Skip to main content

Rule Management in Data Governance: From Rule to Control

Emily Winks, Data Governance Expert, Atlan
Data Governance Expert
Updated:
|
Published:
18 min read

Key takeaways

  • A policy states intent. A rule is a testable statement tied to data elements. A control makes it happen.
  • When two rules apply at once, someone has to choose the order. Atlan documents one: explicit deny overrides grants.
  • Rules that live as versioned code are reviewed, tested, and rolled back like software.

What is rule management in data governance?

Rule management in data governance is the practice of turning a written policy into testable business rules, tying each rule to data elements and their metadata, and enforcing it with a control, automated where possible. A rule needs three things: a data scope, an owner, and a control that acts. In Atlan's AI Context Platform, a data policy takes its scope from an asset's tags, and the owners, terms, and lineage live on that same asset, so you find the rule and its context together.

On this page:

  • Policy, rule, and control, and how they differ
  • How a rule reaches data through tags, terms, and owners
  • Order, ownership, and approval when rules collide or change
  • Rules as versioned code, tested before they ship

Choosing a platform?


Rule management is how a team turns a written policy into a testable rule, ties it to data, and has a control enforce it. As an illustration, take this rule: customer email is masked unless the requester is on the support team.

A policy in a wiki guides people, but it does not change what a query returns until a control acts on it. Atlan’s data policies mask values in Atlan’s own query and preview surfaces, so a rule that must also hold on other query paths needs a matching tag-scoped control in each system that serves those queries.

Gartner published its 2026 Magic Quadrant for Data and Analytics Governance Platforms on 12 November 2026. To see what the report covers and how to get a copy, read our Magic Quadrant explainer.

Check One Action Before an Agent Gets It


Takes one business action, the system that applies it, who may request it, and the rules it depends on. Returns a ready, partial, or absent verdict on six checks, the refusal and failure cases to test first, and the smallest first release. Read the skill.

Paste into a new chat

Use the skill at https://atlan.com/skills/operational-action-readiness-check.md to check whether one business action is ready for an AI agent. Ask me for whatever it needs.

Run once in a terminal

curl -fsSL --create-dirs \
  -o ~/.agents/skills/operational-action-readiness-check/SKILL.md \
  https://atlan.com/skills/operational-action-readiness-check.md

For an agent

curl -fsSL https://atlan.com/skills/operational-action-readiness-check.md
Quick facts
What it is Writing governance rules so they are testable, tied to data, ordered, owned, and enforced
Core parts Rule statement, data scope (tags, terms, columns), owner, control, version history
Control types Masking policy, access deny, quality check, approval workflow, automation
Who owns it A named owner or group per rule, with stewards approving changes
Where rules can live In the platform next to the asset, and as versioned code in a repository
How you know it works Tests pass or fail on each run: data checks, plus a query as each affected role

How does rule management work, from policy to control?

A written policy states intent. No query result changes until a rule and a control exist behind it.

From policy to rule


“Protect customer email” can sit in a document for years while a query still returns the column in full. What is missing is a testable statement, a link to the data it covers, and a control that enforces it.

Rule management covers the whole path. You write the rule, scope it to data, set its order, name an owner, and enforce it with a control.

Why agents raise the stakes


AI agents read and act on data directly. A rule that lives only in a document does not limit what they read.

A rule is easier to enforce where the agent works. That is the argument behind governance rules that AI agents can read and behind embedded data governance, where the rule travels with the data instead of waiting in a binder. Context engineering for AI governance covers the same idea from the agent’s side.

Policy, rule, and control defined


The three terms are easy to blur, so it helps to pin each one down. NIST’s glossary uses policy in the access-control sense: the representation of rules or relationships that makes it possible to determine whether a requested access should be allowed (NIST CSRC glossary, Policy). It defines a security control as “a safeguard or countermeasure prescribed for an information system or an organization” (NIST CSRC glossary, Security control).

Atlan’s working split reads policy as intent and adds the rule as the middle step. That split is Atlan’s framing, not a NIST definition.

Term Plain meaning Example
Policy What the organization intends, set by risk and legal “Protect customer email”
Rule A testable statement tied to data elements “Email is masked unless the requester is on the support team”
Control The safeguard that makes the rule hold A masking policy, a quality check, an approval workflow

The split matters because a policy can be approved once, while rules get edited as data and use change. NIST SP 800-53 Rev. 5 is the catalog where security and privacy controls are named and organized.

A team building an AI governance framework can start by listing policies. It then writes the rules and controls beneath them. Shared evaluation criteria show whether a platform can carry all three.


What components make up a rule, and how does it reach data?

A rule reaches data through the metadata on it. A tag or glossary term lets one rule cover many assets, and an owner is the person or group who answers for it.

Scope by tag


In Atlan, you tag a column as PII and link a purpose to that tag. The purpose’s policies then cover every PII-tagged asset in scope (Atlan docs, Protect data by tag).

A purpose scopes by tag, while a persona scopes by who is asking. Masking applies where Atlan serves data: users run queries in Insights and see sample data in asset previews (Atlan docs, Data policy).

Tags that follow the data


Tags can also follow the data downstream. When propagation is enabled, a tag on a source column flows through lineage to the assets built from it. The scope follows the data to the dashboard and the derived table without anyone retagging it (Atlan docs, What are tags).

Propagation is not on by default, so it is a setting a team decides to turn on.

Terms and owners


Terms and owners ride the same metadata. The glossary term carries the business meaning the rule refers to, and the owner answers for it.

The metadata layer for AI and metadata management for AI explain why these fields matter to agents. The types of metadata for AI agents lays out which kind does which job.

For the glossary side, see semantic layer vs business glossary and managing KPI definitions in a catalog. For stored metadata versus a live layer, see active metadata vs context layer.

Control types: masking options


Once the scope is set, the control is a choice. Atlan’s data policy docs list five ways to mask a column (Atlan docs, Data policy):

  • Show first 4 characters, and replace the rest.
  • Show last 4 characters, and replace the rest.
  • Hash, a consistent hash, so joins across assets still work.
  • Nullify, which replaces the value with null.
  • Redact, which replaces letters with x and digits with 0.

For the email column, a hash keeps the join key usable and replaces the plain address in results. Because the hash is consistent, the same address always maps to the same value, so records stay linkable.

The rule owner decides whether that is acceptable for the column. For tagging many assets at once, see data classification and tag management for AI.


How do rules compare when two apply at once?

When two rules point in different directions, the result depends on an order someone chose, called precedence. Each system sets its own, and no standard order exists.

Atlan precedence


In Atlan, role, persona, and purpose policies stack, and a user gets the union of their grants. An explicit deny in any policy overrides every grant, including the Admin role (Atlan docs, Permissions and data access).

Open Policy Agent precedence


Open Policy Agent, the open-source policy engine, works differently. Its FAQ says allow and deny statements can both be true at once, and that you control explicitly which one takes precedence, for example deny overrides allow.

Statement order does not matter in its language by default. The else keyword gives first-match, ordered evaluation when you want it (Open Policy Agent FAQ).

Habits that keep ordering predictable


Three habits keep rule order predictable. Write each exception as its own rule with its own owner, instead of burying it inside a broader one. Before a rule ships, test what each affected role sees, including the support team.

A separate allow cannot override a deny, so a deny’s scope has to be narrow to begin with. In Atlan, keep deny rules few and clearly named.

Schedule overlapping automations apart as well. Atlan’s playbook docs warn that concurrent playbooks acting on the same assets can cause inconsistencies (Atlan docs, Set up playbooks).

Zero trust data governance treats every access as denied until a rule says otherwise. For the agent case, see runtime AI governance vs reviewing models before deployment.


Who owns a rule, and how does a change get approved?

Each rule needs a person or group who answers for it, and a route that approves changes. Without both, nobody is positioned to correct the rule when data or use changes.

Owners


In Atlan, owners are users or groups and serve as the primary contact for an asset. That is one place a team can record who answers for the rules on that asset. Column-level owners need the Labs setting “Enable owners for columns” turned on (Atlan docs, Add owners).

Approval


Approval is the second half. Atlan’s governance workflows are templated, multi-step approvals that route metadata changes to the right owners. The templates cover change management, new entity creation, access management, and, when the policy center module is enabled, policy approval (Atlan docs, Create governance workflows).

An admin turns on and manages these workflows.

Ownership in a data contract standard


The Open Data Contract Standard models ownership and approval too. It describes team members with roles, including Owner (team). It also describes roles that carry first and second level approvers (roles).

For more on stewardship at scale, see data governance across business domains and AI-ready data owners and stewards. For the enforcement side, see policy enforcement across operational, analytical, and AI use.


How do rules move into pipelines as versioned code?

Moving a rule into versioned code means it is written in a file, reviewed in a pull request, tested in CI/CD, and rolled back like any other change.

Policy as code with Open Policy Agent


The open-source policy engine Open Policy Agent is built on this idea. It decouples policy from the software service so that the people responsible can read, write, analyze, version, distribute, and manage it separately (Open Policy Agent, Philosophy).

Its opa test command runs PASS and FAIL tests and reports coverage (Open Policy Agent, Policy testing). Its CI/CD guidance covers policy checks in pipelines, including pull request checks (Open Policy Agent, CI/CD).

Data contracts


Data contracts apply the same idea to data (see data contracts for AI). The Open Data Contract Standard defines a contract as a machine-readable, version-controlled agreement between a data producer and its consumers, written in YAML.

Agreements are “reviewed, diffed and versioned like code,” and the same document drives validation in CI and at runtime (Open Data Contract Standard v3.2.0). Its quality rules come in four types: text, library, SQL, and custom (ODCS data quality).

Data contracts in Atlan


In Atlan, data contracts are in Private Preview, cover Table, View, and MaterialisedView assets, and map their properties to asset metadata such as owners, certification, terms, and tags. The Atlan CLI can import an existing contract directly or through a CI/CD pipeline. Webhooks fire when a contract is added or updated (Atlan docs, Create data contracts).

Quality rules in Data Quality Studio


Quality rules run as checks with versioned history. In Data Quality Studio, a rule is a SQL-based expectation, a rule set groups related rules on a table, and each run reports passed, failed, or warning.

Execution happens inside the warehouse. Versioned execution and history tracking let a team track changes and roll back if needed (Atlan docs, Data Quality Studio).

When code earns its overhead


Code earns its overhead when a rule changes often, several teams depend on it, or it needs a test before release. Keep a plain-language record beside the code so a non-engineer owner can still read the rule.

Fixing data quality problems at the source with governance and data quality for AI agents show why the check belongs close to the producer. Computational governance names the broader pattern.


What does Gartner say about rule management?

The 12 November 2026 Gartner Magic Quadrant for Data and Analytics Governance Platforms lists rule management as a platform capability. You can get the complimentary report from Atlan.

Gartner describes the capability in one line: “Automate the enforcement of business rules tied to data elements and associated metadata.” The sections above map to it: tags, terms, and owners tie a rule to data elements, and a control carries out the enforcement. Our Magic Quadrant explainer and the companion Critical Capabilities page cover the rest.


How does Atlan approach rule management? A worked example

Take the running example from the earlier sections and place it on one asset. Atlan keeps the policy scope, owner, terms, and lineage together there, for people and agents alike.

The idea


People keep the approval step, and the system carries out what they approve. See what a context layer is, the enterprise context layer, and the business context layer.

The running example, step by step


Here is the running example, step by step. Steps 3 to 8 use features that Atlan’s docs describe.

  1. Policy. The risk team writes “protect customer email.”
  2. Rule. The team states it as “email is masked unless the requester is on the support team.”
  3. Data scope. The email column is tagged PII, with terms and an owner on the asset.
  4. Control. A data policy on a purpose masks the column for the users it governs, with a hash if the same value must stay joinable across assets. The support exception is set through who a policy covers, and the owner confirms what the support team sees by querying as a support user.
  5. Order. An explicit deny overrides any grant, so a deny written for this rule has to be scoped with care.
  6. Owner and approval. A named owner holds the rule, and changes route through a governance workflow.
  7. Check. Data Quality Studio rules test the data itself and keep a visible history. To test the mask, query as a support user and as a non-support user, and confirm each sees the value the rule intends, in a preview and in Insights.
  8. Code. Optionally, a versioned data contract (Private Preview) carries the owner, tags, terms, and quality checks in a repository. The masking policy itself is set in Atlan.

What trusted data for AI looks like and the agent context layer describe the outcome from the agent’s side.

Setup facts


A few setup facts are worth stating plainly. Data policies are not enabled by default and need an admin or the Governance Admin sub-role (Atlan docs, Data policy). Tag propagation is a setting you turn on.

Playbooks apply rule-based automations in bulk, on a schedule from hourly to monthly (Atlan docs, Set up playbooks). Agents can read the same catalog context through an MCP-connected data catalog.

Get the free 2026 Gartner Magic Quadrant for D&A Governance Platforms from Atlan

Get the Gartner report

FAQs about rule management in data governance

1. What is a business rule in data governance?


A business rule is a testable statement tied to specific data elements, such as “customer email is masked unless the requester is on the support team.” It turns the intent of a policy into something a person, a test, or a system can check against the data.

2. What is the difference between a policy, a rule, and a control?


A policy states intent, a rule is a testable statement tied to data, and a control is the safeguard that makes the rule true. That split is Atlan’s working framing. NIST uses “policy” more narrowly, as a representation of rules for access decisions, and defines a security control as a safeguard or countermeasure.

3. What happens when two data governance rules conflict?


No single standard order exists, so someone has to choose one. In Atlan, an explicit deny in any metadata or data policy overrides every grant, including the Admin role. Open Policy Agent leaves precedence to the policy author, for example deny overrides allow.

4. Who is responsible for data governance rules?


Each rule needs a named owner, a person or a group, who answers for it. Stewards review proposed changes, and an approval workflow can route each change to the right owner.

5. Can data governance rules be written as code?


Yes. Policy-as-code tools keep rules in files that are versioned and tested, and the Open Data Contract Standard describes contracts as YAML agreements that can be validated in CI. Data contracts in Atlan are in Private Preview.

6. How do you test a data governance rule?


Run it against data or sample inputs and record a result. Data quality rules report passed, failed, or warning, and policy-as-code tools such as Open Policy Agent run PASS and FAIL tests and report coverage, ideally on every pull request.

7. When should you move a rule from a document into a pipeline?


Move it when it changes often, when more than one team depends on it, or when you need a test before it ships. Keep a plain-language record of the rule beside the code so owners can still read it.

8. What is data governance, and where do rules fit?


Data governance is how an organization decides who owns data, how it may be used, and how those decisions are carried out. Policies state the intent, and rules are the testable middle step that ties each policy to specific data, so a control can enforce it. Without rules, a policy stays a document that nothing checks.

9. How is data governance different from data management?


In Atlan’s working framing, data management runs the data: moving, storing, and preparing it. Data governance sets the decisions about that data, such as who owns it, who may see it, and what quality it must meet. Rules are how those decisions reach the data management work as something a test or control can check.

10. Why does data governance matter for AI readiness?


AI agents read and act on data directly, and a rule that lives only in a document does not limit what they read. Rules that are tied to data, owned, tested, and enforced by a control apply where the agent works. That is what makes the data ready for an agent to use under the organization’s policies.

11. Who approves data governance policies?


Approval sits with the owner the organization designates, often a governance council or a named data owner, and it should be recorded. NIST SP 800-53 has organizations designate an official to manage development and dissemination of each policy. In Atlan, governance workflows route changes through templated multi-step approvals, including a policy approval template when the policy center module is enabled.

12. How often should data governance policies be reviewed?


No standard sets one interval. NIST SP 800-53 has each organization define its own review frequency for policies and procedures, and also review them after defined events. Annual review plus a review after a major change, such as a new AI use case, is a common starting point.

13. What is the difference between a data governance policy and a data governance standard?


A policy states intent and who is accountable. A standard is a rule, condition or requirement that says how that intent is met. NIST’s glossary definition of standard describes it as a rule, condition or requirement covering specifications or procedures. Rules tied to data then put the standard into effect.

14. What happens when someone violates a data governance policy?


That depends on the organization’s own process, which the policy should spell out in advance. NIST SP 800-53 calls for a formal sanctions process for individuals who fail to comply with security and privacy policies. A rule enforced by a control also blocks or masks the access at the time of the request, so the violation does not depend on someone noticing it afterward.

15. How many data governance policies does an organization need?


No standard sets a number. NIST SP 800-53 writes a policy per control family, such as access control and planning, which shows that the count follows the domains and risks an organization covers. Start with the few that matter most, for example access, classification and retention, and add one only when a decision needs an owner.

16. Why is data governance important?


It decides who owns data, who may use it and under what conditions. That matters more when AI agents read data directly, because a rule that lives only in a document does not limit what they read. NIST notes that policies and procedures contribute to security and privacy assurance.


Sources

  1. Gartner, Magic Quadrant for Data and Analytics Governance Platforms, 12 November 2026, Anurag Raj Et Al.
  2. Policy, NIST CSRC Glossary
  3. Security control, NIST CSRC Glossary
  4. Security and Privacy Controls for Information Systems and Organizations, NIST SP 800-53 Rev. 5
  5. Open Data Contract Standard v3.2.0, Bitol
  6. Data quality, Open Data Contract Standard
  7. Team, Open Data Contract Standard; Roles, Open Data Contract Standard
  8. FAQ, Open Policy Agent
  9. Philosophy, Open Policy Agent
  10. Policy testing, Open Policy Agent
  11. CI/CD, Open Policy Agent
  12. Data policy, Atlan Documentation
  13. Protect data by tag (purpose), Atlan Documentation
  14. Permissions and data access, Atlan Documentation
  15. What are tags, Atlan Documentation
  16. Set up playbooks, Atlan Documentation
  17. Add owners, Atlan Documentation
  18. Create governance workflows, Atlan Documentation
  19. Create data contracts (Private Preview), Atlan Documentation
  20. What is Data Quality Studio, Atlan Documentation
  21. Standard, NIST CSRC Glossary

Gartner does not endorse any company, vendor, product or service depicted in its publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner publications consist of the opinions of Gartner’s business and technology insights organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this publication, including any warranties of merchantability or fitness for a particular purpose.

Gartner and Magic Quadrant are a trademark of Gartner, Inc., and/or its affiliates.

Share this article

signoff-panel-logo

Atlan is the Context Layer for AI. It translates business knowledge, including data definitions, working procedures, and governance policies, into context AI can actually use. This knowledge lives in a single Enterprise Data Graph that every team and AI agent can reach.

In Atlan's AI Labs benchmark, adding this context improved AI's text-to-SQL accuracy by 38%.

Atlan is recognized as a Leader across multiple Gartner reports and Forrester Waves, and is trusted by over 400 enterprises representing $10T+ in market cap, including Mastercard, Workday, General Motors, CME Group, HubSpot, FOX, Virgin Media O2, and Elastic.

Bridge the context gap.
Ship AI that works.