AI Access Controls
Apply least privilege: the AI reads only the specific data its task requires, with role- or attribute-based permissions enforced at the document level and every access logged. 97% of AI incidents involved organizations that skipped exactly this.
Least privilege for AI
An AI with access to everything is a single compromised credential away from exposing everything. Least privilege limits both the blast radius and the incentive to attack it.
The same principle that governs human access applies doubly to AI, because AI acts fast and at scale. Grant it read access only to the data its specific job needs, enforce permissions by role, and log every access. When 97% of AI incidents trace back to missing controls (IBM, 2025), least privilege is not best practice, it is the baseline.
Enforce at the document level
Blanket access to a data store leaks everything in it. Document- or record-level access control ensures the AI can only surface what a given user is entitled to see.
A well-built AI assistant should never surface data the asking user could not access directly. Enforce access at the document level so the AI respects existing permissions instead of becoming a backdoor around them.
Frequently asked questions.
Does an AI assistant create a permissions backdoor?
It can, if built carelessly, an AI with blanket data access will surface things the asking user should not see. Enforce document-level, role-based access so the AI inherits and respects each user’s existing permissions rather than bypassing them.