The policy matrix

The policy matrix is your control surface — a single, readable table that governs every AI request. It maps each data class to three settings:

• Minimum trust — the lowest trust tier a destination may be for this kind of data.
• If it can't go there — the fallback action when a destination is below that minimum.
• Internet — the web-access policy for this class.

Fallback actions

Action	What happens when a destination is below the minimum trust
Allow	Send as-is (used when the minimum trust is public-frontier).
Redact	Strip the sensitive values, then send.
Route	Don't send to this destination; route to an approved one, else block.
Block	Never send.
Require approval	Hold for an administrator's decision.

Example default matrix

Data class	Minimum trust	If it can't go there
Source code / secrets	—	Block
PCI / payment data	Enterprise-managed	Route (or redact)
PII / personal data	Enterprise-managed	Route
Financial data	Enterprise-managed	Route
Strategy / board confidential	Enterprise-managed	Route
Public / non-sensitive	Public frontier	Allow

Some things are always blocked

Regardless of the matrix, raw secrets and prompt-injection attempts are blocked at every tier — no destination is trusted enough. The matrix governs everything else.

You own the dial

The matrix is yours to set, per organization. Loosen a class to allow it on more destinations, or tighten it so it only ever reaches your most trusted model. Whatever you choose:

Every change is auditable

Saving the matrix records a policy change in the audit log, with the old and new values. The people who govern AI are themselves governed.

To configure it, see Administration → policy matrix.