Before a regulated enterprise lets an AI system near its data, the CIO has three questions: where does our data go, who can touch it, and can we prove what happened. We answer all three on one page. We send the same page to every CIO, because the answer shouldn't depend on how hard you push. Here is the memo, drawn as a diagram.
The trust boundary is the first thing on the page.
Before we describe a single capability, we draw the dashed line and say plainly: your data does not cross it. Processing happens inside your tenant. We do not train on it. There is no quiet egress to somewhere else. A CIO should be able to point at the line and know the answer to "where does our data go" before reading a word of prose.
Access scopes are named, least-privilege, and revocable.
The system reads what it is explicitly granted and nothing more. Scopes are named — not "broad access," but a specific, listed set. They follow least-privilege by default, and any of them can be revoked in a single action. Revocation is not a support ticket. It is a button the customer holds.
The audit trail is the part legal actually cares about.
For every interaction we record who asked, what was retrieved, what the model actually saw, what it returned, and who approved it. The log is append-only and exportable. When a regulator asks "what happened on this date," the answer is a query, not an investigation.
A governance memo that only the vendor understands isn't governance. It's marketing with a compliance font. The test is whether your CIO can hand it to legal without translating it.
Why one page, and why the same page for everyone.
- A page a busy CIO can absorb in two minutes beats a fifty-page policy nobody finishes.
- The same memo goes to every client. Governance that depends on how hard you negotiate isn't governance.
- A human approves consequential actions. The audit trail records the human's decision alongside the model's output — accountability stays with a person.
— Suhail Abidi. Walnut Creek, CA. June 2026.
Comments