Questions & Answers

TrustGate is an action-clearance and proof layer for consequential AI actions. It binds authority to an exact action and helps account for what the protected target actually did.

That label is too broad on its own. The bounded pattern concerns exact-action clearance, a protected-target check, and outcome handling; it is not a claim of general AI safety.

TrustGate is described as complementary to identity, policy and operational controls. It is not presented as a replacement for IAM or PAM.

No. Existing workflow, orchestration and execution systems can remain in place; whether a target path can be assessed is a separate question.

No. A use case may not need it where existing target controls already close the consequence lifecycle or where the action is not material enough to justify another control.

A bounded Salesforce Developer Edition reference: Salesforce committed a real action, its success response was lost, TrustGate kept the result UNKNOWN, refused protected continuation, and confirmed the outcome only after a fresh Salesforce readback.

No. It is not a customer production deployment and does not certify the full TGAP-PROD runtime.

The external consequence is unresolved. It is not treated as success or failure, and the protected continuation remains fenced until external reality is established.

A timeout does not establish whether the target committed. Retrying can create another consequence, so the proof refused equivalent and conflicting protected continuation before a second protected write.

The proof used target-side evidence, including a fresh Salesforce readback. Separately, qualification showed Salesforce accepted two ordinary Account POSTs with the same canonical business action identity; that endpoint did not deduplicate them.

This reference proves only its bounded action path. Multiple consequences need explicit target-specific assessment and are not covered by a universal claim.

Those controls may close the lifecycle sufficiently. In that case TrustGate may add no value and may be unnecessary.

That is a deployment decision to agree for each environment. The public proof does not certify a production topology.

The proof credentials were ephemeral and excluded from evidence. Production credential location and custody must be agreed within the customer and vendor trust boundaries.

Production should allowlist only the approved target and control endpoints required by the agreed design. The exact list is deployment-specific.

The stated minimum is action, attempt, expected-outcome, lifecycle and evidence state needed for consequence custody. Retention, access and deletion requirements must be qualified.

The demonstrated posture is no silent retry and UNKNOWN when execution is ambiguous. Production recovery behavior and operational ownership must be agreed.

A stop and disable procedure must be agreed and tested for the specific deployment before production admission.

The proof does not establish compromise resistance. Threat model, privileges, containment and recovery need customer-specific security review.

UNKNOWN remains visible and protected continuation stays fenced while external reality is unresolved. The owner, investigation and closure procedure must be agreed operationally.

Changes require a defined qualification and regression process for the target path before use. This proof does not certify future versions.

It does not provide omniscience, distributed atomicity, control of routes outside its custody, compliance certification or universal production safety.

No. A bounded reference review can start without customer production data.

Start with one bounded consequential action, target, route and expected outcome, then agree evidence and responsibilities.

Yes, where the assessed design keeps those systems in their roles and establishes the required target and authority boundaries.

Then TrustGate may be unnecessary. A fit review should identify that early.

A pilot should name its use case, target, evidence, success boundaries, responsibilities and limitations. No outcome is guaranteed in advance.

No. It does not determine compliance or legal admissibility. Those require the relevant human, legal and organizational review.