Why Blackbox AI is a no-go for Customs Compliance

Customs compliance in formal U.S. entries means ensuring the licensed customs broker makes accurate, documented decisions on classification, valuation, and duty. That requires an audit trail that shows the basis for each decision and supports responsible supervision and control.

This is critical as tech vendors in the industry race to add AI capabilities. AI can support customs work – it can organize information, flag exceptions, and help teams work through high volumes of data. But technology does not replace the licensed customs broker’s responsibility for the customs business conducted under their license.

The question is not whether a vendor uses AI or deterministic rules. The question is whether the brokerage can maintain control over entry-relevant decisions and demonstrate how an entry was handled when CBP asks.

U.S. Customs and Border Protection (“CBP”) addressed this issue in January 2026 in HQ H350722. For licensed customs brokers, customs compliance professionals, and trade compliance software vendors working with CBP, that ruling puts the responsibility in plain terms: in customs business the licensed customs broker must play a significant role in determining entry-relevant output.

An audit trail that identifies the basis for each decision is what separates a defensible filing from one that exposes the broker to personal liability and leaves them unable to answer a CBP inquiry. This article looks at why that audit trail matters, what questions brokers should ask software vendors, and how Tru Identity’s platform addresses those requirements.

Why broker control and audit trails matter for customs compliance

Customs Brokers licensed by CBP are personally liable for all entries filed under their license. The CBP HQ H350722 ruling confirms that this responsibility does not change because a brokerage uses AI or other technology.

A brokerage does not need a vendor to disclose every detail of its technical architecture. It does need to know that the system supports broker-controlled decisions, documented exception handling, and a retrievable record of the basis for each entry. When CBP sends a CF-28 fourteen months after a filing asking why a particular HTS code was applied, the broker must produce a timestamped record to explain how the decision was handled.

The audit trail must show the basis for the output, the information and regulatory sources considered, the applicable process, and any documented broker review. If those records do not exist, the brokerage cannot clearly demonstrate how it exercised control over the entry.

The question set every licensed customs broker should bring to a vendor demo

A licensed broker can’t meaningfully evaluate the rest of a trade compliance solution until the vendor’s demo clearly explains its decision architecture and audit-trail documentation.

Architecture

How does the system support broker-controlled decisions, identify exceptions, and preserve the record required to explain an entry later? The system must support accurate commercial invoices and permits and still tie each decision to a specific rule or documented broker decision. If the vendor cannot show how the system handles exceptions and preserves the decision record, then there is nothing to document.

Which party is responsible for recording which version of the rules was in effect at the time of each filing? Tariffs change with very little notice. An entry submitted Thursday morning could be subject to a different Chapter 99 provision than an entry submitted later that day. To address this, the audit trail must record the date and time of each version and document which version of the rules, along with the required documentation tied to that filing, was in effect at the time of filing.

The CF-28 test

Can the software generate the decision rationale and valuation basis for a specific entry filed 14 months ago? A usable historical record should also show the transaction value declared for the merchandise when valuation was part of the filing basis, since proper valuation requires declaring the true transaction value of merchandise. If yes, it is also important to determine whether the response is based on information stored in the audit trail, or whether it would require someone to reconstruct the decisions from email records and spreadsheets within a 30-day timeline.

Confidence threshold

What happens when the confidence level of AI scoring drops below a predetermined level? And does the system automatically route low-confidence exceptions into documented review procedures for evaluation by a licensed broker, or does it just default to the most probable answer?

The answers should determine whether the licensed broker actively participates in reviewing entries or simply provides oversight. Maintaining customs compliance also depends on regular audits and employee training around those procedures. If the vendor cannot clearly answer any of the questions above, then there is probably a compliance gap.

What HQ H350722 means for audit trails in international trade

CBP’s HQ H350722 ruling makes it clear that a licensed customs broker will need to play an important role in defining what data any tool’s decision-making process automatically creates when it generates entry-related output. To be clear, the ruling does not prohibit the use of AI in the customs business. Instead, it sets limits on who determines how AI is used within the customs business, because the purpose of customs compliance is to follow regulations in cross-border trade.

On audit trail compliance, the effects are very direct. A brokerage needs records that show how entry-related decisions were handled and how the licensed broker maintained responsible supervision and control. The record must support the brokerage’s process regardless of whether the workflow includes AI, automation, deterministic rules, or manual review.

Under 19 CFR 111.28, a broker cannot assert supervisory responsibility over the customs business unless he has records. Following these requirements also helps maintain good standing with government agencies. In this case, the audit trail is those records. When a customs broker manages large-volume supply chain accounts with thousands of line-item entries, determining an audit trail’s adequacy becomes a scale issue.

Manually documenting each decision may work well for 50 entries per month, but it will not work for 500 or 5,000 entries. As such, the system must automatically capture the reason for each decision, or there will be no record of the decision basis. Automated audit trails only reduce reliance on manual documentation if they preserve the basis for each decision and the broker’s documented review procedures, which supports fair and secure global trade and fair trade practices by making improper or illegal activity less likely to go undetected.

How Tru Identity builds audit trail discipline into customs business

Tru Identity’s platform ensures the basis for each decision is captured at the moment of filing, and the timestamp applies the regulation in effect at that moment, so users can access the filing basis and supporting record at the moment of review and the audit trail answers the CF-28 before the CF-28 even arrives.

That was intentional from the Tru team. The platform is designed to support broker-controlled compliance workflows at scale, including clear exception handling, confidence scoring where appropriate, and a documented record of the basis for each filing. Each classification decision is logged against the specific CSMS bulletin, ruling, or Federal Register notice that governed it. That defines the scope of the platform’s controls while leaving licensed broker judgment where regulations still require it.

Every point is auditable, and every decision is based on how you want it carried out, making it the most compliant method available. After implementing Tru’s application, GHY reduced onboarding data review time by nearly 90%, from about 1.5 hours per client to under 10 minutes. Reviews are now billable services, turning compliance into a revenue line item rather than a cost item that grows with every headcount.

Frequently asked questions

Q1. How do audit trails protect licensed customs brokers during CBP audits?

Audit trails create verifiable records that document which regulation applied to each classification decision at the time of filing. When CBP issues a CF-28 or conducts a review, the broker needs to demonstrate responsible supervision under 19 CFR 111.28. Weak records can also worsen customs non-compliance consequences beyond the audit itself, including confiscated goods and reputational damage. A timestamped audit trail showing the decision basis and the rule version in effect is the documentation that proves supervision happened. Without it, responsible supervision is a claim with no proof.

Q2. Can AI tools independently make entry-related determinations in formal customs entries?

AI can assist a customs workflow, but a licensed customs broker must have a significant role in specifying entry-relevant output. For importers, AI suggestions may inform review, but broker-controlled decisions still matter when country of origin determination affects tariff calculations or trade agreement treatment. Vendors whose systems do not preserve the decision basis and documented broker oversight leave the broker without proof they made the call. That gap is the exposure the ruling was designed to address.

Q3. How does Tru Identity track which regulation applied at the time of a customs filing?

Tru Identity timestamps every classification decision against the specific regulation in effect when the entry was filed. CSMS bulletins, rulings, proclamations, and Federal Register notices are ingested when they are published, so each filing record traces back to a primary source. When a CF-28 arrives months later, the broker exports a structured record instead of reconstructing the rationale from emails, spreadsheets, or memory.

Q4. What is the difference between AI classification and rule-based classification in customs compliance software?

AI classification can use probabilistic models to help process and prioritize information at scale. Rule-based classification uses deterministic logic that produces the same output every time when the same conditions apply. Tru uses both deliberately, with documented controls, confidence scoring where appropriate, and deterministic rules where one right answer must be repeatable.