Customer data
Agree what is collected, redacted, retained and sent to an approved model. Do not send passwords or production dumps through this public site.
Security & trust
AI may recommend.
Authority stays explicit.
Review the control boundaries before the first connection—not after the first production action.
The authority model
- 01
Evidence informs.
Logs, tickets and model outputs are not permission grants. - 02
PROTECTOR evaluates.
Identity, scope, risk, policy and required approval are reviewed separately from reasoning. - 03
The Gateway enforces.
An enabled adapter must validate the specific authorization before acting. - 04
Verification checks.
An execution result is reviewed against later technical and business evidence.
Data & deployment
Define the boundary, then validate it.
Human review
Record who reviewed the proposal, its scope and its evidence. A changed proposal needs an appropriate re-review.
Measured assurance
Request the installed build’s validation evidence and deployment-specific limitations. This page is not a certification or audit report.
A practical next step
Start with a control review.
Bring a recurring incident, a change challenge or an AI idea. We’ll agree the evidence, boundaries and a useful first outcome.