Fake paystub detection now works only as a layered process: inspect the document for internal inconsistencies, reconcile it against independent evidence such as bank deposits and a direct employment check, and add automated document analysis to catch forgeries the eye cannot. Paystub generators and generative AI have made a convincing forgery a matter of minutes, so no single reviewer, however experienced, can be the control. This guide sets out the red flags, the verification steps regulators and agencies already expect, the reporting duties that follow a confirmed case, and how to build a workflow that scales for a credit union, community bank or lender.
TL;DR
- Forged paystubs are cheap and fast to produce: one auto-lending report puts creation at as little as 90 seconds using any of more than 12,000 generator sites.
- Visual inspection is no longer enough. Generative AI can produce fully synthetic financial documents with no original file or trail to compare against.
- The reliable defense is reconciliation: match stub income to bank deposits, prior employment data and a direct employer verification before closing.
- Suspected fraud that meets the threshold triggers a Suspicious Activity Report, generally within 30 calendar days of initial detection.
- Automated document analysis, with audit trails, lets reviewers spend time on the files that deserve it.
Maintain 100% NCUA & OCC Audit Readiness
Monitor regulatory updates 24/7, check internal credit policies, and generate compliance trails with Erina (AI Regulatory Agent).
Table of Contents
- Why paystub fraud has become a lending problem, not an edge case
- How generators and AI changed the threat
- Red flags a reviewer can still catch
- Verify income independently when a stub is in doubt
- Build a workflow instead of relying on a reviewer
- What to do when you suspect fraud
- Our perspective
- Automating paystub and document checks with RiskInMind
- Governance: make the controls examiner-ready
- FAQ
- Sources
Why paystub fraud has become a lending problem, not an edge case
Income is the number most underwriting decisions rest on, and the paystub is still the document most often used to prove it. That makes it the obvious target. The auto lending segment offers the clearest public data: Point Predictive's 2024 report, as covered by BankInfoSecurity, found that income and employment fraud accounted for $3.6 billion in losses in 2023, or 45% of total fraud-related losses for auto lenders. The same coverage reports that 1 in 10 paystubs submitted contains hard-to-detect forgeries. A later report put auto lending fraud losses at $9.2 billion in 2024, up 16.5% over 2023, with employment-data fraud accounting for $3.9 billion of fraud risk.
These figures describe auto lending, but the mechanics carry over to mortgage, consumer, credit card and small-business underwriting wherever a stub is accepted as proof of income. Credit unions and community banks feel it acutely because they often lack the large fraud operations of national lenders and frequently rely on manual review.
The cost of a miss also extends beyond the loan itself. According to the LexisNexis True Cost of Fraud Study 2025, every dollar lost to fraud costs North American financial institutions an average of more than $5 in total impact, up from $4.00 in 2021, and 44% of institutions still rely primarily on manual processes for fraud prevention. Investigation time, charge-offs, repurchase exposure and remediation all sit behind the headline loss.
How generators and AI changed the threat
A decade ago, a forged stub usually meant an edited template with sloppy arithmetic. Today a generator calculates withholding, year-to-date totals and net pay correctly, applies realistic formatting and can invent a plausible employer. The Association of Certified Fraud Examiners' Top Fraud Trends of 2025 notes that generative AI lets fraudsters create pay stubs, bank statements, invoices and tax records that are entirely synthetic. It also cites Sumsub research reporting a 311% increase in synthetic identity document fraud between Q1 2024 and Q1 2025.
The implication is significant. Traditional defenses compare a submitted document against known templates or prior submissions. When the document has no traceable origin, that comparison has nothing to anchor to. U.S. News reporting on mortgage fraud adds that AI-enabled techniques contribute to 19% of U.S. fraud losses and increase the speed and scale at which fraud is attempted.
The practical lesson is that the question has shifted from "does this document look right?" to "does this document agree with everything else we can establish about this borrower?" A flawless-looking stub is not evidence of authenticity. It may be evidence of a better generator.
Red flags a reviewer can still catch
Human review remains useful for the weaker forgeries and, more importantly, for spotting a mismatch between the stub and the rest of the file. Fannie Mae's best practices for income and employment verifications list several indicators worth building into your checklist:
- Rounded earnings figures where real payroll output would produce irregular amounts.
- Generic job titles that do not match the employer or the borrower's described duties.
- Income that is inconsistent with the borrower's occupation or education.
- Bank deposits that do not match the claimed employer, or direct deposits from a different employer at significantly lower income.
- A recent large pay increase.
- Paystub or bank data that reflects a different employment source or income amount than the disclosed information, for example when compared with the Verification of Employment on Form 1005.
These are file-level signals rather than pixel-level ones, which is why they survive the improvement of generators. A forger can perfect the layout of a stub far more easily than the borrower's entire financial footprint.
Check the stub against what a stub must contain
Pay statement content is set by state law, so the baseline varies. Maryland is a useful example of what a complete statement looks like: the state labor department requires the employer's name, address and telephone number, the payment date, pay-period dates, hours worked (unless exempt), rates and other bases of pay, gross and net pay, and the amount and description of each deduction. A stub missing items the borrower's state requires, or showing an employer address and phone number that cannot be matched to a real business, deserves a closer look. Confirm the requirements for the borrower's state before applying such a test, since they differ.

Internal consistency checks
Beyond content, test the arithmetic and the story the numbers tell:
- Does year-to-date gross equal the per-period gross multiplied by the number of periods elapsed, allowing for raises and irregular pay?
- Do withholding amounts look plausible for the stated earnings and the borrower's state?
- Does the pay frequency match the deposit pattern in the bank statements?
- Does net pay on the stub equal the deposit that actually landed in the account?
The fourth check is the most powerful. A genuine paycheck leaves a footprint that a generated stub cannot easily replicate, and cash-flow analysis of the borrower's statements turns this into a repeatable test. We cover the same logic from the bank statement side in what a bank statement taught us about document fraud.
Verify income independently when a stub is in doubt
When reconciliation leaves a question, move off the document and verify at the source. Agency guidance is explicit about how.
Verbal verification of employment. Fannie Mae requires a verbal VOE for borrowers using employment-related income, Schedule K-1 income with less than 25% ownership, or self-employment income to qualify. For employment-related income, employment must be verified no more than 10 business days on or before the note date. The VOE may be obtained after closing, up to delivery, but if it cannot be obtained before delivery the loan is ineligible for sale to Fannie Mae. Fannie Mae frames the purpose as confirming, as late in the process as possible, that the borrower remains employed as disclosed.
Re-verification under HUD guidance. HUD's third-party verification services mortgagee letter states that re-verification of employment must be completed within 10 days before the note date, and that verbal or electronic re-verification is acceptable. Electronic employment data must be current within 30 days of the verification date. Where alternative current employment documentation is used, the mortgagee must obtain the most recent paystub showing year-to-date earnings and original IRS W-2 forms from the previous two years, and document current employment by telephone, recording the name, title and telephone number of the person verified.
A small operational detail matters here: place the verification call to a number you sourced independently, not the number printed on the stub. A forger who controls the stub often controls the phone number on it.
Escalating when the file will not reconcile. If employment and income cannot be reconciled, Fannie Mae's best practices direct lenders to execute Form 4506-T and obtain further information before closing. The same document recommends reverification of employment, income and assets, targeted quality control selections on transactions with red-flag characteristics, and staff training and information sharing.
Build a workflow instead of relying on a reviewer
The controls above are sound, but applied by hand to every file they are slow, and slowness is what pushes institutions to skip steps. A workable design separates routine files from exceptions.
| Stage | What happens | Who or what handles it |
|---|---|---|
| Intake | Document is captured and checked for tampering signals and metadata anomalies | Automated document analysis |
| Reconciliation | Stub income compared with bank deposits, application data and prior verifications | Automated, with analyst review of flags |
| Independent verification | Employer contacted through an independently sourced number, or electronic re-verification run | Processor or underwriter |
| Escalation | Unresolved discrepancies routed to fraud or compliance, with all evidence attached | Fraud or compliance officer |
| Reporting and QC | SAR decision, documentation retention, targeted quality control sampling | Compliance and QC |
The design principle is that automation sorts and documents, while people decide the exceptions. Document-level analysis is also where machine learning earns its place: the same ACFE and industry reporting that shows human reviewers falling behind describes a problem that is, by nature, about detecting subtle statistical patterns at volume. Our engineering write-up on calibration in document fraud detection explains why threshold tuning matters as much as the model, since a detector that flags everything is as unusable as one that flags nothing.
Fraud rarely arrives as a single forged document. A fabricated stub is often paired with a manufactured identity, and the defenses need to talk to each other. Our piece on synthetic identity detection as a layered defense shows how those layers fit together.
Train staff on the process, not just the red flags
Generators improve faster than checklists, so training should emphasize the verification steps that do not depend on spotting a visual flaw: where to source the employer phone number, when to pull a transcript, and how to document the decision. Information sharing between underwriting, QC and fraud teams, which Fannie Mae also recommends, ensures a pattern seen on one file informs the next.
What to do when you suspect fraud
A suspected forged stub is not only a credit decision. For banks and credit unions, it can become a reporting matter. Under 31 CFR 1020.320, a bank must file a Suspicious Activity Report with FinCEN for suspicious transactions that meet the regulatory threshold. The SAR generally must be filed no later than 30 calendar days after the date of initial detection, with limited additional timing rules. Supporting documentation must be maintained, and SARs are confidential: no bank employee, director or officer may disclose a SAR or information that would reveal its existence, except as authorized.
Three practical consequences follow for your process:
- Record the date of initial detection in the file, since the filing clock runs from it.
- Keep every artifact, including the original submission, verification notes and call logs, in an immutable, retrievable location.
- Restrict who in the organization sees the SAR decision, and make sure frontline staff know not to tell an applicant that a report was or will be filed.
Credit unions and other institutions should confirm which SAR rules apply to them with their compliance officer or counsel, since the regulation cited above is the bank rule and other regulators publish their own parallel requirements.
Our perspective
We built RiskInMind around a simple observation: the paystub is not the problem, the isolation of the paystub is. A document reviewed on its own, by one person, under queue pressure, will eventually lose to a generator that costs the fraudster almost nothing.
Our approach is to treat every income document as a claim that must agree with the rest of the file. That means analyzing the document itself, reconciling it against bank statement cash flow, and recording every step in a granular audit trail that an examiner or internal auditor can follow. Our platform is designed to keep that processing in-house, with zero third-party data exposure, SOC 2® aligned controls and real-time workflows that respond in under half a second, so that a stronger control does not become a slower loan.
For credit unions and community banks, the goal is not to replace the underwriter's judgment but to give it better evidence earlier. When a file reaches a human, it should arrive with the discrepancies already surfaced and documented.
Raj
Automating paystub and document checks with RiskInMind
If your team is still checking stubs by eye and phone call alone, the quickest gain is to automate the first pass. Our AI document fraud detection analyzes submitted documents for signs of manipulation, and our bank statement analyzer gives underwriters the cash-flow view needed to test whether stated income actually arrives in the account.

These tools sit inside the same platform as loan application intake, credit memo generation and regulatory reporting, so the evidence gathered during fraud review flows into the credit file instead of living in a separate system. You can see how the plans are organized on our pricing page, or explore the full platform to see where document analysis fits in your underwriting flow. Institutions weighing a rollout may also find our guide to running a 60 to 90 day underwriting AI pilot a practical template for scoping a trial.
Governance: make the controls examiner-ready
A detection tool is only as defensible as the record it leaves. When an examiner or auditor asks why a loan was approved, or why a flagged file was cleared, you need to show what was checked, what the result was, who decided and when. That requires version-controlled decision logic, retained inputs and outputs, and clear ownership for overrides. Our article on SOC 2 audit trails and governance-first AI underwriting describes how to structure that record. If you are adding any AI vendor to the process, run it through your third-party risk program first; our vendor risk assessment guide for banks and credit unions outlines the questions to ask.
Finally, measure the program. Track the share of files flagged, the share of flags confirmed, time added to cycle, and losses avoided or charged off from income misrepresentation. Fannie Mae's recommendation of targeted QC selections on red-flag characteristics gives a ready sampling method: select files that shared a flag, review them after closing, and feed the findings back into your rules.
FAQ
How can you spot a fake paystub?
Look first for file-level mismatches: rounded earnings, generic job titles, income that does not fit the borrower's occupation, and deposits in the bank statements that do not match the stub's employer or net pay. Then check the stub against your state's required content and verify employment through an independently sourced contact. Visual flaws alone are unreliable, because modern generators calculate withholding and totals correctly.
Can ChatGPT or AI generate pay stubs?
Generative AI can produce fully synthetic financial documents, including pay stubs, according to the ACFE's 2025 fraud trends review. Such documents have no original source file to compare against. That is why lenders should verify income through independent channels rather than depend on how a document looks.
How do lenders verify employment when a pay stub looks suspicious?
They complete a verbal or electronic verification of employment close to the note date, using a contact sourced independently of the stub. Fannie Mae calls for verbal VOE within 10 business days of the note date for employment-related income, and HUD guidance requires re-verification within 10 days of the note date. If income cannot be reconciled, Form 4506-T and further documentation come before closing.
Do we have to file a SAR for a forged paystub?
If the activity meets the reporting threshold, a bank must file a SAR with FinCEN, generally within 30 calendar days of initial detection, under 31 CFR 1020.320. Whether a particular case qualifies is a decision for your BSA officer. Keep the supporting documentation and do not disclose the SAR's existence.
Sources
- Best practices for income and employment verifications, Fannie Mae
- Verbal Verification of Employment, Fannie Mae Selling Guide
- Third Party Verification Services, HUD mortgagee letter
- Pay Stub and Pay Statement Required Information, Maryland Department of Labor
- Top Fraud Trends of 2025, ACFE
- AI Is Making Mortgage Fraud Easier and Easier to Catch, U.S. News
- Fake Paystubs Are Draining Billions from Lenders, BankInfoSecurity
- Auto Lending Fraud Climbs 16.5% to $9.2B in 2024, BankInfoSecurity
- Every Dollar Lost to Fraud Costs North America's Financial Institutions, LexisNexis Risk Solutions
- 31 CFR 1020.320, FFIEC BSA/AML Manual
