Using AI in Trade Compliance: Navigating CBP Ruling H350722 and Beyond

The rapid integration of AI into the global supply chain has proven incredibly beneficial — and it’s significantly complicated matters at the same time.

From automated classification to predictive risk modeling, technology is solving problems that have plagued trade compliance professionals for decades. However, as with any technological frontier, the regulators eventually catch up.

Recently, U.S. Customs and Border Protection (CBP) issued a multifaceted ruling: H350722. It’s a potential line in the sand for how AI and digital platforms can and cannot be used in the execution of customs business. 

The Core of H350722: What constitutes Customs Business?

The crux of the CBP ruling centers on the definition of “customs business” under 19 U.S.C. § 1641. Traditionally, this involves activities related to the entry and admissibility of merchandise, its classification, and the payment of duties.

In H350722, CBP clarified that while online platforms are permissible for the transmission of information, they cannot be used to actively prepare or file customs documents unless the entity behind the platform is a licensed customs broker.

There are three critical takeaways from this ruling that every compliance officer should know:

The Direct Power of Attorney (POA) Requirement

CBP was adamant that a Power of Attorney must be direct between the broker and the importer. It cannot be mediated or funneled through a third-party technology platform in a way that obscures the legal relationship.

The 10-Digit Red Line

This aspect of the ruling applies to HTS classification. CBP suggests that if an AI system or service provider gives a full 10-digit HTS number, it is engaging in customs business if it also uses that code to file an entry from the same technology platform. This implies that such a provider must hold a customs broker’s license. While advisors such as lawyers and accountants have some leeway, the automation of generating the 10-digit code by a non-licensed software provider could be a high-risk activity for the software provider depending on how and where the code is used.

Confidentiality and Chat Functions

CBP expressed concern that AI-driven chat functions within platforms could violate a broker’s duty of confidentiality. To mitigate this, brokers must ensure they have a specific waiver of confidentiality within their POA to allow for the necessary data sharing that powers these AI tools. This liability falls on the broker, not the importer.

Strategic Shifts

While H350722 sets rules for future practice, many importers are still dealing with the economic and operational fallout of past regulatory actions, specifically regarding the International Emergency Economic Powers Act (IEEPA) duties.

With recent court rulings directing CBP to re-liquidate certain entries, the conversation has shifted to the status of refunds and the newly proposed Consolidated Administration and Processing of Entries (CAPE) system. For importers looking to secure their financial position, the advice from legal experts is clear: be in court. Even if a streamlined process for refunds is established, filing a protest is the only way to fully secure your legal standing. 

This type of belt-and-suspenders approach — filing protests even while pursuing court action — ensures you won’t miss any procedural deadlines. Throughout this process, your Automated Commercial Environment (ACE) account remains the system of record; keeping it meticulously updated is not optional.

The Necessity of AI Guardrails

While there’s no shortage of uncertainty in the wake of HQ350722, one thing is clearer than ever: AI use must be accompanied by guardrails.

Large Language Models (LLMs) are incredibly powerful, but they’re inherently imprecise. Because a single incorrect digit in a classification can result in significant penalties, “mostly accurate” isn’t good enough.

This is why having a human in the loop is critical. AI can perform the first pass of an analysis, but a human expert must validate the final output. Furthermore, robust audit trails are critical so you can “show your work” to CBP. This means documenting every input and output. Last, it’s wise to select an AI software solution that’s designed to access the most current regulations unless it is prompted otherwise. 

An AI Adoption Checklist

As your organization moves toward broader AI adoption, consider this high-level checklist to aid compliance:

• Data governance: Is your data stored in the jurisdiction it will be used in? Is it handled by a jurisdiction-specific provider?
• Licensing: Does your technology partner have at least one corporate officer who is a licensed customs broker overseeing the process?
• Risk assessment: Are you using AI for “production” (safe) or “preparation” (requires a license)?
• Confidentiality: If you are a brokerage company, have you updated your POAs to include the necessary confidentiality waivers?

Deep Dive

For a more comprehensive look at these topics, including expert commentary from Todd R. Smith, Anand Raghavendran (KYG Trade), and Richard M. Wortman (GDLSK), watch our recent webinar: “CBP Ruling H350722, IEEPA, and Guardrails for AI.” In this session, we break down the granular details of the CBP ruling and provide practical steps for future-proofing your compliance program.

Conclusion: Building Confidence in Innovation

Online platforms and AI are here to stay, but they must function as tools for the global trade professional, not replacements for the regulated legal frameworks that protect importers, consumers, and federal revenue sources. At KYG Trade, we believe that the most effective compliance programs are those that combine AI-assisted efficiency with expert human judgment. 

Contact us to learn how our AI assistant, Kay, provides expert-level classification, while our human-in-the-loop ethos safeguards accuracy and compliance. You can also request a demo.