Alation brings the data catalog into ChatGPT Work—but not the permission contract

Alation now has a plugin in ChatGPT Work that brings enterprise-catalog context into a conversation. OpenAI’s plugin page says users can discover data through catalog metadata, governance policies, semantic definitions, lineage, quality signals and documentation.

That is a different integration point from the warehouse and BI plugins surrounding it in OpenAI’s data-tool directory. Those products answer questions against data or create analyses. Alation’s advertised role is earlier in the chain: help the agent decide which asset and definition should be trusted before it asks another system to execute a query.

A pre-query control plane

The three examples on OpenAI’s page are revealing. One asks the plugin to find suitable revenue datasets using definitions, quality signals and certification status. Another traces dependencies before a schema change. A third compares competing revenue definitions to explain conflicting reports.

For natural-language analytics, those are not peripheral catalog tasks. They address three common failure points before SQL generation: selecting the wrong table, relying on stale or low-quality data, and silently choosing one of several business definitions for the same metric.

Alation’s own catalog description says its platform records where data lives, how it should be used and how trustworthy it is. It also exposes lineage, shared SQL queries and results from connected data-quality tools. The company says the catalog supports more than 120 connectors, spanning databases, BI systems, files, applications and AI models.

Together, the two pages describe a plausible separation of duties: Alation supplies the discovery and trust context, while a warehouse or BI plugin handles execution and presentation. That can make the catalog a control plane for the question plan rather than just a search box for humans.

What the public page does not say

The launch page is thin on operating details. It does not state which ChatGPT Work plans receive the plugin, how administrators install it, which Alation permissions are carried into a conversation, whether raw data is returned, or whether certification and quality signals merely inform the model or hard-block an action. It also publishes no evaluation showing that catalog grounding improves dataset selection or answer accuracy.

Those omissions matter. A model seeing a “certified” label is not the same as an enforcement layer preventing use of an uncertified asset. Likewise, tracing lineage is useful context, but it does not by itself prove that a downstream query respects source-system permissions.

Teams evaluating the plugin should therefore test four things before treating it as a governance control: whether users can discover assets they cannot otherwise access; how conflicting metric definitions are presented; what happens when quality or certification signals disagree; and whether every catalog lookup and downstream query can be tied to the initiating user.

The launch is still notable because it moves governed context into the same conversational surface where analysis begins. But until OpenAI or Alation documents the permission and enforcement contract, the safest interpretation is narrower: this is a catalog-aware assistant, not yet a demonstrated policy boundary.