AI Insights Module · Governance
AI Policy Center
AI Generated Summary
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
Explain This AI Summary
Model
Policy Compliance Synthesizer v1.6 (rule-based audit over 8 active policies and 247 production models)
Data Sources
Policy registry, model deployment logs, and the last 90 days of retraining events across all 22 departments.
Why 94% confidence
Compliance status was cross-checked against policy owner sign-off records; the only unresolved item is Fraud Detection v4.1's pending Explainability Requirement re-approval, which caps confidence below full certainty.
Suggested Action Rationale
Models with an unresolved policy status carry deployment risk; resolving Fraud Detection v4.1's sign-off before its next release window keeps the fleet at 100% policy coverage.
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.
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.
Model Approval Before Production Deployment
No model version may serve production traffic until a designated approver signs off on its evaluation results.
Quarterly Bias Audit Requirement
Every production model with customer-facing impact is audited each quarter for demographic and segment bias.
Customer Notification for AI-Driven Decisions
Customers are notified when an AI model materially influences a decision made about their account.
Explainability Requirement for Regulated Use Cases
Models used in regulated decisions (credit, fraud, employment) must produce a human-readable explanation on request.
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 |
Compliance Breakdown
Across 6 flagship production models
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
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