Every document in an application is a set of claims about a business. ForgeCheck checks that each claim holds up: in the file, across the whole application, and against what the business says it is. Every flag comes with its evidence.
No sign-up for samples. Your own documents: sign in with a work email.
ForgeCheck is a document fraud detection tool from Coris. It reads every document in a merchant application as claims (who the business is, where it banks, what it holds, what it sells, how it gets paid) and checks each one three ways: does the file hold up, do the documents agree, and does it fit the business? Each check returns a verdict (Clear, Needs review, Not original or Flagged), a score, and the evidence and next step for every finding.
Re-saves after a file was made, and the editors and code libraries that touched it.
Original figures left in the file, under boxes drawn over them.
Signs a document, or part of one, was generated or edited with AI.
Balances, totals and average tickets checked against their parts.
Scan lines and barcodes that disagree with the numbers printed beside them.
Names, owners, addresses and bank accounts compared across every file in the packet.
Where the bank operates, against where the business is registered and licensed.
Average ticket, volume and receivables, against what the business sells and how it gets paid.
AI tools now produce statements whose pixels and totals look right. What they rarely get right is everything else the application says: the account on the check, the address on the processing statement, a bank that operates where the business does.
That’s why ForgeCheck checks the claims, not just the pixels.
Four ForgeCheck results on fictional sample documents, and one real fraud case, anonymized.
The claim: an ending balance of $13,117.03, and every visible total reconciles. The file says otherwise. It was re-saved three times, its build shows PyMuPDF wrote it, and 455 characters of hidden text sit under boxes drawn over the original figures. About $10,200 of withdrawals vanished from the page, not from the file.
See all 7 findings →The claim: $78,232.11 in volume. The file records that figure being added after its first save, with the original still under a box. And the statement’s own average ticket doesn’t equal volume divided by count: printed 31,093.31, the parts give 49,492.86.
See all 5 findings →A photo has no file history to lean on, so ForgeCheck reads the claims printed on the check. Here the barcode or scan line holds a different value from the printed one, though both should come from the same record.
See all 3 findings →Each file can pass on its own. Read together, the voided check and the bank statement claim different business names, accounts ending 2955 and 8171, and addresses in different states. One of them likely belongs to someone else.
See the combined result →The claim: the clinic’s business account is at a small community bank in rural Ohio. The check itself was clean. But the applicant was a chiropractic clinic registered and licensed in Oregon, and the bank has no branches in Oregon. Small clinics bank where they practice. This one didn’t, and the application was fraud.
From a real fraud case a customer shared. Names and identifying details changed.
of 330 DOJ pandemic-loan fraud cases reviewed by the GAO falsified tax, payroll or bank documents. 67% involved a fake or shell business.
in 29,435 pandemic loans tied to identity-theft referrals had the bank account changed before funding.
lost to business email compromise in 2025, often a swapped account on a letter, voided check or W-9. About $123k per complaint.
Every finding also says when it might be innocent. Photos of paper, merged PDFs, a business with two locations: honest merchants produce these too, and ForgeCheck says so instead of calling them fraud.
Check statements and voided checks at onboarding, before the first payout.
Catch doctored processing statements before an application goes to the processor.
Spot-check the documents behind your programs’ approvals, with evidence you can audit.
Underwrite on statements you can trust, without asking reviewers to look harder.
It checks whether a document submitted as evidence, like a bank statement or voided check, was edited, generated or borrowed from another business. ForgeCheck does it by checking the claims each document makes.
Every document says something about a business: its name, bank account, balances, average ticket. ForgeCheck checks whether the file supports each claim, whether the other documents agree, and whether it fits the business.
Yes, when the file carries the evidence: re-saves in an editor, hidden text, or boxes drawn over the original figures, as in our sample bank statement.
It checks every document for signs of AI-made or AI-edited content. And an AI-made statement still has to agree with the rest of the application.
Run the four samples at forgecheck.app.coris.ai with no sign-up. To check your own PDF, JPG or PNG files, sign in with a work email.
Follow us
.svg)
.png)
%201.png)
.svg)