Central AI governance

Verifiable core and a clear expansion path

The existing control core makes server-side policy decisions and records redacted workspace, work order and security events. The next expansion packages will add a central AI inventory and approval centre.

Expansion package 1

In preparation: central AI inventory

The planned governance register will enforce server-side authorisation: only authorised roles will be able to create, edit, approve, or archive entries.

AI system

Name, purpose and model route

In preparation

Accountability

Responsible person and workspace

In preparation

Data class

Processed data and permitted boundaries

In preparation

Risk

Criticality and human control measure

In preparation

Review

Review date and document references

In preparation

Status

Draft, reviewed, approved or archived

In preparation
Verifiable control core

Turn policies into server-side limits

The policy decision evaluates workspace, role, data class, sovereignty mode, model route and permitted tools. The browser cannot set or bypass these approvals.

Example policy evaluationExample
“Confidential data only through a route permitted for the workspace.”Data class + route
“External models only after provider review and approval.”Provider + workspace
“Block shell and connectors for restricted data.”Tool + data class
“Critical outcomes require human review.”Risk + approval
Redacted audit trail

Traceable events without sensitive raw content

Policy, workspace, work order and security events are recorded without placing prompts, credentials or raw content in the audit trail. A filterable interface and approval workflows will be added in expansion package 2.

Workspace event

Workspace, role and work order.

Redacted

Policy event

Decision on route, policy and tool limits.

Redacted

Data class event

Data class and applied sovereignty mode.

Redacted

Security event

Block, approval or revoked permission.

Redacted