NextFin News - Washington’s tariff agenda is moving from setting duty rates to proving who owes them. The administration is developing AI-enabled customs enforcement aimed at detecting import patterns that can mask tariff evasion, a shift that matters because a tariff changes behavior only when officials can connect a shipment’s paperwork to its real origin, ownership and supply chain. Executive Order 14411, signed June 3, gives that enforcement drive a timetable: 90 days for new transparency measures and 180 days for revised importer-eligibility rules.
The immediate story is technology, but the financial question is institutional. Artificial intelligence cannot make a tariff less costly at the border; it can make evasion harder, compliance more data-intensive and the effective tariff burden less negotiable for businesses whose sourcing models depend on opaque intermediaries. That is a structural policy change, not simply another cyclical swing in customs collections. The key uncertainty is execution: automated risk scoring can concentrate attention, but it cannot by itself establish country of origin, substitute for evidence or turn a suspicious pattern into a final legal finding.
As of Aug. 13, 2026, the administration has paired the policy order with a broader customs modernization effort. CBP said in January that it had allocated more than $1 billion from the One Big Beautiful Bill Act for non-intrusive inspection systems, artificial intelligence and other mission-support capabilities. The President’s fiscal 2026 budget separately requested $137 million for additional systems and enhanced non-intrusive inspection capabilities. CBP has also published artificial-intelligence records that include illicit-trade model-card materials dated July 2025 and an AI Operations and Governance directive dated November 2025. The public record establishes an AI-governance and illicit-trade documentation workstream; it does not establish a model’s accuracy or an AI-specific recovery total.
The compliance prize is material even without such a total. From Jan. 20 through Aug. 8, 2025, CBP said its Enforce and Protect Act investigations uncovered more than $400 million in unpaid trade duties and identified 89 cases with reasonable suspicion of duty evasion. Its largest case at the time covered 23 U.S. importers and shell-company networks routing goods through Indonesia, South Korea and Vietnam; CBP identified more than $250 million in revenue for collection. Those are enforcement findings, not cash collected and not an estimate of all evasion. They nonetheless show why trade officials see data linkage, rather than another published tariff schedule, as the policy bottleneck.
The tension is straightforward. Tariffs can create a large incentive to alter paperwork, routing or the identity of an importer. Yet enforcement agencies face an opposite constraint: the volume and complexity of trade entries make manual review selective. The administration’s bet is that better targeting, combined with tougher rules for importers and intermediaries, can narrow that gap. Its success should be judged on upheld cases and durable collections, not on the number of algorithms announced.
The Enforcement Mechanism Is Data, Not Just Detection
The central judgment is that AI changes tariff administration only when it joins disparate records into a usable risk signal before clearance. A customs entry contains a tariff classification, declared value, importer and origin fields. Those fields are necessary, but they can be incomplete, inconsistent or strategically selected. The enforcement value comes from comparing entries with shipment histories, company relationships, product attributes, routing patterns and documentary inconsistencies, then directing human investigators toward transactions least likely to withstand scrutiny.
That is a different task from setting a tariff rate. A rate is a published policy parameter. Enforcement is an evidence problem. For origin-sensitive duties, an officer must assess whether the declared country of origin matches the underlying production and transformation of the good. For undervaluation, the agency must test a declared price against the transaction and surrounding records. For transshipment, it must determine whether a change in route reflects a genuine supply-chain change or an attempt to disguise the origin of merchandise. These questions become network questions when one importer, several intermediaries and multiple countries are involved.
EO 14411 matters because it addresses the information and accountability sides of that network. It directs DHS to revise importer-eligibility regulations, guidance and policies within 180 days, and to enhance customs transparency within 90 days, including through annual enforcement-transparency reports. The order also calls for enforcement of liquidated-damages claims against bonds, restrictions on in-bond utilization, increased audits and maximum penalties for brokers in specified failures of due diligence or cooperation. DHS and the Justice Department are directed to prioritize cases involving misclassification, undervaluation and illegal transshipment, including EAPA investigations.
The order’s more granular provisions reinforce the same mechanism. DHS is directed within 180 days to update the importer-of-record registry, including removal of inactive importers, confirmation of compliance with regulations and disclosures, and risk-based tiers tied to compliance history, enforcement actions and audit results. It also calls for enhanced and recurrent vetting of parties directly involved in imports, including foreign importers, affiliates, customs brokers, custodians of bonded merchandise and freight forwarders. Within 90 days, it calls for steps to expedite and enhance seizure and disposal of non-compliant imports, including higher bond requirements for high-risk shipments where permitted by law.
That combination creates a feedback loop. Better data can identify suspicious entries; tighter importer and broker standards can require more verifiable data; audits, bond claims and penalties can raise the cost of supplying weak data. Each element makes the others more useful. A model flag is less valuable if the importer has no durable records or assets. A higher penalty is less valuable if enforcement teams cannot identify the right entries. The policy is trying to close both gaps at once.
“Never before has CBP identified this many importers evading AD/CVD in a single consolidated EAPA investigation,” Susan S. Thomas, then acting Executive Assistant Commissioner for CBP’s Office of Trade, said in the agency’s 2025 announcement.
The 23-importer investigation demonstrates why the mechanism is not reducible to a dockside inspection. CBP said the case involved shell-company networks and goods routed through Indonesia, South Korea and Vietnam. Whether particular entries evaded duties required a conclusion about the relationships among merchandise, companies and declared origins over time. AI can help triage patterns across a large record set. It cannot eliminate the legal need to test evidence, give parties the process required by law and reach an agency determination.
The distinction matters for businesses. A stronger risk screen can produce more requests for information, audits, holds, bond demands or investigation referrals before a final determination produces any collection. For a compliant importer, the relevant advantage is not immunity from review; it is the ability to produce coherent evidence quickly. For a noncompliant importer, the risk is not only a tariff bill. It is the loss of a business model built on the assumption that records held by different parties will never be connected.
CBP’s stated technology objective is not solely friction. In January testimony, the agency said artificial intelligence would reduce manual image analysis and help increase security while enhancing the flow of legitimate trade and travel. That creates a two-track commercial outcome. Traders with consistent, well-documented supply chains could be positioned for more efficient processing as the system matures. Traders whose records cannot reconcile product, origin and ownership face a deeper information burden. The policy divides enforcement cost by evidence quality, not simply by product category.
The first-order effect is easier to see: suspected evasion can be flagged more efficiently. The second-order effect is more consequential: procurement, customs brokerage, logistics and supplier selection become part of tariff risk management. A sourcing move that appears to diversify production can become costly if the importer cannot document substantial transformation, ownership transparency and consistent product records. That is the step beyond the familiar tariff story.
A Structural Shift With Cyclical Enforcement Results
The long-term change is structural because it is rooted in rules, systems and accountability requirements that do not naturally reverse when trade flows normalize. The short-term results will remain cyclical. Cases take time, tariff schedules can change, import volumes rise and fall, and counterparties adapt. Treating a single recovery figure or enforcement announcement as proof of permanent revenue growth would confuse the cycle with the regime.
Three pieces of evidence support the structural call. First, the June order sets administrative deadlines and a one-year effectiveness-report requirement rather than a temporary operational campaign. Second, CBP’s published records show an established AI-governance and illicit-trade documentation workstream, not a one-off experiment announced after a single case. Third, CBP has allocated more than $1 billion for new inspection systems, AI and other mission-support capabilities, while the fiscal 2026 budget requested another $137 million for additional inspection systems and enhanced capabilities. Capital systems, data practices and importer rules generally persist longer than any one tariff schedule.
The cyclical qualification is equally important. EAPA was created by the Trade Facilitation and Trade Enforcement Act signed in February 2016. Its published timetable runs from initiation at day 0 to interim measures at day 90, a voluntary-submission cutoff at day 200, written arguments at day 230 and final determinations around day 300 or 360. That lag means a detection count is not an instant duty receipt. The 2025 data make the point: CBP’s more than $400 million figure covered Jan. 20 through Aug. 8, while it said the more than $250 million amount identified in its largest case could rise as that investigation expanded. Detection, final determination and collection do not move in lockstep.
A third comparison is operational. Traditional inspections depend on physical capacity and sample selection. Non-intrusive inspection systems add imaging. AI-assisted trade analysis can rank entries before or alongside physical checks. Each layer can improve targeting, but none makes verification unnecessary. The structural development is the combination of data analysis, imaging, registry requirements, recurrent vetting and penalty tools. The cyclical development is the changing flow of shipments, alerts, holds and successful cases through that system.
This separation yields a more useful base case. The durable outcome is a higher compliance fixed cost for importers whose supply chains cannot produce evidence rapidly. The variable outcome is how much revenue CBP ultimately collects in a given year. The first can alter vendor selection, working-capital needs and broker diligence even when the second is volatile. Documentation becomes an operating asset rather than a back-office afterthought.
There is a parallel benefit for firms with credible traceability. A 2025 company announcement by Altana said CBP had selected its AI-powered Product Passports to improve enforcement efficiency and customs processing through visibility and traceability. The company said detailed supply-chain information could be provided before customs filings and used for continuous compliance monitoring. That is a vendor statement, not an agency performance audit. But it captures the policy direction: trusted-trader advantages are increasingly tied to the ability to demonstrate, rather than simply declare, the facts supporting an entry.
The Market Is Not Pricing a Single Tariff Rate
There is no defensible single market-implied probability or consensus tariff outcome to cite for an enforcement program, and pretending otherwise would blur policy risk with a priced security. The more relevant expectation gap is qualitative. Tariff debates often treat the announced rate as the endpoint. Enforcement technology turns it into the start of a process whose cost is distributed unevenly across supply chains.
The first-order exposure falls on importers of goods subject to trade remedies or country-specific tariffs, and on intermediaries whose entries depend on weak origin evidence. Their costs can emerge through delayed clearance, higher bond requirements, audits, broker diligence, legal review and potential penalties before they appear as a line labeled tariff. The second-order transmission runs into inventory planning and margins. A company that needs more buffer stock to absorb inspections, or must replace a supplier because documentation is inadequate, can face a higher cash conversion cycle even when the statutory duty rate has not changed.
The third-order effect reaches purported tariff workarounds. Transshipment’s economic value rests on a gap between a product’s underlying origin and the origin claimed on entry. As enforcement teams connect trade records, factory relationships and product-level information, the expected value of that gap can decline. The policy can then affect trade not only by making imports more expensive, but by making opaque routing less financeable. Lenders, insurers, brokers and logistics providers all have incentives to avoid transactions whose documentation repeatedly produces risk alerts.
For domestic producers competing with imports, the potential benefit is not simply a higher published barrier. It is more consistent application of existing duties. For customs brokers, software providers and trade-data specialists, demand can rise for origin analysis, recordkeeping and pre-entry verification. For retailers, manufacturers and distributors with fragmented global sourcing, the outcome is more nuanced: robust traceability can be an advantage, but systems integration, supplier remediation and inventory buffers can be costly. The policy creates winners and losers within sectors because the relevant variable is evidence quality, not only import dependence.
The strongest counter-thesis is that this is enforcement rebranding rather than a regime change. Customs has long used data analysis; EAPA has existed since 2016; final determinations still need investigators, documentation and due process. An AI system can generate false positives, burden compliant importers and shift trade flows without producing materially more final findings or collected duties. The EAPA timetable supports the core of that critique: determinations can take about 300 or 360 days, so more flags cannot be treated as more cash.
That objection limits claims about near-term revenue, but it does not defeat the structural thesis. The policy does not rely on a model alone. It combines technology spending, importer-registry changes, recurrent vetting, broker obligations, audit authority, bond tools, transparency deadlines and a directive to prioritize illegal transshipment. If those elements operate together, the relevant outcome is not merely more alerts; it is a higher cost of maintaining an unverifiable supply chain.
The structural call has a clear falsifier. After the 90-day and 180-day reforms are operating, CBP’s annual transparency reports should show whether risk-targeted examinations, EAPA case initiations or duty collections rise on a sustained basis relative to the agency’s pre-reform baseline. If the reports show no sustained improvement in those measures, while compliant trade faces more delay or cost, the thesis that the new architecture materially strengthens enforceability is wrong. A system that creates alerts but not enforceable outcomes is an expensive filter.
What Changes Next for Supply Chains and Trade Risk
The short-term horizon is operational. Importers, brokers and freight intermediaries will focus on data completeness, origin documentation and their ability to answer a customs inquiry quickly. That can produce localized delays and compliance expense before a system generates a visible national collection result. Firms with established supplier mapping may be better positioned for a risk-based environment. Firms that rely on layered intermediaries can discover that a low purchase price did not include the full cost of proving origin.
Over the medium term, the meaningful measure is enforcement conversion, not the number of AI tools or policy announcements. The annual transparency reports required by EO 14411, EAPA initiations, interim measures, final determinations, amounts identified for collection and published audit or hold data will provide a more rigorous scorecard. A rise in alerts without a rise in upheld cases would support the false-positive critique. More timely, durable findings paired with repeatable documentation standards would support the case that enforcement capacity has improved.
Over the long term, the structural question is whether supply chains become more legible to government systems. If they do, tariffs can influence production decisions more effectively because shifting a shipment’s route is no longer enough to change its enforcement risk. That does not mean tariffs automatically restore domestic capacity. Labor availability, capital costs, input economics and demand still determine where production settles. It means the policy’s practical barrier may be traceability, which is harder to arbitrage than a published duty rate.
The base case is gradual implementation: data-driven targeting raises documentation standards and redistributes clearance risk, while enforcement results remain uneven because cases and collections lag. The upside case for the administration’s objective is that risk models and importer rules produce sustained EAPA findings and duty collections without materially slowing lawful trade. The downside case is that false positives, uneven data or legal challenges create friction without improving final outcomes. The annual reporting requirement and EAPA results should distinguish those paths.
This is not a forecast that AI will solve tariff evasion. It is a judgment that the enforcement architecture can make tariffs more durable by raising the cost of opaque compliance. If alerts do not convert into upheld cases and collections, the policy will be an expensive filter. If they do, the tariff regime will be enforced through supply-chain data, not just border rates.
Explore more exclusive insights at nextfin.ai.
