Who's Allowed to Use the Good Chairs

Opening Scene

A well-run household doesn’t treat every piece of furniture as equally available to everyone — some rooms and pieces are genuinely open to all, others reserved deliberately for specific occasions or people, matching real, sensible distinctions about use and care. An internal AI tool deserves this same deliberate access control, matching an organization’s actual data sensitivity and role structure, rather than defaulting to uniform access for everyone.

In Plain English

Access control for an internal AI tool means deliberately matching who can query what data and perform what actions to an organization’s actual role structure and data sensitivity classifications, connecting directly to the access control and data security practices covered in this content library’s dedicated data governance series. A tool grounded in retrieval, covered in this content library’s RAG series, needs particular care here, since it might otherwise surface sensitive information to someone without the underlying permission to see it directly.

The Old Way

Before deliberate access control design was standard practice for internal AI tools, permission structures were sometimes applied too loosely:

• Some internal AI tools were deployed with uniform access for all internal users, without genuine consideration of underlying data sensitivity or role-based restrictions.
• There wasn’t yet a well-established practice of ensuring a retrieval-grounded tool respected the same access permissions as the underlying data sources it drew from.
• A tool could sometimes inadvertently surface sensitive information to a user who lacked the underlying permission to see the source document directly.

Deliberate, role-matched access control, respecting underlying data permissions, reflects the access control practices covered throughout this content library’s dedicated data governance series.

What’s Changing (and Why AI Is the Reason)

1. Internal tool access control increasingly respects the same permissions as underlying data sources, connecting directly to the access control practices covered in this content library’s dedicated data governance series.
2. This connects directly to the retrieval grounding covered in this content library’s RAG series, since a retrieval-grounded tool specifically needs to filter results by the querying user’s actual permissions.
3. As this discipline matures, access control is increasingly designed deliberately from the start, not retrofitted after an initial, overly permissive deployment.

The Metaphor, Fully Extended

The Custom Furniture Maker	Access Control Concept
Not treating every piece as equally available to everyone	Not treating every data source as equally available to everyone
Real, sensible distinctions about use and care	Real, sensible distinctions about data sensitivity and role
Some pieces reserved deliberately for specific occasions	Some data reserved deliberately for specific roles or permissions
A household’s genuine, considered structure, not uniform access	An organization’s genuine role structure, not uniform tool access

Quick Recap

• Access control for internal AI tools should deliberately match an organization’s actual role structure and data sensitivity.
• This connects directly to the access control practices covered in this content library’s dedicated data governance series.
• Retrieval-grounded tools specifically need to respect the same permissions as their underlying data sources.
• Access control should be designed deliberately from the start, not retrofitted after overly permissive initial deployment.

Where This Fits in the Series

Article 10 covered deliberate access control design. Article 11 turns to the craftsperson’s own workshop safety rules: broader governance and internal AI policy.