IN Brief:
- A White House report proposes an AI-enabled “Detective Border” system to support customs targeting of suspected illegal transshipment.
- The model would combine shipment, routing, ownership, classification, production-capacity, anomaly, and computer-vision data.
- The report explicitly says enforcement needs to distinguish suspected pass-through trade from legitimate nearshoring and foreign investment.
US Customs and Border Protection could make greater use of linked trade data and artificial intelligence to identify shipments presenting a higher risk of tariff-evasion transshipment, under a White House proposal for an AI-enabled customs targeting system.
The proposed system, referred to as “Detective Border”, would combine shipment data, routing histories, product classifications, company ownership relationships, production-capacity indicators, anomaly detection, and computer vision to identify cargo requiring closer enforcement attention.
The concept is aimed at goods routed through third countries where the declared origin may not reflect where substantive production occurred. Practices examined by the report include relabelling, repackaging, reinvoicing, false origin declarations, and limited processing intended to present goods as originating in a jurisdiction facing lower US tariffs.
The important qualification is that a changed trade route is not evidence of customs fraud by itself. Manufacturers regularly move sourcing, establish new factories, add suppliers, and restructure production networks in response to tariffs, labour availability, geopolitical exposure, freight costs, and customer demand.
The White House report acknowledges that distinction explicitly. Its proposed analytical approach is intended to help customs officers separate legitimate nearshoring and foreign investment from suspected pass-through trade rather than assuming that every increase in exports from an intermediary country represents illegal transshipment.
That makes the quality of the underlying data critical. Shipping records can show where a consignment travelled, but origin rules may depend on what processing took place, the materials used, and whether manufacturing in a particular country met the legal threshold for substantial transformation.
Ownership information can provide another signal. A newly established exporter connected to a producer in a higher-tariff jurisdiction may warrant greater examination when combined with unusual trade growth, but the corporate relationship alone does not determine the legal origin of the goods.
Production capacity adds a different test. If recorded exports of a particular product rise far beyond a country’s plausible manufacturing capability, authorities can use that discrepancy to decide where further documentary or physical investigation is justified.
Routing history can be analysed in the same way. Repeated movement through bonded zones or intermediary ports can be commercially legitimate where goods are consolidated, stored, or genuinely processed, but unusual patterns become more significant when paired with inconsistent values, classifications, origin documents, or corporate links.
Artificial intelligence is therefore better suited to prioritising cases than making the final customs decision. Automated models can analyse volumes of bills of lading, manifests, classifications, trade histories, and ownership data that would be impractical for officers to review manually, then direct human attention towards the strongest combinations of risk signals.
CBP already uses machine learning and analytical tools within targeting and inspection work. The proposed Detective Border architecture would extend that approach by linking a wider range of supply-chain data and applying network analysis across companies, cargo movements, and production locations.
For importers, the practical consequence would be greater pressure on upstream documentation. Businesses sourcing through several countries may need stronger evidence showing where production occurred, which supplier performed each stage, how origin was determined, and why the declared tariff treatment is supported by the manufacturing facts.
Procurement teams may therefore become more closely involved in customs compliance. Supplier selection, bill-of-materials changes, subcontracting, and manufacturing relocation can all affect origin, yet those decisions are often made well before a customs broker receives the documents required to enter the shipment.
Better supplier mapping and traceability would also help legitimate businesses respond to an automated risk flag. A company that can document its production chain, facilities, processing steps, and commercial relationships is better placed to distinguish a genuine sourcing shift from the routing patterns an enforcement model is designed to detect.
The report includes several estimates of potential transshipment exposure, ranging from about $40 billion to $303 billion depending on methodology and definition. It states that those estimates are not additive or directly comparable, an important limitation when assessing the scale of the issue.
That same caution applies to AI targeting. Poorly calibrated models could generate false positives, delaying legitimate cargo and increasing documentation work without improving enforcement. A useful system has to narrow the inspection pool more accurately than existing methods rather than simply creating another automated reason to hold shipments.
The proposed system is therefore as much a data-governance challenge as an AI project. Its value will depend on reliable source information, transparent risk criteria, legal review, and the ability to convert analytical signals into proportionate customs action while leaving legitimate supply-chain diversification able to function.
