FinTechCryptoTask Prompt

AI Prompt for Cross-Border KYB and UBO Discovery

This prompt instructs Arc, the deepidv research agent, to crawl corporate registries in multiple jurisdictions, map direct and indirect shareholders above your ownership threshold, and identify ultimate beneficial owners under FATF guidance. Every UBO and shareholder is then screened against OFAC SDN, EU Consolidated, UN, and HMT sanctions lists plus PEP databases and adverse media. It is built for KYB analysts, fintech onboarding teams, and crypto compliance officers who need a registry-sourced ownership picture before approving a corporate customer.

How to use this prompt

1
Open the Arc agent in your deepidv dashboard and paste the full prompt. It also works in Claude, ChatGPT, or Gemini if you paste in registry extracts yourself, since those tools cannot crawl registries directly.
2
Replace the placeholders: subject entity name and registration number, primary jurisdiction of incorporation, secondary jurisdictions to traverse, and the ownership threshold that triggers stakeholder inclusion (default 25%).
3
Run it and review the structured output: a corporate structure map with effective ownership percentages, a UBO list with HIGH/MEDIUM/LOW confidence scores, screening results per person, and jurisdictional risk flags such as shell-company indicators.
4
Verify every screening hit and ownership claim against the cited registry source (Companies House, EDGAR, ACRA, RCS) before acting. The prompt explicitly tells the model to flag missing or contradictory registry data instead of inventing chains.
5
Apply the recommended KYB risk tier (Standard, Enhanced, or Very High) and re-verification cadence to the customer record, and request the listed documentation gaps from the subject entity.

The prompt

Arc, perform a deep UBO discovery for "[Insert Company Name]". Crawl the corporate registries in [Country A] and [Country B], and map all stakeholders with >25% ownership. Cross-reference these stakeholders against global PEP lists.

INPUT REQUIRED:
- Subject entity name (and registration number if known)
- Primary jurisdiction of incorporation
- Secondary jurisdictions to traverse (subsidiaries, shareholding entities, beneficial owners)
- Risk threshold: ownership percentage that triggers a stakeholder inclusion (default 25%)

OUTPUT, return the following structured response:

1. CORPORATE STRUCTURE MAP
- Direct shareholders (entities and natural persons) with ownership %
- Indirect shareholders through holding companies, with calculated effective ownership
- Trust or nominee arrangements where detected
- Cross-jurisdictional ownership flows (e.g., BVI holdco owns Cyprus subsidiary owns operating company)

2. UBO IDENTIFICATION
- Natural persons with >25% effective ownership (UBOs under FATF guidance)
- Persons with control rights through means other than equity (board appointments, voting agreements)
- Confidence score for each identified UBO (HIGH / MEDIUM / LOW) with the source evidence

3. SCREENING RESULTS
- For each identified UBO and direct shareholder:
  * OFAC SDN check
  * EU Consolidated check
  * UN sanctions check
  * HMT check
  * Country-specific PEP database hits
  * Adverse media surfacings (with source URLs and dates)

4. JURISDICTIONAL RISK FLAGS
- Shell-company indicators (incorporation date <12 months, no operating address, multiple intermediaries)
- High-risk jurisdiction touchpoints (FATF greylist, EU AML list, OFAC sanctions list)
- Beneficial ownership transparency score for each registry crawled

5. RECOMMENDATIONS
- KYB risk tier classification (Standard / Enhanced / Very High)
- Re-verification cadence based on the risk tier
- Open questions where registry data was insufficient
- Documentation gap: what additional disclosures the subject should provide

Cite the registry source for each ownership claim (Companies House, EDGAR, RCS, Singapore ACRA, etc.). Where data is insufficient or contradictory across registries, flag the discrepancy explicitly. Do not fabricate ownership chains; if the registry lookup is blocked or returns empty, say so.

Test it in Claude or another LLM

This prompt is built for the Arc agent inside deepidv, where it performs deep cross-border UBO discovery, crawls corporate registries across two jurisdictions, maps every stakeholder above 25% ownership, and screens them against global PEP and sanctions lists. Here is how to dry-run the same workflow in any LLM first, using a fake corporate structure, before you let Arc crawl real registries against live data.

1
Paste the full prompt into Claude, ChatGPT, or Gemini and replace the opening "Arc," with a role instruction such as "Act as a KYB analyst performing UBO discovery from registry data." Fill the bracketed fields with a fake subject like [Insert Company Name] = ACME-TEST-HOLDINGS LTD and set the two jurisdictions, for example [Country A] = United Kingdom and [Country B] = Cyprus.
2
Paste the synthetic sample data block below so the model has a fabricated ownership chain and registry stubs to reason over, rather than attempting any real registry lookup.
3
Good output for this prompt is a corporate structure map showing direct and indirect shareholders with calculated effective ownership, a UBO list flagging natural persons above 25%, per-UBO screening results across OFAC/EU/UN/HMT/PEP/adverse-media, jurisdictional risk flags such as shell-company indicators, and a KYB risk-tier recommendation. The model should explicitly flag any branch where the (fake) registry data is insufficient rather than inventing an ownership chain.
4
Refine the prompt until the effective-ownership math through the holding layer is shown step by step and the insufficient-data flags appear instead of fabricated chains.
5
Once the output shape is right, run it live in the deepidv dashboard where Arc executes it against real corporate registries (Companies House, EDGAR, ACRA, RCS, and others) and live PEP and sanctions feeds, citing the registry source for each ownership claim.

Synthetic sample data to paste alongside the prompt

Fake test data, safe to share with any LLM. Swap in your own once the output looks right.

KYB SUBJECT (TEST, all fabricated):
Subject: ACME-TEST-HOLDINGS LTD, reg no. TEST-UK-00000000, incorporated UK 2026-03-01 (under 12 months old).
Direct shareholder: DEMO-BVI-HOLDCO (BVI) owns 60% of subject; registry stub FAKE-BVI-REG returns no operating address.
Direct shareholder: Jane Testcase (natural person, jurisdiction Testistan) owns 40% of subject.
DEMO-BVI-HOLDCO is owned 100% by John Sample (natural person) per nominee arrangement note.
Screening seed: "John Sample" returns a fabricated PEP-DB-TEST hit; "Jane Testcase" returns no hits.

Pairs with on deepidv

Sources & further reading

FAQ

What counts as an ultimate beneficial owner in KYB checks?

Under FATF guidance, an ultimate beneficial owner is a natural person who ultimately owns or controls a legal entity, with 25% effective ownership being the most common threshold in national implementations. Control can also arise without equity, for example through board appointment rights or voting agreements, which is why this prompt asks the model to surface non-equity control as well. Many regulators expect firms to calculate effective ownership through intermediate holding companies, not just direct shareholdings.

Can an AI agent really map beneficial ownership across multiple corporate registries?

Yes, when the agent has live research access, as Arc does in the deepidv dashboard, it can pull filings from registries like Companies House, EDGAR, and Singapore ACRA and assemble cross-jurisdictional ownership chains with cited sources. The output is a starting analysis, not a legal determination: registries vary widely in beneficial ownership transparency, so the prompt forces the model to attach confidence scores and flag gaps where registry data is blocked, empty, or contradictory.

Run it with live verification data

These prompts work in any LLM. Inside the deepidv dashboard, Luna, Arbiter, and Arc run them against your real sessions, screening lists, and audit trails.

Book a Demo