The data-access scope map we hand to security.

It's a diagram Silicon Prime hands to security instead of explaining access in prose. Boxes on the left are systems the customer owns; the dashed line is the trust boundary where data would leave the customer's control. The map makes one provable claim: nothing crosses that line unless the diagram says it does. A trust boundary you can draw is one you can audit—if it only lives in a config file, it doesn't exist yet.

Read-only means the agent holds a credential that cannot mutate state—not a promise, the credential literally can't write. Read-write is the narrowest possible scope: the agent can append to the ticket queue but can't delete, read other tenants, or escalate. Never-touched means no credential was ever minted—payments and PII have no key on the ring, so there's nothing for the agent to argue past.

Because policy is advice and a scoped credential is physics. The common mistake is treating access scope as a document saying the AI 'should not' write to production. When security asks for proof of read-only, the team doesn't show the policy—they show that the token issued to the agent has no write grant, plus an audit log proving no write was ever attempted.

The orange dashed line is the only thing on the page that matters to a regulator. Everything to its left is the customer's; everything that crosses it must be named, scoped, and logged. It marks the exact point where data would leave a place the customer controls, making the boundary provable and auditable rather than an assurance buried in prose.

Two ways together: the token issued to the agent has no write grant, so writing is technically impossible, and the audit log shows no write was ever attempted. This pairing—a credential that can't mutate state plus evidence of behavior—is what the team presents when security asks for proof, rather than a policy document stating the AI shouldn't write.

Because the map's value is honesty that security can verify. Everything that crosses the trust boundary must be named, scoped, and logged, so the team would rather hand over a diagram that openly shows a write than a paragraph that conceals one. A drawn boundary is auditable; a buried one isn't, and hiding access only surfaces as risk later.

Never about the model—it's about the data. Security wants to know what the system can read, what it can change, and where data leaves a place the customer controls. That's why Silicon Prime stopped explaining access in prose and started handing over a scope map that answers those questions visually and provably.