Loading model card data…

SR 11-7 Compliant · OCC Guidance · BSA/AML Typologies

Clerk AML Compliance Engine

Model transparency report for bank vendor due diligence. All metrics derived from live system data.

Model Version 1.0.0-rules

Report Generated —

Transactions Analyzed —

Data Governance Tier Minimal

Contents

1 Model Overview 2 Training Data 3 Architecture 4 Performance Metrics 5 Fairness Assessment 6 Limitations 7 Monitoring 8 Human Oversight

Model Overview

Model Name	—
Version	—
Model Type	—
Purpose	—
Regulatory Alignment	—
AI Risk Tier	Significant / High — BSA/AML affects consequential decisions & regulatory reporting

Intended Use Cases

Out of Scope

Training Data

Total Transactions

—

Analyzed in system

Unique Accounts

—

Customer entities

Data Type

Synthetic

No bank customer data

Governance Tier

Minimal

Zero retention

Date Range	—
Data Sources	—
Production Data Used	No — synthetic only
Training on Bank Data	Never
Prompt / Output Retention	Zero retention
Processing Location	US-only

Architecture & Detection Rules

Model Type	—
Approach	—
Risk Scoring	—
Decision Boundary	—
Explainability	—
ML Roadmap	—

Performance Metrics

Precision

—

Of reviewed flags

Recall

N/A

Requires labeled negatives

F1 Score

N/A

Requires recall

FP Rate (Reviewed)

—

Of human-reviewed flags

Flag Rate

—

Of all transactions

Human-Reviewed

—

Flags evaluated by humans

Confusion Matrix (Human-Evaluated)

Based on compliance investigator reviews. FN unknown — no labeled negative ground truth.

Predicted

Actual

Flagged

Not Flagged

Suspicious

—

Confirmed

Unknown

Legitimate

—

Dismissed

—

Not flagged

Alert Volume by Type

Detection Rule Contribution (SHAP-Equivalent Feature Importance)

Importance score = 60% flag frequency + 40% average risk score. Higher = stronger signal for AML detection.

Fairness Assessment

Overall Status:

Not yet run

Methodology: Disparate impact ratio across geographic, account-type, and transaction-size segments. A segment fails if its false-positive-rate proxy exceeds 2× the baseline. Using synthetic data — this framework demonstrates production-ready bias testing infrastructure per MCB AIAUP requirements.

No fairness report generated yet.
Run from Compliance Dashboard → Model Card tab → Run Analysis.

Limitations & Known Weaknesses

Monitoring & Drift Detection

PSI Warning

0.10

Review triggered

PSI Critical

0.25

Retrain triggered

FPR Warning

30%

Sustained 7d

Current Drift

—

Drift Method	—
Monitoring Cadence	—
Retrain Trigger	—
Audit Trail	—
Log Retention	—

Human Oversight (HITL)

HITL Policy	—
Review Workflow	—
Kill Switch	—
Override Logging	—

Mandatory Human Actions

Vendor Contract Requirements

Data Governance Tier	Minimal
Training on Customer Data	Never
Processing Location	US-only
Audit Rights	Available upon request (10 business days notice)
Change Notification	30-day notice for material changes + rollback capability
Subprocessors	—
DPA / BAA	—

Attestation

This model card was generated from live system data on —. All metrics reflect the current production state of the Clerk AML Compliance Engine. For questions, contact your Clerk account representative or email compliance@vaultmind-6.polsia.app.