Trade Compliance AI
AI trade compliance software, and what makes Veredis's approach different
“AI trade compliance” covers a wide range of actual rigor — from AI-assisted summarisation to fully deterministic, auditable decisioning. Here's what the term means in practice, and where Veredis sits.
The problem
Trade compliance decisions — is this counterparty sanctioned, is this good export-controlled, does this shipment need an export licence — used to depend on a human compliance officer manually cross-referencing sanctions lists, export-control regimes, and denied-party registers against a stack of trade documents. That process is slow, inconsistent between reviewers, and produces very little that a regulator or auditor can later reconstruct: why was this transaction cleared, on what evidence, as of what date?
Generic AI tools promise to speed this up, but many do so by asking a large language model to read a document and render a compliance judgment in free text. That's fast, but it inherits exactly the reproducibility problem it was meant to solve — a model asked the same question twice, or a different model entirely, can produce a different answer, with no rule path to point to when someone asks why.
Veredis's approach: document-first, deterministic decisioning
Veredis starts from the actual trade document — an invoice, purchase order, bill of lading, or certificate — and extracts the structured facts a compliance decision actually depends on: parties, goods descriptions, HS codes, jurisdictions, values, and routing. That extraction is where AI genuinely helps: reading messy real-world documents and pulling out the fields that matter.
What happens next is deliberately notleft to a model's judgment. Those structured facts are evaluated against versioned, official-source intelligence — sanctions lists, entity and ownership data, export-control classifications — by deterministic rules. The same inputs produce the same decision, every time, and the rule path that produced it is recorded alongside the decision itself.
The evidence model
Every Veredis decision is recorded in a structured, cryptographically signed dossier: which source versions were consulted, at what timestamp, which entity identifiers were resolved, and the rule path that produced the outcome. That dossier — not a narrative summary — is what a compliance team or auditor can point to later.
- Source versions and timestamps for every check performed
- Entity identifiers resolved against official registries, not free-text name matching alone
- An explicit decision state — clear, needs review, blocked, insufficient data, or source unavailable — rather than a single score
- Hash references and a signed dossier so the record can be verified as unaltered later
Where the limitations are
A deterministic system is only as good as the sources and rules it's built on. Veredis is explicit about when a source could not be checked (source unavailable) rather than presenting an unavailable check as a clean pass — but that also means a decision can legitimately come back needing manual review when official data isn't reachable, not because the transaction itself is risky.
Veredis does not recommend, suggest, or facilitate circumvention of any sanctions, export control, or trade restriction. Any routing or structuring analysis it surfaces is scoped to lawful, non-sanctioned trade flows only, and its output is decision support — not a substitute for qualified legal judgment on genuinely novel or high-stakes fact patterns.
Governance
Compliance-critical logic in Veredis is deterministic and versioned by design — see Intelligence Sourcing & Data Boundary for exactly which signals are queried live versus served from governed, versioned intelligence, and Security for the platform's data-handling and retention posture.
Frequently asked questions
What is AI trade compliance software?+
AI trade compliance software applies machine learning and rules-based automation to the work of assessing trade transactions and documents against sanctions, export-control, and customs regulations. The category ranges widely in rigor — from AI-assisted checklist tools that summarise documents, to deterministic systems like Veredis, where the compliance-critical verdict itself is produced by versioned, auditable rules rather than free-form model judgment.
How is Veredis different from an AI checklist or template tool?+
Veredis's compliance-critical decisions — sanctions matches, export-control flags, controlled-goods classification — are evaluated by deterministic rules against structured, official-source data. AI is used for document extraction and narrative explanation, not as the sole authority for the verdict itself. A checklist tool that asks an LLM 'does this look compliant?' and reports its free-text answer is a fundamentally different architecture, with no reproducible rule path to audit if a decision is later questioned.
Does Veredis replace legal or compliance counsel?+
No. Veredis produces decision support — a structured, evidenced assessment — not legal advice. Qualified sanctions or export-control counsel should be engaged for any transaction assessment before it is relied upon, particularly for close calls or novel fact patterns Veredis's rules were not designed around.
What happens if Veredis can't find data for an entity or transaction?+
Veredis distinguishes between a confirmed clean result and a source that was simply unavailable — it does not silently treat missing data as a pass. Decisions carry explicit states (clear, needs review, blocked, insufficient data, source unavailable) so a reviewer can see exactly why a decision landed where it did, rather than a single opaque score.
Is my document data retained after a scan?+
No, not by default. Veredis extracts the minimum structured fields required for the compliance analysis and does not retain raw uploaded documents, raw prompts, or full transaction narratives as a matter of course — see the Security page for the full data-handling posture.