# ChatGPT Work's Data Agent Queries Your Warehouse Under the Permissions You Already Have

OpenAI added a Data agent to ChatGPT Work on 10 September. Invoked with @Data, it queries Snowflake, BigQuery, Databricks, Redshift and more, builds dashboards from plain language, and enforces the connected account's table, row and column permissions. Sixteen days on, OpenAI still names no price or plan list.

By Neomanex. Updated 2026-09-26.

**Verdict: worth a controlled pilot, not a rollout.** The permission model decides whether you can use the Data agent in ChatGPT Work (not in our directory), not the connector list. OpenAI says queries "enforce the connected account's existing permissions, including table, row, and column restrictions." If that holds, an analyst sees through ChatGPT exactly what they can already see in the warehouse. We have not tested it, and sixteen days after launch OpenAI still publishes no price or plan list. Prove the permissions on one connection first.

## What happened

OpenAI announced the Data agent on 10 September 2026. As of our re-check on 26 September, the announcement says:

| Item | What OpenAI states |
|---|---|
| Where it lives | Listed as "Data" in "the Plugins directory in ChatGPT Work" |
| How you use it | Start a conversation with `@Data` and ask a business question |
| What it does | Connects company data, surfaces insights, builds interactive dashboards from natural language |
| Data sources | "Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB, Snowflake, and more" |
| Files | Can bring files and documents from Google Drive and SharePoint into the analysis |
| Business context | Databricks Genie Ontology, dbt, GitHub, Snowflake Horizon and BI dashboards |
| BI tools | Omni, Oracle BI, Power BI, Sigma, Tableau and ThoughtSpot |
| Early users | NTT Data, Thermo Fisher and other organizations in OpenAI's Alpha program |

## Why it matters

**Access is opt-in per connection and per role.** "Enterprise administrators choose which data connections are available and which roles can use them." Admins make it available or install it for teams through Workspace settings > Plugins. That is the right default for a tool that can read a warehouse.

**Permissions are inherited, not re-modelled.** The agent queries as the connected account. That means your existing warehouse grants are the security boundary. If those grants are loose, the agent exposes the looseness to anyone who can type a question.

**The dashboard layer is the open question.** OpenAI's claim covers queries. It does not say how row and column restrictions carry into a dashboard that is later viewed by someone else.

## What changes for you

- **Data and platform admins:** decide the role-to-connection map before anyone installs the agent for a team.
- **Security teams:** test with a role that has row- and column-level restrictions and try to see past them, in chat and in dashboards.
- **Buyers:** the announcement names no price and does not say which ChatGPT Work plans include the agent. Get both in writing from your account team.

## FAQ

**Which warehouses does the ChatGPT Work Data agent connect to?**
OpenAI lists Amazon Redshift, Datadog, Google BigQuery, ClickHouse, Databricks, MongoDB and Snowflake, "and more".

**Does the Data agent respect warehouse permissions?**
OpenAI says queries enforce the connected account's existing table, row and column restrictions. We have not verified this independently.

**How much does it cost?**
OpenAI has not published a price or a plan list for the Data agent as of 26 September 2026.

Canonical: https://neomanex.com/news/chatgpt-work-data-agent-warehouse-permissions
