AI and reasonable care in trade compliance: What professionals actually need

The numbers tell the story. In 2020,

Meanwhile, the Thomson Reuters US Customs and Border Protection published fewer than 180 regulatory updates for the full year. In 2025, that number crossed 520. And as of mid-2026, the pace is already tracking above 630 for the year.

Meanwhile, the Thomson Reuters 2026 Global Trade Report found that 72% of trade professionals now rank U.S. tariff volatility as their top regulatory challenge — up from 41% just twelve months earlier. Nearly half report increased stress. More than half report increased workload.

Jump to ↓

Interest in AI is rising. Confidence hasn’t caught up.

It’s no surprise, then, that interest in AI has spiked. Forty percent of organizations are now exploring AI for trade management, compared to just 6% two years ago. The appeal is straightforward: if AI can compress hours of regulatory research into minutes, compliance teams get time back to do the work that requires human judgment.

But interest and adoption aren’t the same thing. Actual use of AI-powered software for trade management remains low — and the gap has a reason. Trade compliance professionals understand, better than most, what it costs to get something wrong. Before they hand any part of their workflow to an AI tool, they need to know it will hold up. Not just in practice, but under scrutiny — from leadership, from auditors, from CBP.

That gap — between AI interest and AI confidence — was the starting point for a recent Thomson Reuters webinar, Prompt like a pro: Mastering AI for global trade compliance. The session was led by Andrew Moxon, a licensed U.S. customs broker with more than two decades in the field and Senior Product Marketing Manager for ONESOURCE Global Trade at Thomson Reuters. Over the course of the conversation, Moxon showed what AI can and cannot do in a compliance context, where purpose-built tools outperform general-purpose ones, and how professionals can get reliable, defensible results from the AI they’re already using.

For compliance professionals weighing whether AI belongs in their workflows, the answer starts with understanding what separates AI built for their work from AI that wasn’t — and whether it meets the reasonable care standard trade professionals are held to.

The real question isn’t “Can AI help?” — it’s “Can I stand behind the answer?”

For most business functions, an AI tool that’s right 90% of the time is a productivity win. For trade compliance, that math doesn’t hold. A misclassified HS code, a missed tariff exclusion, or a regulatory research error doesn’t just create rework — it creates exposure. Penalties, delays, reputational risk, and in some cases, personal professional liability.

That’s not a reason to avoid AI. It’s a reason to choose it carefully.

In January 2026, U.S. Customs and Border Protection issued ruling HQ H350722 — the first time CBP directly addressed an AI tool in the context of customs business. The ruling drew a clear line. An AI tool can narrow a product to a six-digit HS heading. That’s permissible research assistance. But if AI is making the final ten-digit determination that goes into an entry — that’s customs business, and it cannot be performed by an unlicensed entity.

The implication for compliance professionals is direct: a human-in-the-loop isn’t optional. It’s a legal requirement. AI can accelerate the research and surface the analysis, but the licensed broker or importer of record makes the call and owns the determination.

This is the right framework — not just because CBP requires it, but because it reflects how the technology actually works best. The question worth asking about any AI tool isn’t only “Is it fast?” It’s “Does it support my ability to verify the output, meet the reasonable care standard, and stand behind it?”

What Fiduciary-Grade AI™ actually means (and why it matters for trade)

Fiduciary-Grade AI™ is the Thomson Reuters standard for how AI should perform in high-stakes professional environments — the kind where outputs influence legal judgments, financial disclosures, regulatory filings, or client advice.

That describes trade compliance. Fiduciary-Grade AI is defined by four principles, each of which carries real weight for trade professionals:

Trained on authoritative sources

General-purpose AI learns from the open web — blogs, forums, Wikipedia, and other publicly available text. That’s useful for many things, but it creates a problem for trade work: the model may not know what a specific tariff rate was on a given date, whether a CBP ruling has been superseded, or how a particular chapter note has been interpreted in practice.

Fiduciary-Grade AI draws on government publications, licensed regulatory databases, and domain-specific content curated by subject matter experts. Every material output traces back to a source a professional can locate, cite, and verify.

Every answer is cited

In a high-stakes workflow, the answer matters — but so does the trail behind it. When an AI tool surfaces a Federal Register notice, a CBP ruling, or an HS chapter note, it should link directly to that source. The professional clicks through, confirms what the document actually says, and documents that review.

That citation chain is an audit trail. It’s the difference between “we used AI to help research this determination” and “we can show you exactly what we relied on and how we verified it.”

Content that keeps pace with the work

Thomson Reuters made more than 155 million changes to its tariff databases in 2025 alone. If an AI tool’s underlying content refreshes quarterly — or even weekly — it can’t keep up with the regulatory pace that trade professionals work at every day. Fiduciary-Grade AI refreshes critical sources within one business day.

Data privacy by design

A compliance team’s classification data, supplier relationships, and entry history are proprietary. Fiduciary-Grade AI treats data privacy as a structural feature, not a configurable option. Customer data does not feed back into model training.

Together, these principles define AI that professionals can use in production — not just in experimentation. The standard isn’t about capability alone. It’s about whether the tool is built for environments where the consequences of being wrong are real.

Where ONESOURCE Global Trade proves its value

The two workflows where trade compliance professionals feel the most pressure — and where they’re most often asked to move faster than manual research allows — are HS classification and regulatory research. These are also where purpose-built AI delivers the clearest, most defensible value.

HS classification

Classification is a volume problem and a judgment problem at the same time. For standard products at scale, the challenge is consistency and speed. For edge cases — new composite materials, products that span multiple headings, items with ambiguous chapter notes — the challenge is knowing where the defensible answer lands and being able to show your work.

ONESOURCE Global Trade addresses both dimensions. Machine learning trained on classification data recognizes patterns across product databases to suggest HS codes. Agentic AI can then cross-reference chapter notes, relevant CBP rulings, and applicable tariff modifications, returning a cited, reviewable result for the professional to evaluate.

That last step — professional review — isn’t a formality. It’s where CBP’s January 2026 ruling places the line. AI does the research and surfaces the analysis. The licensed broker or importer of record validates the output, applies professional judgment, and makes the final determination. That’s what reasonable care looks like with AI assistance.

Regulatory research

When tariff policy shifts, the questions come fast. What changed? When does it take effect? Which of our products are affected? What’s the new landed cost? What documentation does CBP expect to see?

Answering those questions manually — working through Federal Register notices, CSMS messages, executive orders, and CBP guidance — can take hours. And in a tariff environment where meaningful changes are arriving multiple times a week, those hours add up.

ONESOURCE Global Trade Research powered by CoCounsel is an AI-driven research tool built specifically for this work. It synthesizes Federal Register notices, CBP rulings, HS data, and executive orders across sources and returns cited answers the professional can verify against primary documents. Moxon noted during the webinar that it has passed the U.S. Customs Broker examination six times — a useful proxy for the depth of its regulatory training.

Consider a practical example from the webinar that introduced these capabilities: a compliance team needs to identify all active Section 301 tariff exclusions for lithium-ion batteries imported from China. A general-purpose AI might offer background on Section 301 exclusions broadly. An agentic AI built for this work searches the Federal Register for relevant exclusions, cross-references them against the current HS, checks expiration dates, and returns a list of active exclusions with citations — a task that could occupy an afternoon, delivered in minutes.

The cite-and-verify step still belongs to the professional. But the time required to get to that step drops sharply.

Global Classification and Global Trade Research — better together

Global Classification powered by CoCounsel, and Global Trade Research powered by CoCounsel address different parts of the same compliance workflow — and they’re more useful in combination than separately.

Global Classification handles classification at scale. It ingests product descriptions, applies machine learning to suggest HS codes, incorporates chapter notes and applicable rulings, and surfaces results for professional review. It’s built for volume, consistency, and the kind of defensible output that holds up when CBP asks questions.

Global Trade Research handles the research layer. When a regulatory change hits and a compliance team needs to understand what it means, which products it affects, and what response it requires, Global Trade Research is where that synthesis happens — quickly, with cited sources, and across the full range of relevant regulatory inputs.

In practice, the two tools cover the full loop. A team uses Global Classification AI-generated similarity scores to help speed their classification at scale, then turns to Global Trade Research to understand the regulatory context around specific headings, verify edge cases, and pressure-test its reasoning before it goes into an entry. When a tariff change lands, Global Trade Research surfaces the implications; Global Classification helps the team work through any reclassification the change requires.

The professional moves between research and classification as the work demands. The tools are built to support that flow.

How you use AI is part of the defensibility equation

The quality of what an AI tool returns depends heavily on what you ask it. A well-structured prompt to a purpose-built AI tool returns a precise, cited, actionable answer. A vague prompt to the same tool returns something you’ll spend time cleaning up — or worse, something that looks right but isn’t.

A few key practices make the difference.

• Brief the AI the way you’d brief a skilled analyst. Provide the product description, the HTS code if you have it, the country of origin, the transaction type, and the time frame you’re asking about. The more relevant context you include upfront, the more precise the answer — and the less time you spend iterating.
• Anchor your query to a specific date. Tariff rates can change materially week to week. “As of today” and “as of January 15, 2025” may return different answers, and in compliance work, the difference is not academic.
• Review every citation before it informs a determination. This is the step that keeps AI-assisted research defensible. Click through to the source document. Confirm the ruling exists and says what the AI says it says. It takes thirty seconds and catches errors before they become problems.
• Build an audit trail as you go. Capture the query you ran, the answer the AI returned, the citations you reviewed, and the determination you made. That record is your defense in an audit — it demonstrates that AI assisted your research and that a qualified professional validated the output, applied judgment, and made the call. That satisfies reasonable care. Relying on an AI answer without that validation does not.

The standard that matches the trade compliance stakes

Trade compliance has always operated under a higher standard than most business functions. The importer of record is legally responsible for reasonable care under 19 U.S.C. § 1484. The licensed customs broker is professionally accountable for the work performed on a client’s behalf. Errors have consequences that go well beyond a corrected spreadsheet.

AI doesn’t change that. What it changes is the speed at which professionals can research, analyze, and validate — which matters enormously when the regulatory environment is moving as fast as it is right now.

But speed without rigor isn’t a gain — it’s a different kind of risk. The AI tools worth using in this environment are the ones built to support professional judgment, not sidestep it. They cite their sources. They refresh their content at the pace the work requires. They’re designed so that the professional can examine, explain, and defend every output.

That’s the standard Thomson Reuters builds to — Fiduciary-Grade AI, designed for professionals who carry real accountability for the work they do.

When evaluating any AI tool for trade compliance work, the questions worth asking are straightforward: What was it trained on? Does every answer come with a citation? How current is the underlying content? What happens to your data? Does it produce audit-ready outputs? Was it built by people who know this work, or adapted from something built for something else?

The answers will tell you whether the tool is built for experimentation or built for production.

See it in action