Yes, where ownership can be traced on paper. An AI agent can pull registry records, read shareholder registers and trust deeds, trace ownership through each layer and calculate indirect holdings against your threshold. Where control runs through something other than shares, or the chain ends in a jurisdiction with no public register, a person decides who the beneficial owner is.
Why is beneficial ownership hard to establish?
An ultimate beneficial owner is the natural person who ultimately owns or controls a customer. The FATF Recommendations expect firms, as part of customer due diligence, to identify that person and take reasonable measures to verify who they are.
On simple structures this takes minutes: one company, two shareholders, both individuals. The difficulty is in structures with several layers, trusts, nominee shareholders, or holdings split just below the threshold across family members. Each layer needs a document, each document needs reading, and percentages need multiplying through the chain. That is careful, tedious work where errors compound, and it is where an agent earns its place.
What evidence does the agent assemble?
For each customer the agent builds an ownership file:
- The chain: every entity between the customer and the natural persons at the top, with the percentage held at each step.
- Sources for each link: registry extracts, share registers, trust deeds, partnership agreements or the customer's own structure chart, with each source labelled by how reliable it is.
- Calculated holdings: each individual's indirect percentage, worked through the chain and shown step by step.
- Control indicators: voting agreements, rights to appoint directors, powers under trust deeds, and signatories with unusual authority.
- Discrepancies: where the customer's chart disagrees with a public register, or registers disagree with each other.
- Dead ends: layers where no document could be found.
Source reliability matters a great deal here. A registry extract is an official record from an identifiable source. A structure chart the customer drew is a declaration. Where the file rests on declarations alone, the person reviewing it should be able to see that at a glance. The beneficial ownership verification checklist sets out what evidence each type of structure needs.
What decision does a person keep?
| Situation | Who decides |
|---|---|
| Clear chain, all links documented, owners above threshold | Agent records; person reviews by sample or under policy |
| Holdings split just under the threshold among relatives | Person decides whether they act together |
| Control through agreements, not shares | Person decides who controls |
| Trust structures | Person decides which settlors, trustees and beneficiaries to treat as owners under policy |
| No natural person found above threshold | Person decides who holds senior managing control and records why |
| Registry and customer disagree | Person decides how to resolve and whether to report the discrepancy |
The threshold is a floor for identification, not a definition of control. Someone holding a small stake but with the right to appoint the whole board controls the company, and only a person reading the agreements in context should make that call.
What does a traced structure look like?
Take a corporate bank onboarding a logistics company. The customer's chart shows two holding companies above it, each owned by a different family member.
The agent pulls the registry records. They confirm the first holding company is owned by one person. The second is registered abroad, and its register shows a nominee shareholder services firm. The agent cannot see past the nominee. It also finds, in the logistics company's articles, a clause giving the second holding company the right to appoint the chair, who has a casting vote.
The file shows one beneficial owner confirmed, one unresolved, and a control right that makes the unresolved one more important. The onboarding officer asks for the nominee declaration of trust. It names a third family member, not shown on the customer's chart. The officer records both individuals as beneficial owners, notes the control right, and raises the customer's risk rating. The step-by-step method is in how to identify ultimate beneficial owners with an AI agent.
What controls make the result reliable?
- Calculations are shown. Every indirect percentage is worked visibly, so a reviewer can check the arithmetic.
- Each link cites its document. A link with no document is shown as unverified.
- Dead ends are recorded. The file says where the trail stopped, not just who was found.
- Thresholds are set by policy. The agent applies your threshold and flags holdings just below it.
- Accountability for the outcome is named. Supervisors who follow principles like MAS FEAT accountability expect a named owner for each agent's work; the person who signs the ownership finding answers for it.
The general principle is in our guide to AI compliance agents.
Where Pryme fits
In Pryme, each link in an ownership chain reaches the agent stamped with its provenance: a registry record arrives as signed, a structure chart from the customer as declared. The file shows which owners rest on declarations alone. Recording an owner where control is unclear is outside the agent's role scope and is held for a named officer.
Questions people ask
Can the agent re-check ownership when a registry changes?
Yes. It can watch registry filings for existing customers and flag changes in ownership or control for review.
Does the agent verify the identity of the owners too?
It can run identity verification on the individuals it finds, using the same checks as for any customer. Verification and identification are separate steps.
What if the customer refuses to provide ownership documents?
The agent records the refusal. Whether to proceed, decline or exit is a decision for the compliance officer.