AI Security Models Explained

Naming note: despite sharing the name "Security Model," this is unrelated to the Branch Types / Security Models feature under My Account (Workspace) — that one governs sub-organizations; this one governs AI access limits. Don't conflate the two when writing or answering questions. Placement note: the org-wide default system prompt, auto-model-selection toggle, and required opening/closing text settings all live at the bottom of this same Security Models tab, not on their own tab — see How do I configure general AI settings?

An AI access policy (what this tab calls a "security model") is a named bundle of limits and permissions you can assign to one or more roles:

• Max Tokens Per Request and Max Context Length — per-call limits.
• Max Tokens Per Day and Max Requests Per Day — daily caps.
• Allowed Providers and Allowed Models — restrict which connected providers/models this policy's members can actually use.
• Allow Image Upload and Allow MCP Tools — feature-level permissions.
• Assigned Roles — which Security Roles this policy applies to.

A member's actual limits come from whichever policy is assigned to their role(s), the same way rights and permissions cascade elsewhere in Workspace settings.