Governance

Data and AI governance in Amazon SageMaker — SageMaker Catalog for discovery, lineage, quality, and access control.

---

Overview

Amazon SageMaker Catalog (built on Amazon DataZone) provides:

Capability	Purpose
Discovery	Find data and AI assets at scale
Governance	Control access with fine-grained permissions
Collaboration	Share assets across teams and projects
Quality	Monitor data quality and lineage
Compliance	Audit trail and responsible AI

---

SageMaker Catalog

Unified catalog for all data and AI assets:

What’s in the Catalog?

Asset Type	Examples
Structured data	Tables, databases, data products
Unstructured data	Documents, images, files
AI models	Trained models, prompts, agents
BI dashboards	QuickSight/Quick Sight reports
Applications	GenAI apps, notebooks

---

Governance Features

Data and AI Catalog

Feature	Details
Central catalog	Discover all assets in one place
Metadata management	Technical and business metadata
Asset registration	Automatic and manual registration
Search	Find assets by name, tag, or owner

Business Glossary

Feature	Details
Shared definitions	Standardize business terminology
Customizable metadata	Create metadata forms
Classification terms	Tag sensitive data consistently
Governance workflows	Enforce tagging policies

Data Lineage

Track data flow across systems:

Feature	Details
OpenLineage compatible	Industry-standard lineage format
Origin tracking	Where data comes from
Transformation history	How data changes
Consumption patterns	Who uses the data
Impact analysis	Understand downstream effects

Data Quality Monitoring

Feature	Details
Quality metrics	View metrics from AWS and third-party tools
Consumer trust	Show quality scores in catalog
API integration	Integrate external quality signals
Unified portal	Single view of data health

---

Discovery & Search

Data Discovery

Feature	Details
Business context	Enrich technical metadata with descriptions
Auto-enrichment	AI-generated metadata
Quick understanding	Help users find and trust data

Automated Metadata Recommendations

Feature	Details
LLM-powered	AI generates business-friendly names
Descriptions	Auto-generate asset descriptions
Consistency	Improve clarity across assets

Semantic Search

Feature	Details
Natural language	Search using plain English
Intent understanding	Goes beyond keywords
Context-aware	Understands relationships
Relevant results	Returns what you mean, not just what you type

---

Data Products

Package related assets into reusable products:

Feature	Details
Bundled assets	Group related tables, models, dashboards
Shared metadata	Common business descriptions
Unified access	Single subscription request
Consumption tracking	Monitor product usage
Reduced overhead	Fewer individual permissions

---

Access Control

Permission Model

Level	Control
Domain	Top-level organizational boundary
Project	Team-scoped access and collaboration
Asset	Individual table/model permissions
Column	Fine-grained column-level access

Integration with Lake Formation

• Fine-grained permissions at no extra cost
• Row and column level security
• Tag-based access control
• Cross-account sharing

---

Responsible AI

AI Governance Features

Feature	Purpose
Data classification	Label sensitive data
Toxicity detection	Identify harmful content
Guardrails	Apply responsible AI policies
Model cards	Document model behavior and limitations
Bias detection	Identify and mitigate bias

ML Lineage

Track the full ML lifecycle:

• Training data sources
• Model versions
• Experiment parameters
• Deployment history

SageMaker Model Dashboard (Brief)

SageMaker Model Dashboard is a centralized repository and single interface for tracking model governance and performance.

Coverage	Details
Model inventory	Consolidates models in your account, including outputs from SageMaker training jobs
Imported models	Supports models trained outside SageMaker and then hosted on SageMaker
Single stakeholder view	Gives IT admins, model risk managers, and business leaders one place to review model status
Cross-service signals	Aggregates data from multiple AWS services to indicate model health and performance
Deployment insights	Shows endpoint details and deployed model visibility
Batch insights	Includes batch transform job details for offline inference workloads
Monitoring insights	Surfaces monitoring job information for drift, quality, and ongoing model behavior checks

Think of it as a control panel for model inventory, deployment visibility, and risk/performance oversight.

---

Collaboration

Projects

Feature	Details
Team spaces	Isolated collaboration environments
Asset sharing	Publish and subscribe workflows
Centralized or decentralized	Flexible governance models
Self-service	Teams can request access independently

Publishing & Subscribing

Workflow	Description
Publish	Make assets available to others
Subscribe	Request access to shared assets
Approval	Data owners approve/deny requests
Audit	Track all sharing activities

---

Pricing

Component	Pricing
Catalog	Free usage tier available
Metadata storage	Per GB stored
API requests	Per request (with free tier)
Lineage	Included in standard pricing
Lake Formation	No extra cost for permissions

---

TL;DR

• SageMaker Catalog = Central catalog for data + AI assets (built on DataZone)
• Discovery = Semantic search, LLM-powered metadata, business glossary
• Lineage = OpenLineage-compatible tracking of data flow
• Quality = Unified view of data health metrics
• Access control = Fine-grained to column level via Lake Formation
• Collaboration = Projects for teams, publish/subscribe for sharing
• Responsible AI = Data classification, guardrails, bias detection, model cards
• Model Dashboard = Centralized model repository with endpoint, batch transform, and monitoring visibility

---

Resources

SageMaker Catalog 🔴
Data and AI governance capabilities.

Amazon DataZone 🔴
Underlying data governance platform.

AWS Lake Formation 🔴
Fine-grained access control.