Lesson 5 of 10 · Building the context
Governance rules agents can read
Short answer
Governance helps agents only when certification, sensitivity and access rules travel with the context in a form the agent receives, and stay current as the stack changes.
By The Lineage Desk · Best Data Catalogs · Published
Desk research from public vendor pages. Editorial assessment.
2 min read
Most enterprises already have governance: owners, certified datasets, policies on personal data. The problem is that much of it lives in documents, tickets and people's heads. An agent cannot read a wiki page about which revenue table is approved.
What does machine-readable governance look like?
It looks like labels and rules attached to assets. Euno describes turning governance rules into live labels that propagate across its context graph: for example, a Tableau asset is trusted if it queries only gold-layer dbt models, and assets that depend on personal data are flagged as not AI-ready. Collibra describes certified data products that carry meaning, lineage, quality signals and policy constraints. Atlan places governance, glossary and certification in its Enterprise Data Graph, with people certifying what AI agents draft.
Source: Euno blog: AlphaSense x Euno at dbt Summit
Source: Collibra homepage
Source: Atlan homepage
Why should rules be computed, not hand-applied?
A label applied by hand is correct on the day it is applied. When a model upstream changes, the label does not. Rules computed from lineage stay current as the stack changes, which matters more for agents than for people, because agents act on the label without asking.
How should access be scoped?
Access rules have two halves. The data platform already enforces who can query what. The context layer must also avoid putting restricted context into an agent's prompt. AtScale gives each agent its own OAuth identity with row- and column-level security and logs every call. Honeydew propagates user identity and native RBAC inside the warehouse. Euno describes persona-based access to context. DataHub lists access and policy, data contracts, RBAC and SSO.
Source: AtScale: semantic context for AI via MCP
Source: Honeydew homepage
Source: Euno homepage
Source: DataHub Cloud context platform
What about governing the agents themselves?
A newer layer of governance tracks what agents did. Alation's AI Governance, generally available since its 17 September 2026 release, traces agent lineage and connects to agent registries in Amazon Bedrock, Microsoft Copilot Studio, Snowflake Cortex and others. This answers a question auditors will ask: which data did this agent touch to produce this answer?
Source: Alation press release: AIOS expansion with six products
Where do you start?
Write down three rules you already enforce by hand, such as certified sources only for board metrics, no personal data in agent prompts, and finance models owned by finance. Ask each vendor to express them as rules in its product and show how an agent sees the result.
Then change something upstream: mark a source as deprecated or add a personal-data column to a model. Watch whether the labels on downstream assets change on their own, how long that takes, and whether the agent stops using the affected assets. That single test tells you more about governance for agents than any feature list.
Key points
- Governance helps agents only when it travels with the context.
- Rules computed from lineage stay current; hand-applied labels drift.
- Govern both the data agents read and what agents did with it.