# Data governance

Data governance is the system of policies, responsibilities, and controls that determines how data is collected, used, protected, and managed throughout its lifecycle.

Cloud concept. Updated 2026-11-12. Page: https://blowstack.com/tools/cloud-concepts-hub/data-governance

> Data governance makes data use accountable by defining ownership, permitted purposes, quality expectations, access rules, and retention requirements. For foundation model fine-tuning, it helps ensure that training examples are authorized, relevant, representative, consistently labeled, and traceable to their sources.

Practice exams covering Data governance: [AIF-C01 AWS Certified AI Practitioner](https://blowstack.com/practice-exams/aws-certified-ai-practitioner).

## 6 key aspects of Data governance

Effective data governance combines accountable ownership, authorized use, privacy controls, reliable training data, traceability, and lifecycle enforcement.

1. **Ownership and permitted use**: Assign accountable owners and stewards, and document which uses are authorized. Verify training rights independently of whether engineers can technically access the data.
2. **Privacy and access controls**: Classify sensitive data, minimize what is collected, and restrict access according to job responsibilities. Apply encryption and appropriate redaction without assuming that either makes unrestricted reuse acceptable.
3. **Curation, size, and representativeness**: Set acceptance criteria for relevance, completeness, duplication, and coverage of production scenarios. Choose dataset size using model requirements and evaluation evidence rather than assuming that more records always improve results.
4. **Labeling and human-feedback quality**: Provide explicit instructions for target responses and preference judgments, train reviewers, and resolve disagreements. Check annotation consistency and systematic bias before using labels or RLHF feedback.
5. **Lineage and reproducibility**: Track source datasets, transformations, approvals, versions, and the training runs that consumed them. Preserve sufficient metadata to investigate problems without unnecessarily retaining sensitive source content.
6. **Lifecycle and accountability**: Define retention, deletion, exception handling, and periodic review requirements for source and derived data. Record policy decisions and audit evidence, including how data changes affect previously trained models.

## Cloud services for Data governance

Amazon DataZone provides discovery, ownership context, and access-request workflows, while AWS Lake Formation applies permissions to supported data lake resources. Amazon S3 stores restricted source data and approved training datasets, and AWS Glue supplies technical metadata, transformations, and data-quality checks. Amazon Macie helps identify sensitive information requiring review before training. Amazon SageMaker Ground Truth supports controlled labeling workflows when human annotations are needed. Together, these services support governance implementation, but engineers must still define permitted uses, approval criteria, representative sampling, retention rules, and links between dataset versions and training runs.

- **[Amazon DataZone](https://blowstack.com/tools/cloud-services-hub/amazon-datazone)**: Provides a business data catalog, ownership context, and subscription workflows for discovering and requesting access to data. It helps teams communicate approved uses and responsibilities, but catalog entries do not independently establish legal training rights.
- **[AWS Lake Formation](https://blowstack.com/tools/cloud-services-hub/aws-lake-formation)**: Supports centralized permissions and fine-grained access controls for supported data lake resources and query engines. Teams can use it to restrict which users and workloads can access governed datasets.
- **[AWS Glue](https://blowstack.com/tools/cloud-services-hub/aws-glue)**: The AWS Glue Data Catalog stores technical metadata, while AWS Glue Data Quality can evaluate explicit data-quality rules. These capabilities support dataset discovery and repeatable checks before data is approved for training.
- **[Amazon Macie](https://blowstack.com/tools/cloud-services-hub/amazon-macie)**: Discovers sensitive data in supported Amazon S3 objects and produces findings that can inform classification and review. Findings help identify training-data risks, but Macie does not automatically redact datasets or certify that they are anonymous.
- **[Amazon S3](https://blowstack.com/tools/cloud-services-hub/amazon-s3)**: Stores source and curated datasets with controls such as bucket policies, encryption, versioning, and lifecycle rules. Versioning supports dataset traceability, but teams must separately preserve transformation metadata and configure retention for current and noncurrent versions.
- **[Amazon SageMaker Ground Truth](https://blowstack.com/tools/cloud-services-hub/amazon-sagemaker-ground-truth)**: Supports human data-labeling workflows for creating training datasets. Governance still requires suitable worker access, labeling instructions, quality review, and explicit criteria for any collected human judgments.

## What is Data governance

Data governance defines who can make decisions about data and how those decisions are enforced and verified. It covers ownership, stewardship, acceptable use, quality, privacy, security, retention, and accountability. Data management performs activities such as storing, transforming, and deleting data; governance establishes the rules and responsibilities governing those activities. A catalog or access-control system supports governance, but neither replaces organizational policies and accountable owners.

In the cloud, governance commonly begins with an inventory of datasets, their owners, sensitivity classifications, source systems, and approved uses. Engineers translate policy into access permissions, encryption settings, approval workflows, retention controls, and audit records. For example, a customer-support dataset might be approved for operational reporting but not for model training. Access to that dataset does not automatically grant permission to reuse it for fine-tuning.

Preparing data to fine-tune a foundation model requires both governance decisions and technical curation. A team should verify licensing, consent where required, contractual restrictions, and whether the intended training use is permitted. It should identify sensitive information, remove unnecessary fields, and apply suitable redaction or de-identification. Masking names alone does not guarantee anonymity because combinations of remaining attributes can still identify individuals.

Curation should remove corrupted records, duplicates, irrelevant examples, and unsuitable content while preserving examples needed for the intended task. Dataset size must meet the selected model's requirements, but there is no universally correct number of examples. More data increases processing, labeling, and review costs and can worsen results when it introduces noise or bias. A smaller, carefully curated dataset can be more effective than a larger, poorly controlled collection.

Representativeness means covering the users, languages, tasks, and operating conditions expected in production, including important rare cases. Teams should document coverage gaps and evaluate performance across relevant groups. Training, validation, and test sets should be separated appropriately, often by customer, document, or time period, rather than merely by randomly splitting closely related records. This reduces leakage that would otherwise make evaluation results appear better than real-world performance.

For supervised fine-tuning, governance includes clear labeling instructions, qualified reviewers, disagreement resolution, and checks for consistent target responses. Reinforcement learning from human feedback, or RLHF, may use human preferences between model responses to train a reward model that guides subsequent optimization. These preference datasets need documented evaluation criteria, reviewer training, privacy protections, and checks for systematic bias. RLHF is not simply another name for labeling, and it is not required or supported by every fine-tuning workflow.

A practical support-assistant project might retain restricted source conversations separately from an approved, redacted training dataset. Each dataset version records its source, transformations, labeling guidelines, approvals, and associated training runs. This traceability supports investigation and reproducibility. Retention policies must also consider derived datasets and backups; deleting a source record does not automatically remove its influence from an already trained model.

Governance introduces review effort and can slow experimentation, especially when approvals are manual. Reusable policies, automated validation, and risk-based review help reduce that overhead. The goal is controlled, useful data access, not blanket restriction. Encryption and regulatory compliance are important, but neither alone establishes data quality, representativeness, or permission to train a model.
