Skip to content

Examiner First-Day-Letter Response Mapping

Audience: Chief Compliance Officer (CCO), AI Governance Lead, Audit & Examination Liaison
Purpose: Map the most common examiner first-day-letter request types to the framework artifacts that satisfy them.
Use this document when: A FINRA, SEC, OCC, Federal Reserve, FDIC, NCUA, CFPB, or state regulator examination is announced and the firm receives a first-day request letter (sometimes called a "request list," "exam scope letter," or "initial information request") that touches AI agents, generative AI, or model-assisted decisioning.

How to use this mapping

Examiner first-day letters typically request a discrete set of artifact categories — model inventory, validation reports, audit logs, change tickets, training records, risk assessments, policies and procedures, and incident reports. This mapping converts each examiner request type into:

  1. The framework control(s) that govern the artifact
  2. Where to find the source data in this framework, the assessment engine, or the FSI-AgentGov-Solutions companion repository
  3. Suggested evidence format to deliver to the examiner

The mapping is not a substitute for legal counsel review of the actual examiner letter. Implementation requires the firm to validate each artifact against the specific scope and date range the examiner requested. Organizations should retain a copy of the response package and the supporting evidence in WORM-configured storage for the FINRA Rule 4511 / SEC 17a-4 retention period.

Do not over-promise evidence

Only commit to producing an artifact category if the underlying control is implemented and the evidence is retrievable for the requested date range. If a control is partially implemented, document the implementation status (e.g., "Zone 3 agents only, since [date]") rather than producing partial evidence as if it were complete.

Reference letter templates examiners commonly use

The following examiner reference materials provide the canonical request-list templates that this mapping responds to. Pull the current version of the relevant document at the start of every exam — examiner request lists evolve with each cycle.

  • FFIEC IT Examination Handbook — referenced by federal banking regulators (OCC, Federal Reserve, FDIC, NCUA) for technology-risk and model-risk examinations. The Architecture, Infrastructure, and Operations and Information Security booklets contain the canonical first-day request templates. See the FFIEC IT Handbook InfoBase.
  • FINRA examination cycle and Annual Regulatory Oversight Report — current AI-specific examination focus areas are published in the 2026 FINRA Annual Regulatory Oversight Report. Cycle exam request letters reference Rule 3110 supervision and Rule 4511 books-and-records preservation.
  • OCC Comptroller's Handbook — model risk management, third-party risk management, and operational risk booklets define the procedures examiners follow. See the OCC Comptroller's Handbook. Bulletin 2026-13 (the current revision of the legacy Bulletin 2011-12 model-risk guidance) and Bulletin 2023-17 (third-party risk) frame the request scope.
  • Federal Reserve SR letters — SR 26-2 (the current revision of the legacy Fed SR 26-2 (formerly SR 11-7) model-risk guidance), SR 23-4 (third-party risk), and the Federal Reserve SR Letters index frame Reserve Bank examiner request scope.
  • SEC examination request letters — the Division of Examinations publishes annual priorities; recent priorities have referenced AI / predictive data analytics use by registrants. See the SEC Division of Examinations priorities.
  • CFPB supervisory examination procedures — the CFPB Supervision and Examinations program publishes the canonical request templates for ECOA / Reg B fair-lending examinations. See the CFPB Supervision and Examinations program.

First-day-letter request type → framework artifact map

Examiner Request Type Framework Artifact Where to Find It Suggested Evidence Format
Model inventory — list of all AI / ML / generative-AI models, agents, or "automated decisioning" systems in use, with owner, business purpose, and risk tier Control 2.6 - Model Risk Management model registry; agent inventory captured by the assessment engine Model registry export; assessment/output/<run>/assessment-prefilled.md Control 2.6 section; companion agent-inventory solution if deployed CSV / Excel: model ID, agent name, owner, business purpose, zone, risk tier, model type, deployment date, last validation date, current status
Validation reports — pre-deployment validation evidence and ongoing model-performance monitoring Control 2.6 (model validation lifecycle), Control 2.11 (fairness validation), Control 2.5 (QA validation) Per-agent validation packages stored in WORM retention; bias-testing reports per Control 2.11; QA test results per Control 2.5 PDF report packages per agent: methodology, test datasets, statistical results, sign-offs, remediation history, independent-validation attestation (Zone 3)
Audit logs / activity logs — agent invocation logs, prompt/response logs, administrative-action logs, sensitive-data-access logs Control 1.7 - Comprehensive Audit Logging, Control 3.3 - Compliance and Regulatory Reporting, Control 1.5 DLP event telemetry Microsoft Purview Audit; Microsoft Sentinel; M365 Unified Audit Log; Power Platform admin analytics; Application Insights for agent telemetry Date-range filtered audit-log export (CSV / JSON) with SHA-256 integrity manifest; cover memo describing collection methodology and retention configuration
Change tickets / release management evidence — record of agent versions, configuration changes, approvals, and deployment timestamps Control 2.1 - Managed Environments, Control 2.3 - Change Management and Release Planning Change-management system of record (ServiceNow / Jira / Azure DevOps); environment lifecycle workflow logs; release-approval Power Automate flows Change-ticket export with: ticket ID, agent / artifact ID, requestor, approver(s), pre-prod test evidence link, deployment timestamp, post-deployment validation status
Training records — staff AI / governance / compliance training completion, role-based curriculum, and re-attestation cycle Control 2.14 - Training and Awareness Program LMS export (Microsoft Viva Learning / SAP SuccessFactors / Cornerstone); attestation records; role-curriculum mapping document Roster export by role: employee ID (or anonymized handle), role, required curriculum, completion date, attestation date, current status
Risk assessments — agent-level risk assessment, risk-tiering methodology, and re-assessment cadence Control 2.6 (model-risk methodology), Control 2.11 (fairness risk assessment), Framework: Zones and Tiers (risk-tier definitions) Per-agent risk-assessment packages; risk-tiering methodology document; assessment engine output PDF risk-assessment package per in-scope agent: scoring rubric, tier rationale, sign-off, next reassessment date
Policies and procedures — written governance program, AI acceptable-use policy, model-risk policy, third-party-AI policy Framework: Governance Fundamentals, Framework: Operating Model, Framework: Regulatory Framework, Reference: RACI Matrix Firm's published policy library; framework documentation as supporting reference; RACI matrix for role responsibilities Signed and dated policy PDFs with version history; RACI matrix; cross-reference table mapping each policy section to the requesting examiner letter line
Incident reports — AI-related incidents, data-loss events, disclosure errors, model-failure events, customer-complaint pattern Control 1.21 - Adversarial Input Logging, Control 3.4 - Incident Reporting and Root-Cause Analysis Incident-management system of record; SOC ticket queue (Microsoft Sentinel / Defender XDR); customer-complaint system Incident register export for the requested date range: incident ID, classification, agent involved, detection date, containment date, root-cause summary, remediation status, regulatory-notification status
Vendor / third-party AI inventory — list of third-party AI providers, contracts, due-diligence records, and ongoing monitoring Control 2.6 (third-party model coverage), Control 2.7 - Vendor and Third-Party Risk Management Vendor inventory; contract repository; vendor due-diligence packages; ongoing performance monitoring records Vendor inventory export: vendor name, AI service description, contract date, last due-diligence review date, current risk tier, contract clauses for audit / SOC 2 / data-residency / sub-processor disclosure
Books-and-records preservation evidence — WORM retention configuration, record categorization, retention-period evidence (FINRA Rule 4511 / SEC 17a-4) Control 2.13 - Documentation and Record-Keeping, Control 1.9 - Data Retention and Deletion Policies, Control 1.7 Microsoft Purview retention-policy and label-policy export; SharePoint retention configuration; Exchange Online retention configuration Purview policy export (Get-RetentionCompliancePolicy / Get-RetentionComplianceRule); cover memo mapping each retention policy to the regulatory citation it supports
Adverse-action / fair-lending evidence (CFPB, ECOA / Reg B) — principal-reasons disclosures generated for agent-influenced credit decisions Control 2.11 — CFPB / Reg B subsection Per-decision principal-reasons audit trail (model version, feature attributions, reviewer attestation, final notice text) Sample of adverse-action notices with corresponding model-explainability artifacts and reviewer attestations for the requested date range
OFAC / BSA-AML screening evidence (transaction-touching agents) — sanctions-screening result for each agent-influenced transaction Control 1.5 — Sanctions Screening subsection Sanctions-screening platform audit log; agent decision-artifact registry showing screening-result event linkage Per-transaction screening evidence export: counterparty, screening platform request / response IDs, disposition, reviewer attestation where required
Data-loss prevention configuration — DLP policies, sensitivity-label taxonomy, and Power Platform data policies Control 1.5 - DLP and Sensitivity Labels, Control 1.13 - Sensitive Information Types Get-DlpCompliancePolicy / Get-DlpComplianceRule exports; Get-LabelPolicy exports; Power Platform Get-DlpPolicy exports PowerShell-cmdlet output exports with cover memo mapping each policy to its FINRA / SEC / GLBA citation
Access-control and identity evidence — agent identity, RBAC assignments, conditional-access policies for agent users and agents Control 1.1 - Restrict Agent Publishing by Authorization, Control 1.11 - Conditional Access and Phishing-Resistant MFA, Control 1.18 - Application-Level Authorization and RBAC, Control 2.8 - Access Control and Segregation of Duties, Control 2.26 - Entra Agent ID Identity Governance Microsoft Entra ID role assignments; Conditional Access policy export; PIM assignments Entra ID role assignment export per role; Conditional Access policy JSON export; PIM eligibility report; Entra Agent ID inventory export
Continuity / resilience evidence — agent business-continuity testing, fallback procedures, RPO / RTO documentation Control 2.4 - Business Continuity and Disaster Recovery BCP / DR plan; tabletop exercise reports; failover test evidence Most recent BCP test report per in-scope agent: scenario, RTO / RPO target, actual RTO / RPO observed, sign-off
Customer-facing disclosure evidence — disclosures that an AI agent is being used in customer interactions, where applicable Control 2.19 - Customer AI Disclosure and Transparency, Control 2.23 - User Consent and AI Disclosure Enforcement Disclosure-language registry; UI / channel screenshots; consent-management system records Disclosure-text inventory by channel with implementation date and screenshot evidence
Agent inventory — definitive list of all deployed agents with metadata, owner, and lifecycle status Control 3.1 - Agent Inventory and Metadata Management, Control 3.11 - Centralized Agent Inventory Enforcement, Control 1.2 - Agent Registry and Integrated Apps Management Centralized agent registry; M365 Admin Center agent inventory; Power Platform admin agent listing; assessment engine output Agent inventory export with owner, business purpose, zone, deployment date, last review date, lifecycle status
Governance committee minutes — AI governance / model-risk committee charter, agendas, and minutes Framework: Operating Model, Framework: Governance Cadence Committee secretariat records Committee charter PDF; agenda + minutes for the requested date range with attendance list

Response-package assembly checklist

When assembling the actual response package:

  1. Confirm scope and date range with the examination liaison before any artifact is collected. Examiner letters often request "as of" snapshots and a separate "during the examination period" range — these are different queries.
  2. Run the assessment engine against the in-scope tenant for the requested date range. The assessment/output/<run>/assessment-prefilled.md file consolidates Control 2.6 model-registry data, Control 2.11 fairness evidence, Control 1.7 audit-log readiness, and Control 1.5 DLP policy posture in one document.
  3. Produce SHA-256 integrity manifests for every evidence file delivered. Examiner staff increasingly expect verifiable file integrity for digital evidence.
  4. Cover memo — every artifact category should ship with a one-page cover memo describing: what was collected, the collection method, the date range, the data source, and the retention configuration. Do not deliver raw exports without a cover memo.
  5. Privilege review — coordinate with Legal before delivering any artifact that may contain privileged communications, work product, or material covered by attorney-client privilege. Some validation reports and incident root-cause analyses fall in this category.
  6. Retain the response package in WORM storage with the same retention period as the underlying records (FINRA Rule 4511: per record category; SEC 17a-4(f): per record category; firm-specific BSA records: minimum 5 years per 31 CFR § 1010.430).
  7. Update STATE.md / governance committee minutes to record that the examination was opened, what was requested, and what was delivered. Examiner request lists themselves are typically discoverable in subsequent exams.

Gaps to track for future framework expansion

The mapping above identifies categories that the current 78-control framework addresses. Examiner request types that fall outside the current framework — and that are candidates for a future control or playbook — should be tracked in the project backlog and surfaced in the v1.8 roadmap discussion. Examples that have appeared in recent examiner letters but are not yet first-class framework artifacts:

  • Cross-tenant agent-collaboration audit trails (where agents from one tenant are invoked by users in another)
  • Generative-AI training-data lineage and "model card" disclosures for in-house fine-tunes
  • Per-agent carbon / compute attribution (emerging supervisory interest in EU but appearing in US large-bank exam letters)

These are placeholders for future framework work, not gaps in the current response capability.

Updated: May 2026 | Version: v1.6.2 | Audience: CCO / AI Governance Lead / Examination Liaison