Responsible AI
The Sentinel Flow uses AI to assist trade professionals, not to replace them. These principles describe how AI is bounded inside the product, how humans stay accountable, and what customers should expect when they verify a recommendation.
Human-in-the-loop
Any workflow step that touches a customer record, filing, or outbound message requires a human review gate before execution.
Approval gating
Unattended actions on customer-impacting workflows are blocked. Approvals capture reviewer identity and timestamp.
Explainability
Each recommendation ships with the source fields, HTS/agreement references, and calculation basis behind it, so a reviewer can verify before acting.
Bias considerations
Financial computations (recovery, duty-free savings, exposure) are deterministic against structured data — not generated by a language model — which limits model bias in the values customers act on.
Evidence-first philosophy
We treat AI output as a recommendation with citations. If the evidence is not sufficient to justify a filing, the opportunity is surfaced but not advanced.
Transparency
Where AI is used, where it is intentionally not used, and what it cannot decide are all documented on the AI Transparency page.
Known limitations
AI outputs can be incomplete or wrong. Findings should be validated by an authorized reviewer or a licensed customs broker or trade counsel before a filing is submitted.
Continuous improvement
We monitor model behavior, incorporate reviewer feedback, and adjust prompts, guards, and deterministic engines as the tariff and regulatory landscape changes.
Customer verification responsibility
AI outputs are analysis aids. Before filing with a customs authority, correcting a classification, or executing a strategic response, an authorized reviewer — or your licensed customs broker or trade counsel — should confirm the finding. Customers remain responsible for the accuracy of any filing submitted under their account.