Prompt injection in customs and trade compliance AI

Four adversarial injection surfaces in customs and trade compliance AI

1. Commercial invoice and packing list image injection (CBP 19 USC §1592, 18 USC §542)

Commercial invoice and packing list AI processes customs entry commercial invoice document scan and OCR display images, packing list unit count and weight reconciliation display images, HTS tariff schedule classification and duty rate display images, customs entry value and dutiable customs value determination display images, invoice first sale and related party transaction pricing disclosure display images, and AD/CVD order-covered merchandise description and quantity display images from Flexport AI at 10,000+ global brand importer operations processing commercial invoice, packing list, and customs clearance document images through AI-assisted HTS tariff classification and customs entry preparation tools; Thomson Reuters ONESOURCE Global Trade AI at 3,000+ multinational corporation operations processing commercial invoice and customs valuation documentation through AI-assisted landed cost calculation and customs duty optimisation tools; Descartes Systems AI at 26,000+ global supply chain and logistics operations processing commercial invoice OCR and customs classification document images through AI-assisted customs entry accuracy verification and duty calculation tools; and customs broker AI platforms including Expeditors International AI, CH Robinson AI, and Kuehne+Nagel AI at freight forwarding and customs brokerage operations processing commercial invoice and packing list document scan images through AI-assisted customs entry preparation and tariff classification accuracy tools — extracting HTS tariff classification accuracy determinations and customs entry dutiable value compliance assessments from commercial invoice and packing list document scan and OCR display image inputs in AI-assisted customs entry preparation and compliance verification pipelines at multi-modal freight forwarding operations volumes that make individual human customs specialist review of every AI-processed commercial invoice impracticable for large customs brokerage and freight forwarding operations.

The adversarial injection surface is the commercial invoice document scan or packing list OCR display image submission pathway: Flexport AI or Descartes AI commercial invoice and packing list document scan display images submitted through AI-assisted HTS tariff classification and customs entry dutiable value accuracy verification tools for AI customs entry preparation record generation and CBP entry filing input. An adversarially crafted commercial invoice scan image — in which pixel perturbations applied to the invoice unit price and declared customs value indicator display region, the HTS tariff classification accuracy and duty rate visual marker, or the AD/CVD order-covered merchandise description and scope quantity display in a commercial invoice document image cause the AI to classify a commercial invoice evidencing a dutiable value understatement or HTS classification inaccuracy as a compliant commercial invoice not meeting CBP entry accuracy challenge criteria when the actual invoice document evidences a dutiable value discrepancy or tariff misclassification with 19 USC §1592 material false statement civil penalty significance — can suppress a dutiable value or classification indicator that would otherwise generate a CBP entry accuracy review referral, an importer of record duty liability correction, and a customs compliance audit record. In customs brokerage and freight forwarding operations where Flexport AI or Thomson Reuters ONESOURCE AI processes thousands of commercial invoice document images per day without individual human customs specialist examination of every AI-processed invoice before the AI classification governs the CBP entry filing, adversarial suppression of dutiable value and tariff classification accuracy indicators creates 19 USC §1592 civil penalty and 18 USC §542 criminal exposure dimensions.

The 19 USC §1592, 18 USC §542, CBP 19 CFR §163, and CTPAT consequences of adversarially suppressed customs entry accuracy classification in commercial invoice AI span 19 USC §1592 material false statement, act, or omission in entry documentation creating CBP civil penalty authority of up to four times the unpaid duties for fraud, or up to two times the unpaid duties for gross negligence; 18 USC §542 entry of goods by means of false statements creating criminal penalties including imprisonment up to 20 years for knowingly and wilfully making false statements in customs entry documents; CBP 19 CFR §163 recordkeeping requirements applicable to importers of record obligating preservation and production of entry documentation for five years; CBP CTPAT programme security profile requirements applicable to importers and customs brokers that include supply chain security controls for commercial documentation integrity; and 19 USC §1595a civil penalty and forfeiture authority applicable to merchandise imported by means of a false invoice or entry. 19 USC §1592 imposes a tiered civil penalty framework based on culpability level: penalties of up to four times the unpaid duties for fraud, two times for gross negligence, and the unpaid duty amount for negligence — adversarial manipulation of commercial invoice AI that suppresses a dutiable value understatement indicator creates §1592 civil penalty dimensions when the adversarially corrupted AI classification permits CBP entry filing with an understated dutiable value. 18 USC §542 creates criminal exposure for importers who knowingly and wilfully make false statements or material omissions in customs entry documents filed with CBP; adversarially crafted commercial invoice images designed to suppress value misdeclaration indicators in AI-assisted entry preparation create §542 criminal intent dimensions when the adversarial manipulation is performed with knowledge that it will cause the entry filer to submit false customs documentation to CBP. Threshold: 60 for commercial invoice and packing list AI — reflecting 19 USC §1592 material false statement civil penalty, 18 USC §542 customs entry fraud, CBP 19 CFR §163 recordkeeping, CTPAT documentation integrity, and 19 USC §1595a forfeiture dimensions.

2. Export control document image injection (EAR 15 CFR Part 730, ITAR 22 CFR Part 120)

Export control document AI processes export control classification number (ECCN) determination and BIS export licence document scan images, US Munitions List (USML) commodity jurisdiction determination display images, end-use and end-user certificate document scan and signature verification display images, BIS export licence application and licence exception eligibility display images, denied party and restricted party screening result display images, export shipment commercial invoice and export declaration (EEI Electronic Export Information) document scan images, and re-export authorisation and foreign national national security employee access determination display images from Thomson Reuters ONESOURCE Global Trade AI at multinational corporation operations processing ECCN determination, export licence management, and denied party screening document images through AI-assisted export control compliance classification tools; Amber Road AI (E2open) at global supply chain operations processing export licence application, USML commodity jurisdiction, and FMS foreign military sales document images through AI-assisted export compliance classification tools; Descartes Systems AI at freight forwarder and customs broker operations processing EEI Electronic Export Information filing and denied party screening result images through AI-assisted export compliance monitoring tools; and SAP Global Trade Services AI at enterprise manufacturing and distribution company operations processing export control classification and licence management document images through AI-assisted EAR and ITAR compliance classification tools — extracting ECCN commodity classification and BIS/DDTC export licence requirement determinations from export control document scan and display image inputs in AI-assisted EAR and ITAR export compliance management pipelines.

The adversarial injection surface is the ECCN determination display image or BIS export licence document scan image submission pathway: Thomson Reuters ONESOURCE AI or Amber Road AI export control document scan display images submitted through AI-assisted ECCN classification and export licence requirement identification tools for AI export compliance determination record generation and export shipment authorisation input. An adversarially crafted ECCN determination display or end-use certificate scan image — in which pixel perturbations applied to the ECCN classification and controlled country destination indicator display region, the BIS export licence requirement and licence exception eligibility visual marker, or the USML category designation and ITAR-controlled article determination display in an export control document image cause the AI to classify an export shipment to a controlled country or controlled end-user destination as licence-except or EAR99 not meeting BIS export licence application criteria when the actual export control document scan evidences an ECCN classification and destination combination requiring BIS licence under EAR 15 CFR Part 730 — can suppress an export licence requirement indicator that would otherwise generate a BIS export licence application referral, an export hold and compliance review determination, and an export control programme compliance record. In enterprise manufacturer and freight forwarder export compliance operations where AI processes thousands of export control classification and end-use certificate document images per day without individual human export compliance specialist pixel-level examination of every AI-processed document before the AI classification governs the export authorisation and EEI Electronic Export Information filing, adversarial suppression of export licence requirement and USML controlled article indicators creates EAR and ITAR unlicensed export criminal dimensions.

The EAR 15 CFR Part 730, ITAR 22 CFR Part 120, OFAC, and BIS criminal enforcement consequences of adversarially suppressed export licence requirement classification in export control document AI span EAR 15 CFR Part 730 through 774 export control requirements including ECCN classification, BIS export licence application requirements for EAR-controlled dual-use items to controlled country or restricted end-user destinations, and Export Control Reform Act of 2018 (ECRA) P.L. 115-232 enhanced criminal penalties of up to $1 million per violation or imprisonment up to 20 years; ITAR 22 CFR Part 120 through 130 export control requirements including USML category commodity jurisdiction determination requirements, DDTC export licence application requirements for defence articles to foreign persons and foreign countries, and AECA Arms Export Control Act 22 USC §2778 criminal penalties of up to $1 million per violation or imprisonment up to 20 years; OFAC 31 CFR economic sanctions regulations applicable to exports, re-exports, and transfers to sanctioned countries, entities, and individuals including Cuba, Iran, North Korea, Syria, and Russia under the comprehensive country programmes and SDN list targeted sanctions; and BIS and DDTC administrative debarment authority applicable to exporters with repeated or wilful EAR and ITAR violation patterns. ECRA Section 1753 enhanced criminal penalties for wilful EAR violations and AECA 22 USC §2778(c) criminal penalties for ITAR violations create the highest-consequence enforcement exposure in the customs and trade compliance regulatory framework; adversarial manipulation of export control document AI that suppresses ECCN-controlled destination licence requirement or USML commodity jurisdiction indicators creates EAR ECRA and ITAR AECA criminal liability dimensions when the adversarially suppressed export control requirement enables unlicensed export shipments to BIS-controlled country or DDTC-controlled foreign national destinations. BIS has issued denial orders, civil penalties exceeding $300 million, and criminal prosecution referrals for EAR violations including exports to Iran, China, and Russia of dual-use items with ECCN designations requiring BIS licences; adversarially corrupted export control AI that suppresses licence requirement indicators creates the systematic unlicensed export exposure that BIS SNAP-R export licensing and post-shipment verification programmes are designed to prevent. Threshold: 55 for export control document AI — reflecting EAR ECRA criminal penalties up to $1M/violation, ITAR AECA criminal penalties up to $1M/violation, OFAC sanctions programme compliance, BIS debarment authority, and DDTC administrative action dimensions.

3. Certificate of origin and FTA rules of origin image injection (USMCA, GSP, 19 USC §2461)

Certificate of origin and FTA rules of origin AI processes USMCA certificate of origin and regional value content (RVC) calculation display images, GSP Form A certificate of origin document scan and qualifying country content determination display images, EU-US bilateral free trade agreement EUR.1 movement certificate scan images, China factory inspection and non-preferential certificate of origin document scan images, manufacturer’s affidavit and FTA production qualification declaration document scan images, customs ruling and advance ruling request response display images, and tariff preference level (TPL) and tariff rate quota (TRQ) allocation usage display images from Flexport AI at importer operations processing USMCA certificate of origin and FTA qualification document images through AI-assisted duty preference eligibility and rules of origin compliance classification tools; Thomson Reuters ONESOURCE AI at multinational corporation operations processing FTA qualification, certificate of origin, and supply chain country of origin management document images through AI-assisted global FTA duty optimisation and compliance classification tools; Amber Road AI (E2open) at global supply chain operations processing certificate of origin and FTA rules of origin qualification document images through AI-assisted FTA duty preference management and compliance tools; and MIC Customs Solutions AI and Precision Software AI at customs broker and freight forwarder operations processing USMCA, CAFTA-DR, Korea FTA, and Japan FTA certificate of origin and production qualification document scan images through AI-assisted preferential duty eligibility and rules of origin compliance classification tools — extracting FTA preferential duty eligibility classifications and rules of origin qualification determinations from certificate of origin and FTA production qualification document image inputs in AI-assisted preferential duty programme management and customs entry optimisation pipelines.

The adversarial injection surface is the USMCA certificate of origin display image or GSP Form A document scan image submission pathway: Flexport AI or Thomson Reuters ONESOURCE AI certificate of origin document scan display images submitted through AI-assisted FTA rules of origin qualification and preferential duty eligibility classification tools for AI FTA duty preference programme record generation and customs entry preferential duty claim input. An adversarially crafted USMCA certificate of origin or GSP Form A document scan — in which pixel perturbations applied to the USMCA regional value content percentage indicator display region, the GSP qualifying country product and sum of cost content visual marker, or the tariff change rule compliance determination display in a certificate of origin document image cause the AI to classify a non-qualifying good as an FTA-qualifying good meeting USMCA or GSP preferential duty eligibility criteria when the actual certificate of origin evidences insufficient RVC or sum of cost content to qualify under the applicable FTA rules of origin — can suppress a rules of origin non-qualification indicator that would otherwise generate an FTA duty preference claim denial, a customs entry duty payment correction, and an FTA compliance audit record. In importer and customs broker FTA duty preference programme operations where Flexport AI or ONESOURCE AI processes thousands of certificate of origin and FTA qualification document images per day without individual human trade compliance specialist examination of every AI-processed certificate of origin before the AI classification governs the FTA duty preference claim and customs entry filing, adversarial suppression of rules of origin non-qualification indicators creates GSP 19 USC §2461 fraudulent preferential duty claim and USMCA rules of origin evasion dimensions.

The USMCA 19 CFR Part 182, GSP 19 USC §2461, CBP 19 USC §1592, and CBP enforcement consequences of adversarially suppressed rules of origin qualification classification in certificate of origin AI span USMCA agreement rules of origin requirements under 19 CFR Part 182 Annex A establishing tariff change rule, regional value content (RVC), and net cost calculation requirements for USMCA preferential duty eligibility, GSP 19 USC §2461 generalised system of preferences duty-free treatment eligibility requirements including direct import, sum of cost 35% content threshold, and beneficiary developing country status requirements, US-Korea FTA and US-Japan TAFTA agreement rules of origin applicable to manufactured goods imported from Korea and Japan under FTA preferential duty programmes, and CBP 19 USC §1592 civil penalty authority and 18 USC §542 criminal penalty authority applicable to fraudulent FTA preferential duty claims based on false certificate of origin representations. CBP USMCA trade benefit programme enforcement under 19 USC §1592 targets importers who claim USMCA tariff preference for goods that do not meet USMCA rules of origin; civil penalties of up to four times the unpaid duties may be assessed for fraudulent USMCA origin misrepresentation. CBP GSP enforcement targets importers who claim GSP duty-free treatment for goods that do not satisfy the 35% sum of cost content threshold or direct import requirements; fraudulent GSP origin claims create 19 USC §1592 civil penalty and potential duty fraud criminal exposure under 18 USC §542. CBP has authority under 19 USC §1508 to require importers to produce supporting documentation for FTA preferential duty claims including producer certifications, manufacturing cost records, and BOM materials sourcing documentation for five years; adversarially corrupted AI certificate of origin classifications that suppress rules of origin non-qualification indicators create CBP §1508 documentation production obligation compliance failures. Threshold: 60 for certificate of origin and FTA rules of origin AI — reflecting USMCA 19 CFR Part 182 rules of origin compliance, GSP 19 USC §2461 content requirements, CBP 19 USC §1592 FTA duty preference fraud, CBP 19 USC §1508 documentation requirements, and 18 USC §542 customs entry fraud dimensions.

4. Customs tariff classification and AD/CVD entry image injection (19 USC §1592, anti-dumping duties)

Customs tariff classification and AD/CVD entry AI processes HTS tariff schedule classification and duty rate binding ruling display images, CBP Form 7501 entry summary and tariff classification determination display images, AD/CVD order scope determination and antidumping duty rate assignment display images, CBP post-entry amendment and prior disclosure filing display images, customs valuation first sale and transaction value determination display images, CBP Census Bureau SIC code and export statistical classification display images, and CBP drawback eligibility and duty refund calculation display images from Thomson Reuters ONESOURCE AI at multinational importer operations processing HTS tariff classification management and AD/CVD order compliance document images through AI-assisted classification accuracy and duty optimisation tools; Descartes Systems AI at customs broker operations processing CBP Form 7501 entry summary and tariff classification review images through AI-assisted classification compliance and duty payment accuracy tools; Flexport AI at importer of record operations processing commercial invoice and HTS classification determination document images through AI-assisted customs entry accuracy and AD/CVD scope analysis tools; and customs compliance software AI platforms including Precision Software AI, Kewill AI, and MIC Customs Solutions AI at customs broker and importer operations processing HTS binding ruling and duty rate determination document images through AI-assisted tariff classification accuracy and AD/CVD order compliance tools — extracting HTS tariff classification accuracy and AD/CVD scope compliance determinations from CBP entry summary and tariff classification display image inputs in AI-assisted customs entry compliance and duty payment accuracy pipelines.

The adversarial injection surface is the HTS tariff classification binding ruling display image or CBP Form 7501 entry summary classification determination display image submission pathway: Thomson Reuters ONESOURCE AI or Descartes AI customs entry summary and tariff classification display images submitted through AI-assisted HTS classification accuracy and AD/CVD order scope analysis tools for AI customs entry classification compliance record generation and CBP Form 7501 entry summary filing input. An adversarially crafted HTS binding ruling display image or AD/CVD scope determination display image — in which pixel perturbations applied to the 10-digit HTS tariff number and applicable duty rate indicator display region, the AD/CVD order scope and covered merchandise description visual marker, or the antidumping duty deposit rate and exporter-specific margin assignment display in a customs entry tariff classification image cause the AI to classify a customs entry involving AD/CVD order-covered merchandise under an HTS tariff number outside the AD/CVD order scope as if the merchandise were not subject to AD/CVD order deposit requirements when the actual CBP entry merchandise description and HTS classification evidences scope and origin facts requiring AD/CVD antidumping duty rate deposit — can suppress an AD/CVD scope indicator that would otherwise generate an AD/CVD deposit requirement, a CBP cash deposit calculation, and an AD/CVD order compliance record. In importer of record and customs broker operations where AI processes hundreds of customs entry summary classification determinations per day involving potential AD/CVD order-covered merchandise without individual human trade compliance specialist examination of every AI-processed entry classification before the AI determination governs the CBP entry summary filing, adversarial suppression of AD/CVD order scope and applicable duty deposit indicators creates 19 USC §1592 fraudulent customs entry and AD/CVD evasion statute dimensions.

The 19 USC §1592, Enforce and Protect Act (EAPA), 19 USC §1517 AD/CVD evasion, and Commerce Department AD/CVD order enforcement consequences of adversarially suppressed AD/CVD scope classification in customs tariff AI span 19 USC §1592 material false statement civil penalty authority applicable to importers who understate AD/CVD duty deposit obligations through false HTS classification or scope misrepresentation; the Enforce and Protect Act of 2015 (EAPA) 19 USC §1517 authorising CBP to initiate investigations into alleged evasion of AD/CVD orders and impose remedial measures including interim measures and official determinations of evasion; Commerce Department scope ruling and circumvention determination authority under 19 USC §1675a applicable to products alleged to be later-developed merchandise covered within the scope of existing AD/CVD orders; and CBP CTPAT programme suspension authority applicable to importers with patterns of AD/CVD compliance failures. EAPA 19 USC §1517 defines evasion as entering covered merchandise into the customs territory by means of a material and false statement or omission with respect to, inter alia, the applicable HTS classification, the country of origin, or the applicable quantity; adversarial manipulation of customs entry tariff classification AI that suppresses AD/CVD scope indicators creates §1517 EAPA evasion finding dimensions when adversarially corrupted HTS classifications permit AD/CVD order-covered merchandise to enter the US without required antidumping duty deposits. CBP AD/CVD enforcement has resulted in multi-million dollar penalty actions against importers with fraudulent customs entry misclassification patterns; adversarially corrupted customs tariff classification AI that enables systematic AD/CVD order evasion through suppressed scope indicators creates the equivalent exposure of a wilful customs misclassification fraud scheme. Anti-circumvention investigations under 19 USC §1677j examine whether products not explicitly covered by existing AD/CVD orders are assembled or processed in ways designed to circumvent the order; adversarially suppressed HTS classification AI scope determinations that enable AD/CVD circumvention create §1677j anti-circumvention determination and retroactive AD/CVD duty liability exposure. Threshold: 65 for customs tariff classification and AD/CVD entry AI — reflecting 19 USC §1592 material false statement, EAPA 19 USC §1517 AD/CVD evasion, Commerce Department circumvention determination, CBP CTPAT programme suspension, and 18 USC §542 customs entry fraud dimensions.

Integration: customs and trade compliance AI image ingestion with Glyphward pre-scan

Customs and trade compliance AI image ingestion flows from Flexport AI and Thomson Reuters ONESOURCE AI commercial invoice and packing list document scan and OCR display channels, Amber Road AI and SAP GTS AI export control ECCN determination and BIS export licence document scan interfaces, Flexport AI and ONESOURCE AI certificate of origin and FTA rules of origin qualification declaration document processing systems, and Descartes AI and MIC Customs Solutions AI customs entry HTS tariff classification and AD/CVD scope determination display processing platforms into commercial invoice accuracy and customs entry dutiable value classification AI, export control ECCN commodity jurisdiction and licence requirement identification AI, FTA rules of origin qualification eligibility assessment AI, and customs tariff classification and AD/CVD scope compliance determination AI pipelines. Insert Glyphward’s pre-scan at the ingestion boundary before AI-generated output is committed to CBP entry summary filing records, export shipment authorisation determinations, FTA duty preference programme records, or AD/CVD duty deposit compliance records:

import asyncio
import base64
import hashlib
import os
import uuid
from enum import Enum
from pathlib import Path

import httpx

GLYPHWARD_API_KEY = os.environ["GLYPHWARD_API_KEY"]
GLYPHWARD_SCAN_URL = "https://glyphward.com/v1/scan"

# Customs & trade compliance AI — CBP 19 CFR §163 recordkeeping;
# 19 USC §1592 material false statement; 18 USC §542 customs entry fraud;
# EAR 15 CFR Part 730 (ECRA $1M/violation); ITAR 22 CFR Part 120 (AECA $1M/violation);
# OFAC 31 CFR sanctions; USMCA 19 CFR Part 182; GSP 19 USC §2461;
# EAPA 19 USC §1517 AD/CVD evasion.
THRESHOLD_COMMERCIAL_INVOICE_PACKING_AI  = 60  # Flexport/Descartes; 19 USC §1592; §542
THRESHOLD_EXPORT_CONTROL_DOCUMENT_AI     = 55  # ONESOURCE/Amber Road; EAR ECRA; ITAR AECA
THRESHOLD_CERTIFICATE_OF_ORIGIN_FTA_AI   = 60  # Flexport/ONESOURCE; USMCA; GSP §2461
THRESHOLD_CUSTOMS_TARIFF_ADCVD_AI        = 65  # ONESOURCE/Descartes; EAPA §1517; §1592


class CustomsTradeComplianceAIContext(str, Enum):
    COMMERCIAL_INVOICE_PACKING_AI  = "commercial_invoice_packing_ai"  # Flexport, Descartes, CH Robinson
    EXPORT_CONTROL_DOCUMENT_AI     = "export_control_document_ai"     # ONESOURCE, Amber Road, SAP GTS
    CERTIFICATE_OF_ORIGIN_FTA_AI   = "certificate_of_origin_fta_ai"  # Flexport, ONESOURCE, MIC
    CUSTOMS_TARIFF_ADCVD_AI        = "customs_tariff_adcvd_ai"        # ONESOURCE, Descartes, Precision


def threshold_for(context: CustomsTradeComplianceAIContext) -> int:
    mapping = {
        CustomsTradeComplianceAIContext.COMMERCIAL_INVOICE_PACKING_AI:  THRESHOLD_COMMERCIAL_INVOICE_PACKING_AI,
        CustomsTradeComplianceAIContext.EXPORT_CONTROL_DOCUMENT_AI:     THRESHOLD_EXPORT_CONTROL_DOCUMENT_AI,
        CustomsTradeComplianceAIContext.CERTIFICATE_OF_ORIGIN_FTA_AI:   THRESHOLD_CERTIFICATE_OF_ORIGIN_FTA_AI,
        CustomsTradeComplianceAIContext.CUSTOMS_TARIFF_ADCVD_AI:        THRESHOLD_CUSTOMS_TARIFF_ADCVD_AI,
    }
    return mapping[context]


async def scan_customs_trade_compliance_ai_image(
    image_path: str | Path,
    context: CustomsTradeComplianceAIContext,
    trade_entity_hash: str,    # SHA-256 of entry number, AWB, shipment ID, or ECCN determination ref
    compliance_ref: str,       # e.g. "ENTRY-CBP-2026-44821", "EXP-BIS-2026-88441", "USMCA-2026-331"
    clearance_session_id: str, # customs entry session, export shipment batch, or FTA review session
    client: httpx.AsyncClient,
) -> dict:
    """
    Scan a customs or trade compliance AI image for adversarial injection payloads
    before forwarding to commercial invoice accuracy verification, export control ECCN/
    USML licence requirement identification, FTA certificate of origin rules of origin
    qualification assessment, or customs tariff classification and AD/CVD scope determination AI.

    Raises AdversarialCustomsTradeComplianceAIImageError if score meets threshold:
      - COMMERCIAL_INVOICE_PACKING_AI:  threshold 60; 19 USC §1592; 18 USC §542
      - EXPORT_CONTROL_DOCUMENT_AI:     threshold 55; EAR ECRA; ITAR AECA; OFAC sanctions
      - CERTIFICATE_OF_ORIGIN_FTA_AI:   threshold 60; USMCA §182; GSP §2461; §1592 fraud
      - CUSTOMS_TARIFF_ADCVD_AI:        threshold 65; EAPA §1517; §1592 civil; §542 criminal
    """
    image_bytes     = Path(image_path).read_bytes()
    image_b64       = base64.b64encode(image_bytes).decode()
    image_sha256    = hashlib.sha256(image_bytes).hexdigest()
    client_scan_id  = str(uuid.uuid4())
    threshold       = threshold_for(context)

    resp = await client.post(
        GLYPHWARD_SCAN_URL,
        headers={"Authorization": f"Bearer {GLYPHWARD_API_KEY}"},
        json={
            "image": image_b64,
            "source": context.value,
            "metadata": {
                "customs_trade_context":  context.value,
                "trade_entity_hash":      trade_entity_hash,
                "compliance_ref":         compliance_ref,
                "clearance_session_id":   clearance_session_id,
                "client_scan_id":         client_scan_id,
                "image_sha256":           image_sha256,
            },
        },
        timeout=8.0,
    )
    resp.raise_for_status()
    result = resp.json()

    audit_record = {
        "trade_entity_hash":     trade_entity_hash,
        "compliance_ref":        compliance_ref,
        "clearance_session_id":  clearance_session_id,
        "customs_trade_context": context.value,
        "scan_id":               result["scan_id"],
        "client_scan_id":        client_scan_id,
        "image_sha256":          image_sha256,
        "score":                 result["score"],
        "flagged_region":        result.get("flagged_region"),
        "threshold":             threshold,
        "action":                "blocked" if result["score"] >= threshold else "allowed",
    }
    await write_customs_compliance_audit_record(audit_record)

    if result["score"] >= threshold:
        raise AdversarialCustomsTradeComplianceAIImageError(
            f"Customs trade compliance AI image blocked [{context.value}]: "
            f"scan_id={result['scan_id']} score={result['score']} "
            f"entity={trade_entity_hash} ref={compliance_ref}"
        )
    return result


async def write_customs_compliance_audit_record(record: dict) -> None:
    """Persist audit record to customs and trade compliance regulatory documentation store (stub)."""
    import json, sys
    print(json.dumps(record), file=sys.stderr)


class AdversarialCustomsTradeComplianceAIImageError(Exception):
    """Raised when a customs or trade compliance AI image exceeds the adversarial injection threshold."""
    pass

Call scan_customs_trade_compliance_ai_image() with CustomsTradeComplianceAIContext.COMMERCIAL_INVOICE_PACKING_AI before forwarding Flexport AI, Descartes AI, or CH Robinson AI commercial invoice and packing list document scan and OCR display images to HTS tariff classification and customs entry dutiable value accuracy verification AI — with compliance_ref linking the Glyphward scan to the CBP entry number for 19 USC §1592 material false statement civil penalty compliance, 18 USC §542 customs entry fraud prevention, and CBP 19 CFR §163 recordkeeping documentation. Call with CustomsTradeComplianceAIContext.EXPORT_CONTROL_DOCUMENT_AI for Thomson Reuters ONESOURCE AI, Amber Road AI, or SAP GTS AI ECCN determination and BIS export licence document scan images before EAR/ITAR licence requirement identification and OFAC sanctions screening AI, with trade_entity_hash as the SHA-256 of the export shipment reference for EAR ECRA criminal penalty compliance, ITAR AECA criminal penalty compliance, and OFAC SDN sanctions programme audit trail documentation. Call with CustomsTradeComplianceAIContext.CERTIFICATE_OF_ORIGIN_FTA_AI for Flexport AI, ONESOURCE AI, or MIC Customs Solutions AI certificate of origin and USMCA/GSP rules of origin qualification declaration document scan images before FTA preferential duty eligibility classification AI, with clearance_session_id for USMCA 19 CFR Part 182 rules of origin, GSP 19 USC §2461 content requirements, and CBP §1508 FTA supporting documentation audit trail. Call with CustomsTradeComplianceAIContext.CUSTOMS_TARIFF_ADCVD_AI for ONESOURCE AI, Descartes AI, or Precision Software AI customs entry summary HTS tariff classification and AD/CVD scope determination display images before anti-dumping duty scope and deposit calculation AI — with compliance_ref for EAPA 19 USC §1517 AD/CVD evasion prevention, Commerce Department circumvention determination compliance, and 18 USC §542 customs classification fraud audit trail. Get early access

Coverage matrix

Frequently asked questions

Further reading

• Supply chain and logistics AI prompt injection — related adversarial attack surface covering supply chain management AI with procurement, inventory, and distribution logistics dimensions applicable to customs and trade compliance contexts where global supply chain visibility and trade document processing intersect.
• Maritime shipping AI prompt injection — related regulatory framework covering ocean freight and port operations AI with CBP, CTPAT, IMO, and trade compliance dimensions directly applicable to customs and trade contexts where container shipping documentation and port of entry clearance intersect with commercial invoice and certificate of origin processing.
• Financial document AI prompt injection — related adversarial attack surface covering financial document processing AI with SEC, GAAP, and banking regulatory dimensions applicable to customs trade finance contexts where letters of credit, commercial invoices, and cross-border payment documentation intersect with financial document accuracy requirements.
• Free tier — 10 scans/day, no card required — start scanning customs and trade compliance AI documents at development volumes before committing to a production plan.