (per-service)Masking policy tests; non-prod data scans
C-DP-04
administrative
Retention, deletion and legal hold per record class
Record classes carry retention schedules and deletion procedures; legal hold suspends deletion with auditable scope.
process + toolingDocument & Content Service Lakehouse (raw / validated / curated / serving) Backup & Immutable Backup Platform Collaboration & Productivity SuiteRetention schedule; deletion-job evidence; legal-hold register
C-DP-05
preventive
Data residency overlay (configurable)
Residency constraints (pending OQ-02) enforced via region allowlists in landing-zone policy and storage placement rules. No residency claim is made until requirements are confirmed.
policy-as-codeCloud Landing Zones & Account Structure Lakehouse (raw / validated / curated / serving) Hosted Foundation Model ServicesRegion-restriction policy export; placement audit

Cloud controls

IDControlEnforcementApplies toEvidence
C-CLD-01
preventive
Landing-zone guardrails as code
Org-level policies: allowed regions/services, mandatory tags, encryption defaults, logging on, public-exposure bans - evaluated on every deployment.
policy-as-codeCloud Landing Zones & Account Structure Policy-as-Code EnginePolicy pack version; violation trend; exemption register with expiry
C-CLD-02
preventive
IaC-only provisioning with drift detection
Production changes flow through version-controlled IaC and GitOps reconciliation; manual console changes are detected and treated as incidents.
GitOpsIaC & Environment Provisioning (GitOps) Cloud Landing Zones & Account Structure Container Platform (managed Kubernetes)Drift alerts; console-change audit; IaC coverage metric
C-CLD-03
preventive
Hardened baselines, patching and admission control
CIS-aligned images (baseline reference, not certification), patch SLAs, image scanning, and admission policies rejecting unscanned/unsigned workloads.
platform configurationContainer Platform (managed Kubernetes) VM Estate (rehost landing) Artifact & Container RegistryBaseline scan results; patch SLA attainment; admission-rejection logs
C-CLD-04
detective
Continuous posture and workload protection
CSPM misconfiguration detection against guardrails plus workload runtime protection; findings queue with remediation SLAs.
platform configurationCNAPP (CSPM / CWPP) & Vulnerability Management Cloud Landing Zones & Account Structure Container Platform (managed Kubernetes)Open-findings ageing; SLA attainment; recurring-finding analysis

Supply-Chain controls

IDControlEnforcementApplies toEvidence
C-SC-01
preventive
Security scanning gates in CI
SAST, SCA, secret scanning, container and IaC scanning on every build; severity thresholds block release with documented emergency path.
pipeline gatesCI/CD Pipelines Source Control & Code ReviewPipeline gate config; blocked-release log; emergency-release records
C-SC-02
preventive
SBOM, signing and provenance verification
Every artifact ships an SBOM, is signed, and carries provenance attestation; cluster admission verifies before scheduling.
pipeline + admissionCI/CD Pipelines Artifact & Container Registry Container Platform (managed Kubernetes)Attestation coverage; admission verification logs
C-SC-03
preventive
Branch protection and deploy separation of duties
Protected branches, mandatory review, no self-approval to production; deploy approvals separated from authorship.
platform configurationSource Control & Code Review CI/CD PipelinesBranch-protection export; approval audit samples
C-SC-04
corrective
Vulnerability lifecycle with expiring exceptions
Findings triaged by severity with remediation SLAs; exceptions require owner, compensating control and expiry date.
process + toolingCNAPP (CSPM / CWPP) & Vulnerability Management CI/CD Pipelines ITSM & CMDBSLA attainment; exception register review

Integration controls

IDControlEnforcementApplies toEvidence
C-INT-01
preventive
Contract and schema governance
APIs and events are contract-first; registry-enforced compatibility (backward) with major-version process for breaking changes.
registry + CISchema & Contract Registry API Management & Developer Portal Event Streaming BackboneCompatibility-check logs; breaking-change approvals
C-INT-02
corrective
DLQ, replay and poison-message handling
Every consumer has bounded retries then DLQ with owner alerting; replay procedures documented and rehearsed.
platform configurationEvent Streaming Backbone Message Queue Broker Fulfillment Orchestration Service Notification Service B2B Gateway (EDI / AS2)DLQ depth dashboards; replay drill records
C-INT-03
preventive
Idempotent consumers and deduplication
At-least-once delivery assumed everywhere; consumers dedupe on event/command ids; producers use outbox pattern.
service implementationFulfillment Orchestration Service Notification Service Document & Content Service Stream ProcessingIdempotency tests in CI; duplicate-processing incident trend
C-INT-04
preventive
Governed file transfer integrity
All partner/bank file exchange via MFT/B2B gateway: encryption, checksums, non-repudiation logs, window monitoring - no ad-hoc scripts.
platform configurationManaged File Transfer B2B Gateway (EDI / AS2) Trading Partners (EDI / API)Transfer logs with checksums; ad-hoc-transfer scan
C-INT-05
detective
Reconciliation on financial and partner flows
Independent reconciliation (counts + amounts) between systems on billing, payment and partner document flows; discrepancies alert to business owners.
scheduled jobsBilling & Invoicing Service ERP (Finance & Procurement SoR) Payment Service Provider B2B Gateway (EDI / AS2)Reconciliation reports; discrepancy ageing

Ai controls

IDControlEnforcementApplies toEvidence
C-AI-01
preventive
All model access through the AI gateway
No application holds provider keys; gateway enforces per-consumer authN/Z, model allowlists, quotas and logging with redaction.
gateway configurationAI Gateway & Model Access Control Hosted Foundation Model Services Self-Managed Model Serving RAG Retrieval Service (authorization-aware) Agent Orchestrator & PlannerDirect-provider-egress scan (target zero); gateway policy export
C-AI-02
preventive
Prompt-injection defense and output validation
Input/output guardrails: injection heuristics, content safety categories, sensitive-data redaction, tool-output sanitization; scheduled red-team suites.
inline filters + testingAI Safety Guardrails & Content Filtering AI Gateway & Model Access Control Agent Orchestrator & PlannerRed-team results; filter-block metrics; regression suite in CI
C-AI-03
preventive
Authorization-aware retrieval with provenance
RAG retrieval filters by caller entitlements before context assembly; responses carry citations; only cataloged, classified sources are embedded.
service implementationRAG Retrieval Service (authorization-aware) Vector Store & Index Embedding & Chunking Pipeline Data Catalog, Lineage & MetadataEntitlement-bypass tests; provenance coverage metric
C-AI-04
preventive
Agent tools are allowlisted, scoped and sandboxed
Agents act only via registered tools with least-privilege grants, transaction limits and sandboxed execution; no broad production permissions by default.
platform configurationAgent Tool Registry & Execution Sandbox Agent Orchestrator & Planner API Gateway (Edge)Tool grant register; sandbox escape tests; permission diff audits
C-AI-05
preventive
Human approval and kill switch for high-impact actions
Governance-defined thresholds route actions to human approval; kill switch disables agents/models per scope with rehearsed procedure (RB-12).
workflow + platformHuman-in-the-Loop Approval Service Agent Orchestrator & Planner AI Gateway & Model Access ControlApproval audit trail; kill-switch drill records
C-AI-06
detective
AI evaluation, drift and cost monitoring
Scheduled evals against golden sets (incl. hallucination checks), drift monitors, token/inference budgets with alerts and fallback routes.
platform configurationAI Observability & Cost Controls ML Platform (training, registry, evaluation) AI Gateway & Model Access Control Cost Management & Allocation ToolingEval score trends; drift alerts; budget-vs-actual reports

Operations controls

IDControlEnforcementApplies toEvidence
C-OPS-01
detective
Standard telemetry with correlation ids
OpenTelemetry-conformant metrics/logs/traces with propagated correlation ids on every service and platform; SLOs with named owners.
instrumentation standardTelemetry Pipeline (OpenTelemetry) Observability Platform (metrics / logs / traces) Container Platform (managed Kubernetes) API Gateway (Edge) Order Management ServiceInstrumentation coverage; orphan-alert audit (every alert has an owner)
C-OPS-02
detective
Synthetic and real-user monitoring of key journeys
Synthetic probes for order placement, login and partner document submission plus RUM on channels; DR checks reuse synthetics.
platform configurationSynthetic Monitoring, RUM & DEM Customer Web Portal Partner PortalSynthetic uptime history; journey coverage list
C-OPS-03
detective
SIEM onboarding is part of go-live
Every new platform/service onboards security-relevant logs to SIEM before production; detections mapped to a public threat framework.
process + toolingSIEM & Security Analytics Cloud Landing Zones & Account Structure API Gateway (Edge) AI Gateway & Model Access ControlSource-onboarding checklist evidence; detection coverage map
C-OPS-04
corrective
Automated response with human gates
Containment playbooks (token revoke, host isolate, account disable) automated with human approval for destructive/irreversible steps.
SOAR playbooksSOAR & Response Automation SIEM & Security Analytics Workforce Identity ProviderPlaybook run logs; approval records; response-time trend

Resilience controls

IDControlEnforcementApplies toEvidence
C-RES-01
corrective
Backups with immutability and tested restores
Policy-driven backups with immutability windows on critical sets; restore tests on schedule - restore evidence is the control, not job success.
platform configurationBackup & Immutable Backup Platform Operational Databases (per-service) Lakehouse (raw / validated / curated / serving) Legacy Core Business System (Monolith)Restore test records with timings; immutability config export
C-RES-02
corrective
Isolated cyber-recovery vault
Delayed, immutable replication of tier-1 recovery data to a vault with separate credentials and access path; assumes primary backup compromise.
platform configurationCyber Recovery Vault Backup & Immutable Backup PlatformVault isolation review; recovery drill from vault
C-RES-03
corrective
Tiered DR runbooks, rehearsed
Per-tier failover/failback sequences with dependency ordering and data-consistency checks; rehearsals produce timed evidence against placeholder objectives until BIA sets real ones.
process + toolingDR Orchestration & Recovery Runbooks Order Management Service Operational Databases (per-service) Event Streaming BackboneDR exercise reports; gap actions tracked
C-RES-04
preventive
Multi-AZ by default for critical tiers
T1/T2 services and stateful platforms deploy across availability zones; single-AZ deployments require an expiring exception. Multi-REGION posture is a separate open decision (OD-01) - not assumed.
platform configurationContainer Platform (managed Kubernetes) Operational Databases (per-service) Event Streaming Backbone Lakehouse (raw / validated / curated / serving)AZ-distribution audit; exception register

Governance controls

IDControlEnforcementApplies toEvidence
C-GOV-01
administrative
Decisions and exceptions are recorded and expiring
Material choices captured as ADRs; deviations tracked as exceptions with owner, compensating control and expiry; both reviewed on cadence.
processArchitecture Review Board Architecture Repository & StandardsADR log; exception register ageing
C-GOV-02
detective
Tagging, allocation, budgets and anomaly detection
Mandatory ownership/cost tags enforced at deploy; budgets and anomaly alerts per product; showback (chargeback pending org decision) and unit-economics reporting.
policy-as-code + toolingCost Management & Allocation Tooling Cloud Landing Zones & Account Structure Policy-as-Code EngineTag coverage; anomaly response times; budget variance reports
C-GOV-03
administrative
Control evidence retention
Control test results, approvals, review minutes and drill records retained per audit policy in referenceable systems (ITSM/repo), not chat threads.
process + toolingITSM & CMDB Architecture Repository & Standards SIEM & Security AnalyticsEvidence-retrieval spot checks

Privacy controls

IDControlEnforcementApplies toEvidence
C-PRV-01
preventive
Privacy overlay: consent, minimization, DSR readiness
Consent flags respected across profile, analytics and AI pipelines; data minimization at collection; data-subject-request hooks (access/delete) wired to stores. Applicable regulations pending OQ-03 - modeled as a configurable overlay, no compliance claim.
configurable overlayCustomer Profile Service Customer 360 Data Product Embedding & Chunking Pipeline Document & Content ServiceConsent propagation tests; DSR drill records

6 · Security operations

Telemetry: identity, endpoint, network, cloud control-plane, gateway, B2B and AI-gateway logs onboard to the SIEM as part of every go-live (C-OPS-03) — a platform without detections does not pass its gate. Detection: engineering mapped to a public threat framework with coverage tracked; detections are code-reviewed like software. Response: SOAR playbooks automate enrichment and containment with human approval on destructive steps (C-OPS-04); incident flow per RB-16 with evidence preservation and honest notification decisions (legal/privacy in the loop). Hunting & review: hypothesis-driven hunts on cadence; every Sev-1/2 gets a blameless PIR whose actions are tracked to closure; the risk register and this threat model are updated from real events.

7 · AI-specific security

The source (S41/S42) names AI ambition but no AI threat model — this section is a proposal filling that gap. Complete mediation applies twice: models only via the AI gateway (C-AI-01: allowlists, virtual keys, quotas, redacted logging) and actions only via registered tools (C-AI-04: least-privilege grants, transaction limits, sandbox, full audit). Inputs and outputs pass guardrails (C-AI-02: injection heuristics, content safety, sensitive-data redaction, tool-output sanitization) with scheduled red-team suites. Retrieval is entitlement-filtered with provenance (C-AI-03) so RAG cannot become a data-leak path. High-impact actions require human approval (C-AI-05) and every scope has a rehearsed kill switch (RB-12). Evaluation, drift and token-cost monitoring (C-AI-06) make AI behavior and spend observable like any other production system. Autonomy is graduated — assist → recommend → bounded-auto — by AI-governance decision with evidence, never by default (ADR-14).

8 · Privacy, records & regulatory overlays

Applicable regulations are unknown (OQ-02/OQ-03) and no compliance is claimed (A-11). The architecture prepares configurable overlays instead: consent propagation and DSR hooks (C-PRV-01), retention/deletion/legal-hold per record class (C-DP-04), residency placement rules (C-DP-05), and evidence retention (C-GOV-03). When the organization confirms its regulatory set, overlays are activated and mapped — the traceability document records that activation as a model change, not a slide claim.