How to Spot a Fake Shareholder Register

What a Fake Shareholder Register Actually Is

A shareholder register is the company's own record of who holds its shares, how many, and of what class. UK law calls it the register of members. Section 113 of the Companies Act 2006 requires every company to keep one, with each member's name and address, the date they became a member, the date they ceased to be one, and, for a company with share capital, the shares held by each member, distinguished by number and class, with the amount paid. Delaware calls the same record the stock ledger. DGCL § 219(c) defines it as the records in which the names of all stockholders of record, the number of shares registered to each, and all issuances and transfers are recorded, and makes it the only evidence of who may vote at a stockholder meeting.

A cap table is a different object. It is a model of ownership, usually a spreadsheet or a platform export, and it often mixes issued shares with options, warrants, SAFEs, and convertible notes on a fully diluted basis. In a KYB packet, all three arrive under the same label: proof of who owns the applicant and in what percentage.

A fake one makes that ownership claim without being the company's actual record. It comes in two production families. Edited-after-creation starts from a genuine register, a real platform export, or counsel's ledger, then changes a name, a share count, or a percentage. Generated-from-scratch builds the table to a target, so every row sums correctly because the totals were written first. Neither is the company's books.

The motive is almost always the threshold. Under 31 CFR 1010.230, a covered institution must identify each individual who owns 25 percent or more of the equity interests of a legal entity customer, plus one individual with significant responsibility to control or manage it. An owner shown at 24 percent disappears from the first prong. Open Ownership's analysis of thresholds notes that any threshold can be exploited by keeping stakes just below the line, and the Russian Laundromat example shows the line moving and the stakes moving with it. UBO identification is the judgment the register is supposed to support, not a replacement for it.

Tells That Still Work, and the Ones That Don't

The useful tells on a shareholder register are arithmetic and cross-document, not visual. A register is a table of numbers, and numbers have to agree with other numbers the applicant does not control as easily.

The basis switch deserves its own sentence. The CDD rule counts equity interests, and a cap table that reports a founder at 22 percent fully diluted may show the same founder above 25 percent of issued shares once unexercised options and unconverted notes come out of the denominator. That is not always fraud. It is always a reason to recompute.

The tells that fail are the ones that look official. A company seal, a "certified true copy" stamp, a director's signature, and a cap-table platform's logo all survive an edit, because the edit happens to the rows and leaves the furniture alone. A screenshot of a cap-table dashboard is the weakest form of all: it has no export history and can be altered in a browser before capture.

A related-party cluster is the most common honest-looking fake. Spouses, adult children, a trust, and a holding company each sit at 15 to 24 percent, and no single row crosses the line. The FATF guidance on beneficial ownership of legal persons says thresholds should not exceed 25 percent, that countries may consider combined ownership interests such as shareholders working together, and that individuals who exercise ultimate control should be identified regardless of whether their shares exceed any threshold. A clean register can still hide a controller.

How Detection Actually Works

Detection on a shareholder register reads the file's story, then reconciles its numbers against documents the applicant produced at different times for different reasons.

Production method asks how the file was made. A native export from a cap-table platform, a law firm's ledger printed to PDF, a spreadsheet saved as PDF from an office suite, and a page assembled in a design tool leave different structural traces. A register that claims to be a platform export but carries an office-suite producer string was handled after it left the platform.

Timestamp trail asks whether that history fits the claimed date. Registers change with every transfer, so a recent save is normal. A file saved minutes before upload, with the percentage column as the last edit, is a different story from a ledger that accumulated entries over years.

Issuer matching ties the register to the entity. Legal name, registration number, jurisdiction, and share classes should match the certificate of incorporation or articles, and authorized shares should support the issued total. For an LLC, the member schedule in the operating agreement is the comparable record, and the percentages there have to agree with the register the applicant calls a cap table.

Consistency is the core check for this document type. Holder names, share counts, and certificate numbers on the register should match any stock certificates in the packet. Percentages should match the beneficial ownership certification. The owner on the register at 24 percent should not be the same person who signed the application as CEO with no other control person named. A generated register tends to be internally perfect and externally wrong, so inconsistency across documents fails and internal consistency alone passes nothing.

Model artifacts are what a generated table leaves: uniform row spacing no export produces, rounded percentages that do not divide from the share counts, and column labels that match no known platform or statute. Recycled patterns close serial reuse: one register template, one set of holder addresses, or one percentage split across applicants that claim no common owner. Those six checks are classes of evidence, not a recipe for a cleaner fake.

What to Do with a Suspect File

Stop treating the upload as the register. Recompute every percentage from share counts on an issued-equity basis, then aggregate holders who share an address, a surname, an entity parent, or a signatory. If any person or related group reaches 25 percent, identify them as a beneficial owner and treat the certification as unreliable. 31 CFR 1010.230 lets an institution rely on customer-supplied ownership information only when it has no knowledge of facts that reasonably call that information into question. An arithmetic gap is such a fact.

Ask for the record from its keeper, not from the applicant's inbox. A native export sent by the cap-table administrator, the company secretary, or outside counsel is stronger than a PDF forwarded by the person being verified. Ask for transfer history covering any holder whose stake changed in the past year. A transfer that dropped someone from 30 to 24 percent just before onboarding is a question for the customer, and the answer goes in the file.

Apply the control prong regardless of what the percentages say. Identify the individual with significant responsibility to control, manage, or direct the entity, and check whether that person also appears on the register under a different name or through an entity. UBO mapping across layered holdings is where a 24 percent direct stake turns into a 40 percent indirect one.

For a scored second look at the file itself, check a document in the Watchdoc playground. The rest of the packet, from formation filings to bank letters, is covered in KYB document fraud detection.

Where Sphinx Fits

Sphinx scores a shareholder register the same way it scores every other onboarding document. Watchdoc runs the six checks, shows the x-ray of how the file was made and edited, and returns a verdict a reviewer can defend. Across 1 million documents processed, Watchdoc reaches a 94.3% correct verdict, catches 2.8x more forgeries, clears clean files in under 28 seconds, and is 54% less expensive, at $0.45 per document with no seats and no platform fee. The Watchdoc playground is free for the first file, with no email required, on a 50-document free tier.

Frequently Asked Questions