CAIBots · Production Implementation Guide

§1

System Architecture

How It Works in Production

The core mental model · What runs · How intelligence is generated · What the BSA Officer receives

The Core Mental Model

The CAIBots KYC/AML Agentic AI is not a checklist tool or a rules engine with an AI wrapper. It is a continuously running compliance intelligence layer that sits between your customer data infrastructure and your BSA officer's desk. The fundamental shift: the analyst's job changes from researcher to decision-maker. The agent does all investigation, evidence assembly, regulatory mapping, and documentation. The human approves or rejects the conclusion. Every critical action is gated behind a mandatory human-in-the-loop checkpoint.

§1.1 · Visual Reference

Full System Architecture Diagram

Interactive diagram: all 8 layers from AML platform through BSA Officer outputs. Trigger horizons, orchestrator, 7 agents, 5 source tools, HITL gates, and compliance outputs — all in one view.

View Architecture →

§1.2 · Three Trigger Horizons Running Simultaneously

Onboarding

New customer application · Beneficial owner added · Authorized signatory change · Product upgrade (consumer → business banking) · New correspondent relationship

Perpetual KYC

Risk score drift >15 points · Adverse media NLP alert · Sanctions / PEP watchlist hit · Significant transaction velocity change · Ownership structure change · Source of funds anomaly · Scheduled periodic review due (12 / 24 / 36 months by risk tier)

Transaction Monitoring

AML rule engine alert (Actimize / Verafin / your rule engine) · CTR threshold event >$10K cash · Structuring velocity flag · High-risk jurisdiction wire · Correspondent banking anomaly · 52 FATF typology pattern match · FinCEN 314(a) batch match

Key Principle — Event-Driven, Not Calendar-Driven

pKYC is event-driven, not calendar-driven. A genuinely low-risk customer who never changes never consumes analyst time. Only customers whose risk profile actually moves generate work. This is the source of the 60–70% reduction in pKYC workload.

§1.3 · The 7 Sub-Agents — What Each Does in Real Time

IAL-IDENTIdentity VerificationJumio · Onfido · IDEMIA · Socure · Alloy

Authenticates identity documents, validates biometric liveness, computes synthetic identity score (Socure/Alloy), and establishes NIST 800-63-3 IAL2/IAL3 assurance level. Detects synthetic identities, document forgery, and identity mismatch across channels. Runs in seconds against live identity APIs.

UBO-CHAINBeneficial OwnershipD&B ORBIS · Bureau van Dijk · OpenCorporates · FinCEN BOI

Traverses the full UBO chain to natural persons in real time. Enforces 25% threshold for standard CDD and 10% for EDD-elevated cases. Detects circular ownership, shell company layering depth, nominee structures, and trust/bearer share arrangements. Cross-references the FinCEN BOI registry under CTA 2024.

ADV-MEDIAAdverse MediaDow Jones · Refinitiv World-Check · LexisNexis · ComplyAdvantage

Continuous NLP across 300+ monitored sources. Classifies sentiment, allegation type, jurisdiction, and severity on a 0–100 adverse media score. Network propagation: UBO adverse media contaminates the customer's network risk score — guilt-by-association risk travels through ownership edges automatically.

5-DIM RISKRisk ScoringSynthesized from All Agent Outputs · Five Dimensions

Synthesizes inputs from all other agents into a single 0–100 score across five dimensions: (1) Geographic risk — FATF blacklist/greylist, HIFCA designations; (2) Product/channel risk — private banking, correspondent, wire-heavy; (3) Customer type risk — PEP proximity, cash-intensive, NBFI; (4) Behavioral risk — transaction velocity, source of funds; (5) Network risk — related-party SAR history, adverse ownership. Routes to SDD (<30), CDD (30–65), or mandatory EDD (>65 or any PEP/sanctions adjacency).

OFAC-SANCSanctions ScreeningOFAC SDN · EU Consolidated · UN Security Council · HMT · World-Check

Real-time screening — not batch — against all major sanctions lists. Fuzzy name matching handles transliteration variants. Screens 2-hop PEP network proximity. OFAC SDN cache TTL is 1 hour — system never relies on stale list data for more than 60 minutes. Immediate blocking on any SDN hit.

FATF-52TMTransaction MonitoringActimize · Verafin · Fiserv · 52 FATF Typologies

Pattern-matches against all 52 FATF money laundering typologies in real time. For structuring cases, calculates total avoided CTR reporting amount, maps counterparty network, and identifies the precise period of suspicious activity. Evaluates SAR filing obligation under 31 U.S.C. §5318(g). Integrates bidirectionally with your existing AML rule engine.

REG-DOCSRegulatory DocumentationFinCEN BSA E-Filing · OpenText / Documentum ECM · LangChain LCEL

Assembles the complete examination-ready compliance file: CDD narrative memo, EDD memorandum, SAR narrative (FinCEN-compliant format with activity description, subject identification, transaction schedule, and statutory citations), denial/exit letter, and the mandatory action checklist. Every finding cited to a specific regulatory source. Output structured for direct ECM archival.

§1.4 · Five Peer Source Tools — No Hierarchy

All 7 agents query from the same 5 source tools simultaneously. No tool has precedence — conflicts trigger weighted synthesis at the orchestration layer.

Tool	Technology	What It Provides	Used For
[VECTOR-RAG]	Pinecone + pgvector (DR) · LangChain LCEL	FinCEN CDD Rule, FFIEC manual, FATF 40 Recommendations + 52 typology playbooks, OFAC framework, FinCEN guidance 2012–2025. Chunk-level citations returned with every determination.	All cases — regulatory grounding, SAR narrative generation, EDD memo drafting
[API-CALLS]	Jumio/Onfido/IDEMIA · OFAC SDN · Dow Jones · D&B ORBIS · FinCEN BOI	Live identity verification scores, real-time sanctions list results, PEP/adverse media scores, UBO chain data from external registries.	All cases — identity, sanctions, PEP screening run in real time on every trigger
[SQL-DB]	Snowflake · 36-month transaction window	Customer risk score history, transaction patterns (velocity, geography, counterparty), SAR/CTR filing history, prior KYC review outcomes.	CDD/EDD/TM — behavioral risk dimension, structuring analysis, prior case context
[KNOW-GRAPH]	Neo4j AuraDB Enterprise	UBO ownership chains (25%/10% thresholds), PEP network proximity (2-hop), SAR subject linkage, correspondent banking nested chain mapping.	EDD, correspondent banking, 314(a) — network risk dimension, circular ownership detection
[MEMORY]	Redis (In-Context + KV Cache) · In-Weights (fine-tuned)	Prior compliance decisions (Redis), current session state (In-Context), OFAC cache TTL 1hr / PEP cache TTL 4hr (KV Cache), institution-calibrated SAR style (In-Weights).	All cases — institutional precedent, session continuity, cache efficiency, narrative style

§1.5 · What the BSA Officer Receives

When the pipeline completes, the BSA officer's workbench is pre-populated with a complete, examination-ready evidence packet. The officer's job is to evaluate and decide — not to research.

Complete identity confirmation with IAL assurance level and synthetic identity score

UBO structure diagram with traversal path, ownership percentages, and jurisdiction flags

Risk score (0–100) with five-dimension breakdown and historical drift chart

Adverse media score with source citations and allegation classification by type

Sanctions screening result (clear or hit) with list source and fuzzy match confidence score

Transaction pattern analysis with specific dates, amounts, and FATF typology mapping

Draft CDD or EDD memorandum — examination-ready format, with regulatory citations

Draft SAR narrative (if applicable) — FinCEN-compliant with activity description, subject identification, transaction schedule, and statutory citations (31 U.S.C. §5318(g), 31 C.F.R. §1020.320)

Mandatory action checklist with all applicable regulatory deadlines and escalation timers

HITL gate buttons: Approve · Escalate · Deny — every action timestamped and immutably logged

BSA Officer Efficiency — What Changes

Agent pipeline runtime: <10 seconds. EDD review time: 4.5 hours → 55 minutes (80% reduction). CDD review time: 90 minutes → 12 minutes (87% reduction). The analyst is no longer building the case file — they are evaluating a completed one. The 70%+ SDD auto-clear rate for low-risk retail means the majority of onboarding volume never reaches an analyst at all.

§2

Responsibility Assignment

Who Does What — Client vs. CAIBots

Every production rollout task assigned by owner · Colour-coded by responsibility · Read this before §3 onwards

The sections that follow describe how each capability works in production. This section answers the prior question: who is responsible for making it happen. Every task in a live deployment falls into one of three categories — the client institution owns it, CAIBots delivers it, or both share it. Misunderstanding this split is the most common cause of production delays.

CLIENT: OWNS CLIENT: PROVIDES CLIENT: REVIEWS CAIBOTS: BUILDS CAIBOTS: DELIVERS SHARED

⚠ Cannot Be Delegated to CAIBots — Under Any Structure

BSA Officer authorization on all SAR filings, EDD determinations, and 314(a) responses. Selection and payment of the independent model validator (SR 11-7). Response to regulatory examiners. Compliance policy decisions including all risk threshold sign-offs. These are legal obligations of the regulated institution.

1 · Data Access & Integrations

Task	Client	CAIBots	Notes
Data Vendor Contracts
Contract & pay for identity verification vendor (Jumio/Socure/Onfido)	OWNS	CONFIGURES	Client selects per procurement policy. CAIBots integrates to whichever vendor is selected.
Contract & pay for sanctions feed (Refinitiv/Dow Jones/LexisNexis)	OWNS	CONFIGURES	Client may have existing contract. CAIBots adapts to the client's contracted feed.
Contract & pay for adverse media provider	OWNS	CONFIGURES	CAIBots has pre-built integrations for Dow Jones RCA, Refinitiv World-Check, LexisNexis Bridger.
System Access
Provide read-only API access to core banking platform	PROVIDES	BUILDS	Client security team provisions credentials. CAIBots provides integration spec. No write access ever.
Provision UAT/sandbox environment with anonymised production data	OWNS	BUILDS	Client provisions. CAIBots deploys application stack. No real PII in non-production.
Engage AML platform vendor (Actimize/Verafin) for API integration layer	OWNS	BUILDS	Client initiates with their AML vendor. Their professional services engagement is a separate cost.
FinCEN Connections
FinCEN BSA E-Filing account & credentials (SAR 30-day filing / CTR 15-day filing)	OWNS	N/A	Client must have existing BSA E-Filing relationship. BSA Officer is always filer of record.
FinCEN 314(a) batch file delivery to CAIBots pipeline (14-day response window)	PROVIDES	BUILDS	Client provides access to incoming batch files. CAIBots builds 14-day response pipeline.

2 · SR 11-7 Model Governance

Task	Client	CAIBots	Notes
Register agents in institution model inventory	OWNS	DELIVERS	Client MRM registers each agent. CAIBots provides model descriptions, inputs/outputs, and intended use.
Produce Model Development Document (MDD) for each agent	N/A	DELIVERS	CAIBots writes. Client MRM reviews and formally approves before any production traffic.
Provide 12–24 months closed cases with known outcomes (back-testing dataset)	PROVIDES	BUILDS	Client extracts closed SARs/EDD/SDD decisions. CAIBots builds back-testing framework. Min 500 cases.
Engage & pay for independent model validation firm (Oliver Wyman/Promontory/KPMG)	OWNS	DELIVERS	Non-negotiable per SR 11-7. Client hires and pays ($150K–$400K). CAIBots provides all technical docs.
Produce Model Validation Report (MVR)	N/A	REVIEWS	Produced by the independent validator. CAIBots reviews for factual accuracy only.
Deliver quarterly model performance reports	N/A	DELIVERS	CAIBots delivers. Client MRM reviews. Escalates to Model Risk Committee if thresholds breached.
30-day advance notice of any material model change	N/A	DELIVERS	CAIBots contractual obligation. Client MRM reviews before any change goes live in production.

3 · Infrastructure & Security

Task	Client	CAIBots	Notes
Decide & provision cloud deployment model (Private Cloud/On-Prem/SaaS)	OWNS	BUILDS	Client CISO and CTO decide. CAIBots provides Terraform/IaC templates for all three deployment models.
Complete TPRM/Vendor Due Diligence questionnaire	N/A	DELIVERS	CAIBots completes the client's questionnaire. Client TPRM reviews and approves. Allow 4–8 weeks.
Provide SOC 2 Type II report	N/A	DELIVERS	CAIBots delivers current report. Client CISO decides acceptability.
Third-party penetration test	OWNS	DELIVERS	Client engages pen test firm. CAIBots provides architecture docs and remediates critical findings.
Integrate with client Identity Provider (Okta/Azure AD) for SSO/MFA	PROVIDES	BUILDS	Client provides IdP configuration. CAIBots builds SAML 2.0 or OIDC integration.
Build immutable audit log of all agent actions and HITL approvals	N/A	BUILDS	CAIBots builds. Tamper-evident, append-only. 5-year retention per BSA requirements (31 C.F.R. §1020.430).

4 · BSA Compliance Configuration & HITL

Task	Client	CAIBots	Notes
Define risk tier cutoff scores (SDD/CDD/EDD thresholds)	OWNS	CONFIGURES	BSA Officer defines. CAIBots configures. Documented in SR 11-7 model card. Cannot change without formal process. See §5.
Define pKYC drift trigger threshold (+N points fires auto-refresh)	OWNS	CONFIGURES	BSA Officer sets. Default +15 points. Must be formally signed off before production. See §5.
Provide SAR narrative style guide & 50–200 prior SAR examples (redacted)	PROVIDES	CONFIGURES	CAIBots fine-tunes SAR model on institution's filing style. Quality target: >85% rated B or higher.
Define & sign off risk dimension weights (5 dimensions)	OWNS	CONFIGURES	BSA Officer reviews and approves. Rationale for any deviation from defaults must be documented. See §5.1.
Define mandatory HITL checkpoints	OWNS	CONFIGURES	Default: all SARs, all EDD escalations, all 314(a) responses. BSA Officer may add more. See §6.
Review and authorize every SAR narrative before FinCEN filing	OWNS	N/A	BSA Officer responsibility only. Cannot be delegated or automated under any circumstances. See §6.
Train and certify all BSA staff on the HITL Workbench	SHARED	DELIVERS	CAIBots delivers training programme. Client BSA management certifies completion before live access.

5 · UAT & Go-Live Authorization

Task	Client	CAIBots	Notes
Provision UAT environment with anonymised production data	OWNS	BUILDS	Client provisions. CAIBots deploys and executes all 12 required UAT scenarios.
BSA Officer review and sign-off of all 12 UAT scenarios	REVIEWS	DELIVERS	BSA Officer personally reviews each scenario output. See §7 and App B.
CCO formal approval before production promotion	OWNS	N/A	Client CCO only. CAIBots is not a party to this internal approval.
Phase 1 Shadow Mode (30–60 days minimum)	OWNS	BUILDS	Client defines duration and runs parallel process. CAIBots builds shadow run capability.
Formal authorization memo for each go-live phase (Phase 2–4)	OWNS	N/A	BSA Officer and CCO sign each phase authorization. No phase advances without written client authorization.

6 · Indicative Timeline & Key Client Dependencies

Phase	Duration	Critical Client Dependencies	CAIBots Deliverables
Integration & Setup	Wks 1–6	API credentials for all data feeds · Core banking sandbox · AML platform API	Integration architecture · Schema templates · Connection testing
SR 11-7 Setup	Wks 4–12	Back-testing dataset · MRM engaged · Independent validator selected	Model Development Document · Back-testing framework · Performance report
Infrastructure & Security	Wks 4–16	Cloud environment provisioned · Pen test scheduled · TPRM reviewed · SOC 2 accepted	IaC templates · Security hardening · Audit log · Failover
BSA Configuration	Wks 8–18	BSA Officer available · SAR style guide provided · Risk thresholds signed	Risk calibration · SAR fine-tuning · HITL workflow · pKYC config
UAT	Wks 20–26	BSA Officer time for 12 UAT scenarios · CCO available	UAT execution · Defect remediation · Sign-off docs
Shadow Mode	Wks 26–34	BSA Officer parallel review participation	Shadow run · Comparative analysis reports
Phase 2: SDD Live	Wks 34–42	BSA Officer formal authorization memo	SDD auto-clear in production · Monthly performance reports
Phase 3: Full EDD/SAR	Wks 42–52+	CCO + BSA Officer authorization · Model Validation Report received	Full pipeline live · Ongoing monitoring · Quarterly model reports

🔑 Critical Path: What Causes Most Delays

Data vendor contract cycles (8–16 weeks) and independent model validator engagement (4–8 weeks to select and contract) are the most common causes of deployment delay. Both must be initiated at project kickoff — not after engineering begins. The client dependencies in Weeks 1–12 are the critical path.

For the full responsibility matrix covering Infrastructure & Security, detailed UAT requirements, and the Pre-Go-Live sign-off checklist, see the CAIBots Production Rollout RACI document (supplied separately). Appendix B of this guide contains the complete authorization checklist.

§3 Pre-Deployment Readiness

Complete every item in this checklist before infrastructure spend begins — most deployment delays originate in weeks 1–6 client dependencies

Critical Path Warning

Data vendor contracts (8–16 weeks) and independent model validator engagement (4–8 weeks to select, contract, and schedule) are the two most common causes of deployment delay. Both must be initiated at project kickoff — not after engineering begins.

§3.1 · Data Vendor Readiness

Requirement	Owner	Typical Lead Time	Notes
Identity verification vendor selected and contracted (Jumio / Onfido / Socure / IDEMIA)	Client	4–8 weeks	CAIBots has pre-built connectors for all four. Client selects per procurement policy.
Sanctions screening feed contracted (Refinitiv / Dow Jones / LexisNexis)	Client	4–12 weeks	Client may have existing contract. CAIBots adapts to contracted feed.
Adverse media provider contracted (Dow Jones RCA / Refinitiv World-Check / ComplyAdvantage)	Client	4–8 weeks	Pre-built integrations available for all three. API credentials required.
Beneficial ownership data access confirmed (D&B ORBIS / OpenCorporates / FinCEN BOI)	Client	2–4 weeks	FinCEN BOI registry is public API — no contract. D&B / ORBIS require client subscription.
AML platform API integration confirmed with vendor (Actimize / Verafin / Fiserv)	Client	6–16 weeks	Client must initiate with their AML vendor. Vendor professional services engagement is a separate cost not included in CAIBots pilot fee.
Core banking platform sandbox API credentials provisioned	Client	2–4 weeks	Read-only access only. CAIBots never requires write access to core banking systems.

§3.2 · Regulatory & Governance Readiness

Requirement	Owner	Action Required
FinCEN BSA E-Filing account active and BSA Officer confirmed as filer of record	Client	BSA Officer must have active FinCEN credentials. CAIBots cannot be filer of record under any structure.
FinCEN 314(a) batch file delivery mechanism confirmed	Shared	Client provides access to incoming 314(a) batch. CAIBots builds 14-day response pipeline. Response window is mandatory.
Independent model validator identified and engaged (SR 11-7)	Client	Non-negotiable per SR 11-7. Oliver Wyman, Promontory, KPMG, or equivalent. Budget: $150K–$400K. Timeline to engage: 4–8 weeks.
BSA Officer formally designated and available for calibration sessions	Client	BSA Officer must sign off on all risk thresholds before any production case is processed (§6). Allow 2–4 days of focused calibration time in weeks 8–12.
AI Governance Policy drafted and submitted for board/senior management approval	Shared	CAIBots provides template. Client legal and compliance teams customise and approve.
TPRM / vendor due diligence questionnaire completed for CAIBots and Anthropic	Shared	CAIBots completes client's questionnaire and provides SOC 2 Type II, pen test summary, DPA. Allow 4–8 weeks for client TPRM review.

§3.3 · Data Asset Preparation

Closed Case Library (500+ cases minimum)

Client extracts closed KYC/AML cases with known outcomes: SDD approved, CDD approved, EDD approved with enhanced monitoring, SAR filed, account exited. Each record must include: risk score at time of decision (if available), outcome label, case type, and product line. CAIBots strips all PII before ingestion into the fine-tuning pipeline.

Format: CSV or JSON. Fields required: case_id, outcome, case_type, risk_tier, sar_filed (Y/N), edd_triggered (Y/N). PII handling: Do not include customer names, SSNs, or account numbers — CAIBots does not need them for calibration.

SAR Narrative Style Guide (50–200 examples)

Provide 50–200 redacted SAR narratives representing the institution's filing voice and typology coverage. These are used to fine-tune the SAR Narrative Agent on institution-specific language, statutory citation style, and activity description format. Quality target: ≥85% of generated narratives rated B or higher by BSA Officer on first review.

Redaction: Remove customer names, account numbers, and counterparty names. Replace with tokens: [CUSTOMER], [ACCOUNT], [COUNTERPARTY]. Preserve: amounts, dates, jurisdictions, FATF typology language, statutory citations.

§4 Infrastructure & Integration Setup

Network architecture, API integration sequence, security hardening, and deployment topology — covering weeks 1–16 of the implementation timeline

§4.1 · Deployment Topology

Option A: SaaS (Recommended for pilot)

CAIBots orchestration layer runs on CAIBots managed cloud (AWS us-east-1 or eu-west-1). Client data never leaves the client's VPC — only pseudonymised tokens and non-PII signals cross the boundary. Agent pipeline output is returned to the client's AML platform via secure API. Setup time: 2–4 weeks. No infrastructure spend required from client.

Option B: Private Cloud / On-Premises

CAIBots deployed inside client's AWS/Azure/GCP tenant or on-premises Kubernetes cluster. CAIBots provides Terraform / Helm chart templates. Client infra team provisions. Requires: GPU-capable node pool for LLM inference (or Anthropic API egress), Pinecone or pgvector deployment, Redis cluster, Neo4j AuraDB or self-hosted. Setup time: 6–10 weeks.

§4.2 · Network Security & Egress Requirements

Destination	Protocol	Port	Purpose	Direction
`api.anthropic.com`	HTTPS/TLS 1.3	443	LLM inference (SAR narrative, EDD memo, risk reasoning)	Egress only
Pinecone API endpoint	HTTPS/TLS 1.3	443	Vector RAG retrieval — regulatory corpus	Egress only
OFAC SDN API (treasury.gov)	HTTPS	443	Sanctions list — real-time screening	Egress only
Jumio / Onfido / Socure	HTTPS/TLS 1.3	443	Identity verification — biometric & document	Egress only
Dow Jones / Refinitiv World-Check	HTTPS/TLS 1.3	443	PEP screening & adverse media	Egress only
D&B ORBIS / OpenCorporates	HTTPS/TLS 1.3	443	Beneficial ownership UBO traversal	Egress only
Client AML platform (Actimize/Verafin)	HTTPS/TLS 1.3	443	Bidirectional: alert intake + case output	Bidirectional
Client core banking (read-only)	JDBC / REST	Client-defined	Transaction history — Snowflake or equivalent	Inbound to CAIBots
FinCEN BSA E-Filing	HTTPS	443	SAR/CTR filing (BSA Officer action — not automated)	BSA Officer direct
OpenText / Documentum ECM	REST / CMIS	443	Compliance file archival	Egress to ECM

All egress traffic routes through the client's API gateway. No inbound connections from CAIBots-managed infrastructure to client systems. CAIBots operates as an API client — not a server — from the client network's perspective.

§4.3 · Security Architecture

Secret Management

All API credentials stored in HashiCorp Vault (or AWS Secrets Manager / Azure Key Vault). No credentials in environment variables, configuration files, or logs. Vault leases rotated every 24 hours. Production API keys issued at go-live and rotated immediately from sandbox values. Key access logged with full audit trail.

Encryption

At rest: AES-256 for all stored data including Pinecone vectors, Redis cache, audit log. In transit: TLS 1.3 minimum for all API calls. TLS 1.2 deprecated. Certificate pinning enforced on Anthropic API and sanctions feed connections. Audit log: Tamper-evident append-only (WORM) — 5-year retention per 31 C.F.R. §1020.430.

Failover & Disaster Recovery

RTO: <4 hours for full pipeline restoration. RPO: <1 hour (Redis persistence + Snowflake point-in-time recovery). Anthropic API has multi-region failover (us-east / us-west). Pinecone has replica sets. If CAIBots pipeline is unavailable, the client's existing AML rule engine continues to operate — CAIBots is additive, not a dependency for existing regulatory obligations.

Access Control & SSO

SAML 2.0 or OIDC integration with client's identity provider (Okta / Azure AD / Ping Identity). MFA mandatory for all BSA Officer workbench access. Role-based access: BSA Officer (HITL decisions), Compliance Analyst (read-only), AML Tech Lead (configuration), CAIBots Engineer (deployment only — no access to production case data). Privileged access reviewed quarterly.

§4.4 · Integration Sequence (Weeks 1–6)

Week 1–2: Core Banking Read Access. Establish Snowflake or equivalent read-only connection to transaction history. Validate 36-month lookback query performance (<5s target). Confirm column-level masking on direct PII identifiers.

Week 2–3: AML Platform Bidirectional Connector. Connect Actimize/Verafin/Fiserv alert feed to CAIBots event ingestion endpoint. Configure case output webhook to return to client case management. Test with 10 synthetic alert payloads. Validate bidirectional data flow.

Week 3–4: External API Connections. Activate and test each vendor API in sequence: (1) identity verification, (2) OFAC SDN, (3) PEP/adverse media, (4) beneficial ownership. Validate API response times, error handling, and rate limit compliance. Load Redis cache with initial OFAC TTL (1hr) and PEP TTL (4hr) configuration.

Week 4–5: RAG Corpus Load. Load Pinecone vector store with regulatory corpus: FinCEN CDD Rule, FFIEC examination manual, FATF 40 Recommendations + 52 typology playbooks, OFAC compliance framework, FinCEN guidance letters 2012–2025. Validate retrieval accuracy on 20 test queries. Confirm MMR hybrid retrieval returns top-8 relevant chunks.

Week 5–6: Knowledge Graph & ECM. Load Neo4j with existing beneficial ownership data for all commercial accounts. Validate UBO traversal on 50 test entities. Connect OpenText/Documentum ECM write integration. Test immutable file archival with 5 synthetic CDD files. Confirm WORM storage configuration.

§5

Agent Calibration

Risk Appetite Tuning

Default weights are a starting point — not production configuration · BSA Officer sign-off mandatory before any live case

§5.1 · Risk Score Dimension Weights

Geographic

20%

10–30%

Increase for institutions with significant cross-border volume or HIFCA-designated market footprints. Decrease for purely domestic retail banks.

Product/Channel

20%

15–30%

Increase for institutions with private banking, correspondent banking, or crypto-adjacent product lines. Standard weight for basic retail/commercial.

Customer Type

25%

20–35%

Increase for institutions with high PEP exposure (international private banking), NBFI relationships, or MSB customer segments.

Behavioral

20%

15–25%

Increase for transaction-intensive businesses (MSBs, cash-heavy merchants). The primary signal for structuring detection.

Network

15%

10–20%

Increase for institutions with complex correspondent banking networks. Most computationally intensive dimension — increase only where network is a primary risk vector.

Mandatory BSA Officer Action

Risk score thresholds and dimension weights must be reviewed and signed off by the designated BSA Officer before any production case is processed. This sign-off constitutes the institution's documented risk appetite calibration under SR 11-7 model governance. Document the rationale for any deviation from default weights — examiners will ask.

§6

BSA Officer Workflow

Mandatory HITL Gates

These five decisions are hardcoded as mandatory HITL gates — they cannot be bypassed, automated, or delegated to the agent under any circumstances

RequiredEDD Determination

Review the complete evidence packet. Choose: (A) Approve onboarding with enhanced monitoring, (B) Request additional documentation — specify what, (C) Decline and initiate exit. Document rationale in the case file. Cannot proceed without this decision.

RequiredSAR Filing Approval

Review the complete SAR narrative draft. Verify: activity description is accurate and complete, subject identification is correct, transaction schedule matches raw data, statutory citation is appropriate. Sign and submit via FinCEN BSA E-Filing. The agent never submits a SAR — the BSA Officer is always the filer of record.

RequiredCustomer Exit / Account Closure

Authorize the exit letter and confirm the documented rationale. Ensure exit does not constitute a tipping-off violation under 31 U.S.C. §5318(g)(2). Coordinate with relationship manager if required. Log the decision with full rationale in the case file.

RequiredSanctions Hit Resolution

On any OFAC SDN hit: immediately block the transaction/account, notify CISO and Legal within 1 business hour. Initiate OFAC blocking order. File OFAC report within 10 business days. Document all actions with timestamps. The agent executes the block — the officer manages the regulatory reporting.

AdvisoryCorrespondent De-Risking

Review the correspondent bank risk assessment and de-risking recommendation. Consider nested correspondent chain risk, SWIFT routing path exposure, and HIFCA designation status. Decision is advisory — the BSA Officer's judgment governs. Recommended: consult with Legal for significant relationships.

§7 BSA Officer Workflow — Detailed Procedures

Step-by-step operating procedures for BSA Officers — SAR filing, case escalation, overrides, and the FinCEN BSA E-Filing submission sequence

§7.1 · SAR Filing Procedure — Step by Step

Workbench notification. BSA Officer receives a HITL notification in the CAIBots workbench. The case badge shows: typology classification (e.g., FATF T-04 Structuring), risk score, 30-day SAR clock countdown, and pre-populated evidence packet. The clock starts on the date suspicious activity was identified by the agent — not the date the officer receives the case.

Evidence review. Officer reviews the complete evidence packet: transaction schedule (dates, amounts, counterparties), FATF typology match rationale, beneficial ownership diagram, adverse media score, and the agent-generated SAR narrative draft. All regulatory citations are chunk-level — click any citation to see the source regulatory text.

Narrative review and editing. Officer reviews the draft SAR narrative for accuracy: (a) activity description correctly characterises the suspicious conduct; (b) subject identification fields are complete and accurate; (c) transaction schedule matches raw transaction data; (d) statutory citation is correct (31 U.S.C. §5318(g) for banks; 31 U.S.C. §5318(g) + §5330 for MSBs). Officer edits directly in the workbench — all edits are tracked with timestamps.

FinCEN BSA E-Filing submission. Officer clicks "Prepare SAR for Filing." The workbench pre-populates FinCEN Form TD F 90-22.47 fields from the case file. Officer reviews all fields in the FinCEN BSA E-Filing portal, makes any final adjustments, and submits. The filing confirmation number is automatically captured and logged to the case file. Note: The CAIBots system never accesses the FinCEN BSA E-Filing portal. The officer submits directly.

Post-filing logging. Officer records the BSA E-Filing confirmation number in the CAIBots workbench. The case status updates to "SAR Filed." The 30-day clock is cleared. If this is a continuing suspicious activity case, the system sets a 90-day reminder for the next continuing SAR filing evaluation.

§7.2 · CTR Filing Procedure

CTR obligations arise when a single cash transaction — or multiple same-day cash transactions at the same institution — exceed $10,000. The CAIBots Transaction Monitoring Agent automatically identifies CTR-qualifying events and pre-populates FinCEN Form 104.

CTR Auto-Population

The FATF-52TM agent calculates the total qualifying cash amount, identifies the customer and account, and pre-populates CTR Form 104. The BSA Officer reviews and submits within 15 calendar days of the triggering transaction. The workbench shows a 15-day filing clock alongside any open CTR obligation.

Structuring Alert vs. CTR Obligation

If the system detects transactions structured to avoid the $10,000 threshold — multiple transactions just below $10,000 across a short window — the FATF-52TM agent flags a structuring alert (FATF T-04 typology) and evaluates SAR obligation under 31 U.S.C. §5324. The structuring alert and any CTR obligation are handled as separate HITL workbench items.

§7.3 · Case Escalation Path

Trigger	Escalates To	Timeline	Required Action
OFAC SDN exact match — any customer or transaction counterparty	CISO + Legal + CCO	Within 1 business hour of agent flag	Immediate account block. OFAC blocking order initiated. 10 business day OFAC report. Voluntary self-disclosure evaluation.
PEP with adverse media score >70 + EDD recommendation	CCO + Relationship Manager	Within 24 hours	CCO reviews de-risking recommendation. RM coordinates customer communication. No tipping-off under 31 U.S.C. §5318(g)(2).
SAR with potential law enforcement referral indicators (terrorism finance, human trafficking)	CCO + Legal + Law Enforcement Liaison	Immediate — before SAR filing	Legal reviews for voluntary referral to law enforcement. 314(b) voluntary information sharing evaluated. SAR filed within 30-day window.
314(a) match confirmed	BSA Officer + CCO	Within 14-day FinCEN response window	Mandatory search of accounts and transactions. FinCEN 314(a) response submitted. Match → contact FinCEN. No match → no response required.
BSA Officer override rate exceeds 15% in any rolling 30-day period	MRM Committee + CAIBots	Next monthly review	Calibration review. Agent scoring analysis. BSA Officer rationale review. Threshold recalibration if warranted.

§7.4 · Override Workflow

When a BSA Officer disagrees with an agent determination, the override workflow captures the decision for model governance:

Override Documentation

Officer selects "Override Agent Finding" and completes a structured form: (1) which agent finding is being overridden, (2) the officer's determination, (3) the reason for disagreement (structured dropdown: incorrect data, stale information, institution policy, regulatory interpretation, other). The override is timestamped and immutably logged alongside the original agent output.

Override Feedback Loop

All overrides are reviewed in the monthly BSA/AML operations meeting. Patterns are analysed: if overrides cluster around a particular typology, dimension, or data source, this signals a calibration opportunity. Override rate >15% in any dimension triggers a mandatory calibration review by CAIBots and BSA Officer before the next monthly cycle.

§9

Regulatory Examination Readiness

What Examiners Will Ask

OCC · Federal Reserve · FDIC · NCUA · FinCEN · DOJ — prepare these responses before the examination

Who makes the compliance decisions?

The human BSA Officer makes every material compliance decision. The agent prepares evidence and drafts documentation. No SAR is filed, no account is exited, no sanctions hit is resolved without explicit BSA Officer authorization. This is documented in the HITL policy and demonstrable in the audit trail for every case.

How do you validate that the AI is accurate?

Through continuous operational monitoring (§8.1), quarterly model performance reviews, and annual independent model validation under SR 11-7. All metrics are tracked with documented targets. All material calibration decisions are made by the BSA Officer and documented with rationale.

How is the system's reasoning explainable?

Every determination is traceable to: a specific data input (with source tool), a specific regulatory citation (chunk-level RAG citation), and a specific reasoning step. The full agent output JSON is archived with every case file and reproducible on demand within 24 hours of examiner request.

What happens when the system is wrong?

BSA Officers can override any agent finding. All overrides are logged with rationale. Override patterns feed the monthly calibration review. A BSA Officer "Disagree with Agent" workflow flags cases for model improvement. The system cannot override a human — only the reverse.

How is the model governed after go-live?

Through the SR 11-7 model governance framework: continuous operational monitoring (§8), quarterly back-testing, quarterly performance reports delivered by CAIBots, and annual independent model validation by a third-party firm. Material model changes require 30-day advance notice and BSA Officer sign-off before deployment. The full governance trail — Model Development Documents, Model Validation Report, change register — is maintained and available within 24 hours of examiner request.

What is the model change management process?

CAIBots provides 30 calendar days advance notice of any material model change (agent logic, scoring weights, LLM version). The institution's MRM function reviews and must approve before production deployment. Patch updates (bug fixes, regulatory corpus updates) receive 7-day notice. All changes logged in the quarterly model change register. No change goes live without written client approval.

What is your vendor concentration risk posture?

CAIBots is additive — not a dependency for existing regulatory obligations. If CAIBots is unavailable, the client's existing AML platform (Actimize/Verafin) continues to operate. The fallback process — manual analyst processing reverts to pre-CAIBots workflow — must be documented in the HITL policy and tested annually in the business continuity plan. RTO is <4 hours for full pipeline restoration.

What documents should be available on Day 1 of an examination?

Model Development Documents for all 7 agents · Current Model Validation Report (or engagement letter) · HITL policy (BSA Officer + CCO approved) · AI Governance Policy (board/senior management approved) · Quarterly performance reports (last 4 quarters) · Back-testing results (last 4 quarters) · BSA Officer certification records · Audit trail extracts for 10 sampled cases · Vendor due diligence files for CAIBots and Anthropic · Data Processing Agreements with all API vendors · OFAC compliance policy update · SAR filing procedure update.

§11

Model Risk Management

Back-Testing & SR 11-7 Framework

Back-testing must be performed quarterly and documented for SR 11-7 model governance

Risk Score Rank-Ordering

Quarterly

Rank all cases in prior quarter by model risk score. Confirm SAR-filed cases and EDD escalations concentrated in upper score deciles. Pass: Gini coefficient >0.65

SDD Auto-Clear Accuracy

Quarterly

Of all SDD auto-cleared cases (score <30), identify any requiring SAR/EDD within 90 days. Pass: false negative rate <1%. Any breach triggers mandatory threshold recalibration.

SAR Narrative Quality

Monthly

BSA Officer rates 20 randomly sampled SAR drafts: A=file-ready no edits, B=minor edits, C=significant revision, D=incorrect. Pass: >85% in A or B. Two consecutive months below 80% triggers mandatory re-tuning cycle.

OFAC False Negative Test

Monthly

Inject synthetic SDN-matching entities into screening pipeline. Pass: 100% detection rate. Any miss is a Severity 1 incident in test or production.

Adverse Media Recall

Quarterly

Measure recall on held-out entities with known enforcement actions. Pass: >92% recall. Precision tracked but secondary to recall for compliance purposes.

pKYC Drift Detection

Quarterly

Compare agent risk score recalculation vs. independent manual reassessment on sample. Pass: mean absolute error <8 risk score points.

§11 Regulatory Clocks & Filing Automation

Every mandatory filing deadline tracked in the CAIBots workbench — no clock starts without the BSA Officer seeing it; no clock expires without an escalation

Zero Tolerance — Missed Deadlines Are Severity 1 Incidents

Any missed SAR, CTR, OFAC, or 314(a) deadline is an immediate Severity 1 incident requiring CCO and Legal notification within 1 business hour. The CAIBots workbench provides T-14 day escalation alerts and daily countdown for all open obligations. The system tracks — humans are responsible.

§11.1 · Mandatory Filing Clock Summary

Obligation	Deadline	Trigger	CAIBots Automation	Human Action Required
SAR Filing 31 U.S.C. §5318(g) · 31 C.F.R. §1020.320	30 calendar days from date suspicious activity identified	FATF-52TM typology match · AML rule engine alert · BSA Officer discretionary	30-day countdown clock visible in workbench from day of identification. T-14 alert. T-7 escalation to CCO. SAR narrative pre-drafted and evidence packet assembled.	BSA Officer reviews narrative, verifies all fields, submits via FinCEN BSA E-Filing. Agent never submits. Records confirmation number.
SAR — Unknown Subject Extension 31 C.F.R. §1020.320(b)(1)(ii)	60-day extension if subject cannot be identified	BSA Officer documents inability to identify subject within 30-day window	Workbench allows BSA Officer to log extension with rationale. Clock resets to 60-day countdown. Extension reason documented in case file.	BSA Officer documents extension rationale. Must exhaust identification efforts within 60-day window.
Continuing Activity SAR FinCEN SAR Guidance 2012	90-day cycles for ongoing suspicious activity	SAR filed on a subject with continuing suspicious activity	After SAR filing, system sets 90-day continuing activity reminder. Agent monitors customer for continued suspicious pattern. Auto-flags at day 75 for BSA Officer review.	BSA Officer evaluates whether suspicious activity is continuing. Files continuing activity SAR if warranted within each 90-day cycle.
CTR Filing 31 U.S.C. §5313 · 31 C.F.R. §1010.311	15 calendar days after qualifying cash transaction	Cash transaction >$10K single or aggregated same-day · FATF-52TM CTR threshold event	15-day countdown clock. CTR Form 104 auto-populated with customer data, amount, date, account. BSA Officer notified same day as trigger event.	BSA Officer reviews pre-populated Form 104. Submits via FinCEN BSA E-Filing within 15 days. Exempt person designation available — BSA Officer decision only.
OFAC Blocking Report 31 C.F.R. §501.603	10 business days from date of block	OFAC-SANC agent SDN exact or fuzzy match confirmed by BSA Officer	Immediate account block executed on BSA Officer confirmation. 10-business-day clock starts. Blocking order report pre-drafted with: blocked property description, OFAC list citation, blocking date, customer information.	BSA Officer confirms block, reviews report, submits to OFAC. Annual blocked property report filed by June 30 each year for any property remaining blocked.
FinCEN 314(a) Response 31 C.F.R. §1010.520	14 calendar days from FinCEN batch receipt	FinCEN 314(a) batch file received and processed by pipeline	14-day clock starts on batch receipt. System performs mandatory search of all accounts and transactions (5-year lookback). Match results surfaced to BSA Officer workbench. T-7 escalation alert.	BSA Officer reviews match results. Match → contact FinCEN using designated contact number. No match → no response required. Response documented in case file.

§11.2 · Workbench Clock Dashboard

The BSA Officer workbench displays a real-time compliance clock dashboard showing all open filing obligations sorted by days remaining. Color coding: Red = <7 days · Amber = 7–14 days · Green = >14 days. Any clock in red state triggers an automated CCO notification. Any Severity 1 incident (clock at 0 without filing confirmation) triggers immediate escalation protocol per §8.

§11.3 · Escalation Protocol at T-14 Days

T-14 Days: First Escalation

Automated workbench alert to assigned BSA Officer. Email notification to BSA Officer and BSA Officer supervisor. Case re-prioritised to top of workbench queue. No action required from CCO at this stage unless BSA Officer is unavailable.

T-7 Days: CCO Notification

Automated email to Chief Compliance Officer and Legal. Daily workbench reminders to BSA Officer. If BSA Officer is on leave or unavailable, the CCO must designate a qualified alternate — this must be pre-documented in the HITL policy before go-live.

T-0: Severity 1 Incident

If a filing clock reaches zero without a confirmed submission: immediate Severity 1 incident declaration. CCO and Legal notified within 1 business hour. Root cause analysis within 24 hours. Evaluate whether voluntary disclosure to the relevant regulator is appropriate. Document the failure and remediation in the MRA tracker.

§12

PII Engineering

GDPR & Data Residency Architecture

Where PII flows · Where it does not · LLM inference boundary · GDPR controls · Cross-border transfers

PII Boundary Architecture — The Core Design Principle

The CAIBots platform is architected as a PII-minimizing system. No customer PII is stored by CAIBots and no PII is transmitted in LLM inference prompts. SAR prompts contain transaction patterns (amounts, dates, counts), entity type classification, jurisdiction, FATF typology, and statutory references — customer names and account numbers are NOT included in prompts.

Customer Database (Snowflake)

PII Present

Source of truth. CAIBots has read-only access with column-level masking on direct identifiers (SSN, DOB, full account number). Only non-PII tokens (customer_id, amount, jurisdiction_code, risk_tier) transmitted to the agent pipeline.

Agent Orchestration Layer

Pseudonymised

All customer references use internal customer_id tokens. No name, address, SSN, or direct identifier in orchestration messages. Entity resolution happens at output rendering only, within the institution network boundary.

OFAC / Sanctions API Calls

Name Only

Only entity name and jurisdiction transmitted for screening. No account numbers or SSNs. Equivalent to a standard database search — covered by existing vendor DPAs.

LLM Inference (Anthropic API)

No Direct PII

SAR prompts contain: transaction patterns (amounts, dates, counts), entity type classification (MSB, LLC), jurisdiction, FATF typology, statutory references. Customer names and account numbers are NOT included in prompts. SAR narrative generated from structural case data, not raw customer PII.

Vector RAG (Pinecone)

No PII

Regulatory documents only: FinCEN guidance, FATF reports, FFIEC manual, internal policy. No customer data stored. Closed cases used for fine-tuning are PII-stripped before ingestion (§5.2).

Knowledge Graph (Neo4j)

Pseudonymised

Nodes use customer_id and entity_id tokens. Entity names stored only for external companies and public officials — already public information.

ECM Archive (OpenText)

Full Case PII

Final CDD/EDD/SAR file contains full PII as required by BSA record retention. Within institution ECM infrastructure. CAIBots writes the file; institution owns and retains it per 31 C.F.R. §1010.430 (5-year minimum).

Audit Log

Metadata Only

Contains: event_id, timestamp, agent_name, action_type, output_hash. No customer data. Tamper-evident and append-only.

§13.2 · GDPR Cross-Border Transfer Mechanism

For institutions in the UK, EU, or EEA processing data through the Anthropic API (US-based), the following transfer mechanism applies:

Standard Contractual Clauses (SCCs)

CAIBots executes EU Standard Contractual Clauses (Module 2: Controller-to-Processor) with Anthropic for any EU/EEA customer data processed through the LLM inference layer. As noted in §12, no direct PII is transmitted to the LLM — only pseudonymised structural case data. Anthropic's DPA and SCCs are available upon request. Supplementary measures include: TLS 1.3 encryption in transit, pseudonymisation at the orchestration layer, and no LLM training on customer data.

Right to Erasure — BSA Conflict Resolution

GDPR Article 17 right to erasure conflicts with BSA 31 C.F.R. §1010.430 5-year retention obligation. Resolution: the BSA retention obligation supersedes erasure requests for AML case records. The institution should document this conflict in its GDPR Records of Processing Activities (RoPA) and in the customer-facing privacy notice. CAIBots pseudonymises case records at the orchestration layer — the ECM archive containing PII is owned by the institution and subject to its retention policy.

DPIA Requirement (GDPR Article 35)

Using AI to systematically process personal data for compliance decisions almost certainly triggers a Data Protection Impact Assessment (DPIA) requirement under GDPR Article 35. CAIBots provides a DPIA template covering: processing purposes, data flows, necessity and proportionality analysis, risk assessment, and mitigation measures. The DPIA must be completed by the institution's Data Protection Officer before go-live in EU/EEA jurisdictions.

Data Retention Schedule

CDD/EDD/SAR case files: 5 years minimum per 31 C.F.R. §1010.430. CTR records: 5 years. Agent audit logs (metadata only): 5 years (append-only, no deletion). Redis cache: TTL-based — OFAC 1hr, PEP 4hr, no persistent PII. RAG corpus (Pinecone): No PII — regulatory documents only, retained indefinitely. Fine-tuning dataset: PII-stripped before ingestion, retained for model governance purposes.

§8

Ongoing Operations

Governance & Operational Monitoring

Continuous monitoring obligations post go-live — reviewed in monthly BSA/AML operations meeting

SDD Auto-Clear Rate

>70% low-risk retail

If <65%: review risk score thresholds. If <60%: mandatory calibration review before next monthly cycle.

EDD False Positive Rate

<15%

If >20%: recalibrate risk dimension weights. Excessive false positives indicate over-sensitivity in scoring model.

SAR Draft Quality (A+B)

>85%

If <80%: trigger re-tuning cycle on SAR narrative model. Pause SAR draft use until threshold restored.

OFAC False Negative Rate

0% — Always

Any OFAC false negative is a Severity 1 incident. Immediate escalation to CCO, CISO, and Legal. Root cause analysis within 24 hours.

Pipeline Latency (CDD)

<90 seconds

If >120s average: investigate API bottlenecks. Scale orchestration layer if sustained above threshold.

HITL Decision Time (EDD)

<55 min mean

If >90 minutes: investigate workbench bottleneck or analyst capacity. Target is agent-enabled — delays indicate workflow issue.

Regulatory Deadline Breaches

0 — Always

Any missed SAR/CTR/OFAC/314(a) deadline is a Severity 1 incident. Immediate CCO and Legal notification.

Appendix B

Pre-Go-Live Authorization Checklist

All items below must be checked and the responsible party documented before any production traffic is processed

Infrastructure

All external API credentials provisioned and tested (Jumio/Onfido, OFAC, Dow Jones, D&B ORBIS)

AML platform bidirectional connector live and validated with 10 synthetic test cases

Snowflake transaction history query performance validated (<5s for 36-month lookback)

Neo4j graph loaded with existing beneficial ownership data for all commercial accounts

Pinecone regulatory corpus loaded, embedded, and retrieval accuracy validated

Redis cache operational with correct TTL configuration (OFAC 1hr, PEP 4hr, 314a 24hr)

ECM archive (OpenText/Documentum) write connection tested with immutability confirmed

API key vault operational, keys rotated to production values, sandbox keys revoked

Audit logging confirmed: all agent actions and HITL decisions writing to immutable log

Network segmentation confirmed: api.anthropic.com accessible only from orchestration layer

Calibration

In-weights fine-tuning completed on closed case library (minimum 500 cases)

SAR narrative quality validated: >85% of BSA Officer ratings at B or higher

Risk score thresholds reviewed and signed off by BSA Officer (document reference: ____)

Risk dimension weights documented with rationale for any deviation from defaults

Parallel run completed: minimum 4 weeks, material disagreement rate <10%

False negative rate confirmed: <2% on all case types during parallel run

All 12 UAT scenarios passed with BSA Officer evaluation

Governance

AI Governance Policy adopted and approved by board or senior management

Model validation plan executed or formally scheduled within 90 days of go-live

HITL policy written and approved by BSA Officer and CCO

BSA Officer team fully certified (all passed 10-case certification test with >90% accuracy)

Third-party vendor due diligence completed for CAIBots, Anthropic, and all API vendors

Data Processing Agreements executed with all API vendors

FinCEN 314(a) response procedure updated to reflect agent-assisted search workflow

OFAC compliance policy updated to reflect 1-hour TTL cache protocol

SAR filing procedure updated to reflect agent-drafted narrative with mandatory BSA Officer approval

Model Risk & Validation

Independent model validation plan executed or formal engagement confirmed with validator (Oliver Wyman / Promontory / KPMG) — Phase 3 (EDD/SAR) must not proceed without Model Validation Report received

OFAC false negative incident response runbook documented and tested (who to call, timeline, voluntary disclosure evaluation)

Exam document inventory prepared — Model Development Documents, HITL policy, AI Governance Policy, quarterly performance reports, back-testing results, DPAs, and BSA Officer certification records are ready to provide within 24 hours of examiner request

Sign-Off — All Required Before Production Traffic

BSA Officer signature obtained — Date: _____________

Chief Compliance Officer signature obtained — Date: _____________

AML Technology Lead signature obtained — Date: _____________

CISO signature obtained — Date: _____________

Data Governance Officer signature obtained — Date: _____________

Legal / Regulatory Counsel acknowledgment obtained — Date: _____________

Ready to Begin Implementation?

CAIBots implementation engineers are available for a 30-minute architecture scoping session. We will map this pipeline to your specific AML platforms, core banking systems, and compliance team workflow. A scoped 90-day production pilot typically begins within two weeks of the architecture call.

contact@caibots.com · +1 (609) 721-2815 · caibots.com · Princeton, NJ

§14 Platform Scope & Extension Roadmap

Current production scope, what is explicitly out of scope, and the expansion roadmap for institutions moving beyond the KYC/AML pilot

§14.1 · Current Production Scope (v2.1)

✓ In Scope — v2.1 Production

Retail, commercial, and institutional customer onboarding (CDD/EDD/SDD)
Perpetual KYC — event-driven monitoring and risk score refresh
AML transaction monitoring — 52 FATF typologies, SAR obligation assessment
OFAC SDN screening — real-time, exact and fuzzy match, 2-hop PEP network
FinCEN 314(a) batch match and 14-day response pipeline
SAR narrative generation — FinCEN-compliant, BSA Officer review and filing
CTR auto-population — Form 104 pre-fill, 15-day clock tracking
Correspondent banking EDD — SWIFT chain risk, nested correspondent, HIFCA
CTA 2024 BOI compliance — FinCEN BOI registry, non-compliance flag
BSA Officer HITL Workbench — all 5 mandatory HITL gates
SR 11-7 model governance — MDD, back-testing, quarterly performance reports

✗ Out of Scope — v2.1

Mortgage origination KYC (Horizon 2 roadmap)
Wealth management / private banking suitability (Horizon 2)
Crypto / digital asset AML — blockchain transaction monitoring (Horizon 3)
Trade finance AML — documentary credit, TBML full pipeline (Horizon 3)
Multi-jurisdiction simultaneous filing (EU AMLA + FinCEN)
Real-time payment fraud detection (separate capability)
Customer-facing identity verification UI (client's onboarding channel)
Account opening workflow and decisioning engine
Direct law enforcement reporting (via client's existing process)

§14.2 · Extension Roadmap

Horizon	Capability	Target	Key Additions
Horizon 2 12–18 months post-pilot	Mortgage KYC · Wealth Management · Multi-jurisdiction pKYC	H2 2026 – H1 2027	Mortgage origination CDD/EDD pipeline. UHNW PEP screening enhancement. Simultaneous FinCEN + FCA/BaFin/AMLA output generation.
Horizon 3 18–30 months	Crypto/Digital Asset AML · Trade Finance TBML · Predictive Risk	2027	On-chain transaction monitoring (Chainalysis / Elliptic). Documentary credit TBML detection. Predictive risk — flags emerging risk before typology trigger fires.
Horizon 4 Platform phase	Cross-Institution Network · Real-Time Regulatory Change Management	2028+	314(b) voluntary sharing network. Automated regulatory change ingestion — when FinCEN publishes new guidance, RAG corpus updates within 24 hours.

§14.3 · Institution Expansion Path

Phase 1 — Pilot (Weeks 1–52): Retail CDD/SDD auto-clear + OFAC screening. Shadow mode, then SDD live, then full EDD/SAR pipeline. 90-day pilot structure per the CAIBots Pilot Structure document.

Phase 2 — Full Production (Months 13–18): Commercial banking EDD. Correspondent banking SWIFT chain risk. 314(a) automated response. Full pKYC across all risk tiers. SAR narrative in production use.

Phase 3 — Enterprise Platform (Months 19–30): Additional business lines (mortgage, wealth, crypto). Multi-jurisdiction regulatory output. Fine-tuning refresh on updated closed case library (annual). Horizon 2 capabilities as available.

Appendix A — Integration API Reference

REST endpoint catalog, authentication, event schemas, and webhook configuration for AML technology leads and integration engineers

API Key Security

All API keys are provisioned through the CAIBots implementation team during the Week 1–6 integration phase. Keys are environment-scoped (sandbox / staging / production). Production keys are issued at go-live and must be stored in your secret management system (HashiCorp Vault / AWS Secrets Manager). Never share keys across environments.

A.1 · Authentication

API Key (Request Header)

Authorization: Bearer <your-api-key>
Content-Type: application/json
X-CAIBots-Environment: production

All requests require HTTPS/TLS 1.3. HTTP is rejected. Rate limit: 100 requests/minute per key. 429 responses include a Retry-After header.

Webhook Signature Verification

X-CAIBots-Signature: sha256=<hmac-hash>
X-CAIBots-Timestamp: <unix-timestamp>

Verify HMAC-SHA256 signature. Reject webhooks older than 300 seconds. Return HTTP 200 within 10 seconds or CAIBots retries with exponential backoff (3 attempts).

A.2 · Core Endpoints

Endpoint	Method	Purpose	Key Fields
`/v2/cases/submit`	POST	Submit a KYC/AML trigger event to the pipeline	`trigger_type` (onboarding\|pkyc\|transaction), `customer_id`, `event_data`, `priority`
`/v2/cases/{case_id}/status`	GET	Poll case status and retrieve completed output	Response: `status`, `risk_score`, `dd_level`, `hitl_required`
`/v2/cases/{case_id}/output`	GET	Retrieve full evidence packet, SAR draft, regulatory citations	Returns: evidence JSON, SAR narrative, deadlines[], HITL gate status
`/v2/hitl/{case_id}/decision`	POST	Submit BSA Officer HITL decision	`decision` (approve\|escalate\|decline\|request_docs), `officer_id`, `rationale`
`/v2/screening/ofac`	POST	On-demand OFAC SDN screening	`entity_name`, `entity_type`, `jurisdiction`. Returns: `match_status`, `match_score`
`/v2/audit/{case_id}/log`	GET	Retrieve immutable audit trail for examiner request	Chronological event log: timestamps, agent actions, source calls, HITL decisions, output hashes
`/v2/metrics/performance`	GET	Retrieve operational metrics for governance reporting	Params: `from`, `to`, `metric_type` (sdd_rate\|sar_quality\|false_negative\|latency)

A.3 · Webhook Events

Event Type	Trigger	Key Payload Fields
`case.pipeline_complete`	All 7 agents complete — ready for BSA Officer review	`case_id`, `risk_score`, `dd_level`, `hitl_required`, `deadlines[]`
`case.hitl_required`	Case routed to mandatory HITL gate	`case_id`, `hitl_type`, `deadline_date`, `clock_days_remaining`
`deadline.alert_t14`	T-14 day filing obligation alert	`case_id`, `obligation_type` (SAR\|CTR\|OFAC\|314a), `days_remaining`
`deadline.alert_t7`	T-7 day alert — CCO notification triggered	`case_id`, `obligation_type`, `deadline_date`, `cco_notified`: true
`ofac.block_executed`	OFAC SDN match confirmed and account blocked	`case_id`, `blocked_entity`, `ofac_list`, `block_timestamp`, `report_deadline`
`model.performance_alert`	Operational metric breaches threshold	`metric_type`, `current_value`, `threshold`, `action_required`

A.4 · Error Codes

Status	Code	Meaning	Action
400	`INVALID_PAYLOAD`	Required field missing or malformed	Check `error_detail` field for specific field name. Validate against schema.
401	`AUTH_FAILED`	API key invalid, expired, or wrong environment	Verify environment scope (sandbox vs. production). Contact CAIBots implementation team.
429	`RATE_LIMITED`	100 req/min limit exceeded	Implement exponential backoff. Use `Retry-After` header value.
503	`PIPELINE_UNAVAILABLE`	Agent pipeline temporarily unavailable	Retry after 60 seconds. Client AML platform continues operating independently during downtime.