A ChatGPT Enterprise database connection needs five layers, not one credential

An enterprise team rarely needs ChatGPT to "access the database."

It needs specific employees and workflows to answer approved questions from current internal data.

That is a different architecture.

A production connection needs five layers:

1. Identity
   Resolve the human, organization, agent/session, and policy context behind the request.

2. Capability
   Expose approved business questions with explicit parameters, periods, dimensions, and result shapes.

3. Data scope
   Use read-only roles and governed views that encode joins, metric definitions, sensitive-field exclusions, and tenant boundaries.

4. Execution control
   Apply statement timeouts, row limits, cost budgets, cancellation, and bounded retries. Bind execution to the reviewed plan.

5. Evidence
   Return source, metric/schema version, period, filters, freshness, row count, truncation, and a query or trace ID.

Write actions should use a separate path with validation, idempotency, preview, approval, and compensation.

The connector is useful.

But a connector without identity, scope, execution policy, and evidence is just a credential with a conversational front end.

Full architecture: ChatGPT Enterprise database connection