Governance is the structure for responsibly building and operating AI systems. This page covers frameworks, model cards, and policies.

1. Model Card Template

# Model Card for [Model Name]

## Model Details
  - Model name: [Name]
  - Developed by: [Organization]
  - Model date: [YYYY-MM-DD]
  - Model version: [Version]
  - Framework: [PyTorch/TensorFlow]

## Model Use
  - Primary use: [What it does]
  - Intended users: [Who should use]
  - Out-of-scope uses: [What NOT to do]

## Data
  - Training data: [Size, source, characteristics]
  - Evaluation data: [Holdout test set]

## Performance
  - Accuracy: 92.5%
  - Precision: 94.2%
  - Recall: 90.1%
  - Performance by demographic group: [Include fairness metrics]

## Limitations & Biases
  - Known limitations
  - Ethical considerations
  - Known biases and mitigation strategies

## Recommendations
  - Speed vs accuracy tradeoffs
  - Best practices for use
  - How often retrained

2. Model Risk Assessment

from enum import Enum
from datetime import datetime

class RiskLevel(Enum):
    LOW = "low"
    MEDIUM = "medium"
    HIGH = "high"
    CRITICAL = "critical"

class ModelRiskAssessment:
    """Assesses and tracks model risks"""

    def __init__(self, model_name: str):
        self.model_name = model_name
        self.risks = []

    def add_risk(self, category: str, description: str, level: RiskLevel):
        """Adds a risk"""
        self.risks.append({
            "category": category,
            "description": description,
            "risk_level": level,
        })

    def generate_report(self) -> str:
        """Generates risk report"""
        report = f"# Model Risk Assessment\n"
        report += f"Model: {self.model_name}\n\n"

        for i, risk in enumerate(self.risks):
            report += f"### Risk {i+1}: {risk['category']}\n"
            report += f"- {risk['description']}\n"
            report += f"- Level: {risk['risk_level'].value}\n\n"

        return report

# Usage:
assessment = ModelRiskAssessment("Loan Approval Model")
assessment.add_risk("Fairness", "May discriminate against protected groups", RiskLevel.HIGH)
assessment.add_risk("Data Quality", "Training data contains biased samples", RiskLevel.MEDIUM)
print(assessment.generate_report())

3. Internal AI Policy

# Internal AI Policy

## Core Principles
  1. Fairness: Minimize bias and discrimination
  2. Transparency: Explain decisions
  3. Accountability: Clear ownership
  4. Privacy: Protect user data
  5. Security: Prevent misuse

## Model Approval Process
  - Development: Model card, fairness assessment
  - Testing: Human evaluation, performance validation
  - Deployment: Audit trail, monitoring setup
  - Monitoring: Weekly checks, quarterly audits

## High-Risk Categories
  Automatic decisions affecting:
  - Credit/lending
  - Employment
  - Criminal justice
  - Healthcare

  Requires: Ethics review + human approval

## Data Governance
  - Data minimization
  - Regular quality audits
  - Retention limits
  - Anonymization where possible

4. Audit Trail

import json
from datetime import datetime

class AuditTrail:
    """Maintains complete audit trail"""

    def __init__(self, model_id: str):
        self.model_id = model_id
        self.audit_file = f"audit_trail_{model_id}.jsonl"

    def log_event(self, event_type: str, details: dict):
        """Logs an event"""
        entry = {
            "timestamp": datetime.now().isoformat(),
            "model_id": self.model_id,
            "event_type": event_type,
            "details": details,
        }

        with open(self.audit_file, "a") as f:
            f.write(json.dumps(entry) + "\n")

# Usage:
audit = AuditTrail(model_id="loan_model_v2")
audit.log_event("model_training", {"accuracy": 0.925})
audit.log_event("fairness_audit", {"demographic_parity_diff": 0.05})