SOC 2 Type I

A Type I examination asks one question. On a named date, were the controls designed to meet the criteria the auditor selected? This starter answers that question for the AI traffic that passes through the gateway. It gives you a control map and evidence inputs. It does not give you an audit opinion.

Teams reach for this starter when an auditor asks how AI traffic is governed and no answer exists yet. The task is to show that every request passes a defined set of checks, and that the record of those checks survives review. Access, logging, encryption and incident handling appear as fields that you complete.

The configured chain inspects the prompt before the model ever sees it. Injection screening runs first, so a crafted instruction cannot reach the role check and disable it. Role enforcement then rejects any call that arrives without a verified user and organization identity. Redaction masks national identifier patterns next, and a block list stops passwords, private keys and access tokens from leaving the request.

On the way back, output scoring holds the response to a combined score of 0.85, with relevance at 0.78 and accuracy at 0.80. An answer under 120 characters or a single sentence fails the same gate. A response that misses any floor is blocked rather than delivered, so a reviewer sees a refusal instead of a weak answer.

Audit logging sits at the end and marks each evaluated request. It never blocks, and it does not by itself store, retain or seal anything. Storage, retention and evidence export are separate settings that your team owns.

Scope stays with you. Select the trust services categories in play, name the system boundary, and agree the examination date with your auditor. Encryption, availability and confidentiality appear as control areas because the report asks about them. The gateway supplies evidence only for the parts of those areas that it handles. Choose the Type II starter when an auditor must test behavior across a review window rather than on a single day.