Logo

AI Insights Module · Governance

AI Policy Center

AI Generated Summary

Confidence 94% Generated 12 min ago

247 models are governed by 8 active AI policies across Preadmin. Human-in-the-loop compliance holds at 100% for high-risk decision models, and training-data retention adherence sits at 98.6% across connected pipelines. One model — Fraud Detection v4.1 — is pending policy re-approval after a recent retraining event. Recommendation: complete the Explainability Requirement sign-off for Fraud Detection v4.1 before its next scheduled deployment window.

Business Impact

$2.1M compliance risk mitigated this quarter

8

Active AI Policies

247

Models Governed

2

Pending Approvals

98.6%

Policy Adherence Rate

Active AI Policies

6 of 8 shown

Human-in-the-Loop for High-Risk Decisions

Requires a human reviewer to confirm any AI-driven decision above a defined risk threshold before it takes effect.

Active Owner: Sarah Chen

90-Day Training Data Retention

Raw training data used by production models is purged 90 days after model training completes, unless a legal hold applies.

Active Owner: Dana Ruiz

Model Approval Before Production Deployment

No model version may serve production traffic until a designated approver signs off on its evaluation results.

Active Owner: Marcus Lee

Quarterly Bias Audit Requirement

Every production model with customer-facing impact is audited each quarter for demographic and segment bias.

Active Owner: Priya Patel

Customer Notification for AI-Driven Decisions

Customers are notified when an AI model materially influences a decision made about their account.

Draft Owner: Jordan Kim

Explainability Requirement for Regulated Use Cases

Models used in regulated decisions (credit, fraud, employment) must produce a human-readable explanation on request.

Active Owner: Ana Rossi

Model Policy Compliance

Compliance status per production model against active policies

Model Policies Applied Status
Churn Prediction v3.2 6 of 8 policies Compliant
Revenue Forecasting v2.7 5 of 8 policies Compliant
Fraud Detection v4.1 7 of 8 policies Pending
Customer Segmentation v1.9 6 of 8 policies Compliant
Demand Planning v2.3 5 of 8 policies Exception Granted
Sentiment Analysis v1.4 4 of 8 policies Compliant
6 of 247 models shown View all models

Compliance Breakdown

Across 6 flagship production models

98.6% policy adherence rate View audit trail

Model Approval Workflow

Pending sign-offs before production deployment

Fraud Detection v4.1 — Explainability Requirement re-approval

Retrained on updated transaction data; requires sign-off before the next scheduled deployment window.

Requested by Marcus Lee · 2 days ago

Demand Planning v2.3 — Bias audit exception renewal

Existing exception to the Quarterly Bias Audit Requirement expires this cycle and needs renewal or remediation.

Requested by Dana Ruiz · 5 days ago

2 pending approvals

Policy Change History

Sarah Chen updated Human-in-the-Loop for High-Risk Decisions

Added a finance approval step for transactions above $50,000.

3 days ago

Marcus Lee approved Model Approval Before Production Deployment

Signed off on Churn Prediction v3.2 for production release.

1 week ago

Priya Patel completed Quarterly Bias Audit Requirement

Closed out the Q2 bias audit for Customer Segmentation v1.9 with no findings.

2 weeks ago

Jordan Kim drafted Customer Notification for AI-Driven Decisions

Opened for review; not yet active across production models.

3 weeks ago

4 changes in the last 30 days