Document fraud in the US: AI-made IDs, synthetic identities and IRS tax transcripts

US regulators now treat AI-generated identity documents as a mainstream threat. In November 2024 FinCEN warned that suspicious activity reports describing fraudulent identity documents, many made with generative AI, had been rising since 20231. In February 2026 the creator of OnlyFake, a site that produced at least 10,000 digital fake IDs, pleaded guilty in New York2.

By the TamperCheck research teamLast updated 4 cited sources

Check with the issuer first

Social Security number match through the Social Security Administration, with the holder's consent

Where there is no official check

Payslips have no official check in the United States. Teams rely on Prior-year IRS transcripts through IVES, bank deposits. TamperCheck reads those files and marks the region that was edited.

10,000+

digital fake IDs produced by the OnlyFake site, including passports for 56 countries, according to prosecutors

US Attorney's Office, SDNY, Feb 2026

$182m

of possible fraud in about 3,000 suspicious activity reports on potentially synthetic identities filed in 2021 alone

FinCEN data, via FedScoop (GAO review), 2021 data

From edited licences to generated ones

FinCEN's alert FIN-2024-Alert004 describes criminals using generative AI to alter or create images of driver's licences and passports, then using them to open accounts that receive and launder the proceeds of other fraud1. Its suggested red flags are practical: a photo that a reverse image search matches to galleries of AI-generated faces, inconsistencies between identity documents, and a customer who cannot complete a live verification check1.

OnlyFake turned this into a product. Prosecutors say the site produced at least 10,000 digital fake IDs, including passports for 56 countries, in exchange for about $1.2 million in cryptocurrency2. Its output was an image of a document lying on a surface, which is exactly what many remote onboarding flows ask customers to upload. A system that only checks whether an image looks like a licence has no defence against an image made to look like one.

Synthetic identities need consistent paperwork

A synthetic identity combines a real Social Security number, often a child's or an unused one, with an invented name and date of birth. FinCEN recorded about 3,000 suspicious activity reports on potentially synthetic identities in 2021, involving about $182 million of possible fraud, and the GAO notes that this is likely an undercount because many losses are booked as credit losses4.

Each synthetic identity needs a supporting set of documents: a licence, a utility bill, pay stubs and a W-2 that all agree. That set is where it can be caught. Each document may be well made on its own, but the issuers behind them have no record of the person.

Income is verified against the IRS, not the PDF

US mortgage lending already has a check that bypasses the income document altogether. Through the IRS Income Verification Express Service, a borrower authorises the lender, using Form 4506-C or their IRS online account, to receive tax transcripts directly from the IRS3. A pay stub or W-2 that disagrees with the transcript has been edited, or the income was never reported.

The limit is timing and coverage. Transcripts reflect what was filed, which may be a year old, and they do not cover the recent pay periods a lender also wants to see. That gap is filled by pay stubs and bank statements, which have no issuer-side check and are where forensic review earns its place.

What this means for your underwriting and onboarding team

Use IVES and eCBSV where the file and the borrower's consent allow. For the uploads that no issuer will confirm, such as licences, pay stubs and bank statements, TamperCheck looks for edited and generated images across 300+ checks and names the field that was changed, so a flagged file comes with the reason it was flagged.

How to verify US documents at the source

  • With the taxpayer's consent on Form 4506-C or through their IRS online account, a participating lender receives transcripts directly from the IRS and compares them with the documents supplied3.

    What it does not prove: Covers filed years only. Current-year pay stubs still need checking against bank deposits.

  • Social Security number

    SSA electronic Consent Based SSN Verification (eCBSV)

    Approved financial institutions, with the holder's consent, check whether a name, date of birth and SSN match SSA records. The answer is yes, no, or a death indicator4.

    What it does not prove: A match confirms the combination exists. It is not identity proofing, and SSA says it should not be used to authenticate the person4.

  • Employers compare I-9 information with Social Security and Department of Homeland Security records after hiring.

    What it does not prove: It confirms the record, not the document; a genuine identity attached to someone else's photograph can still pass.

  • Registered government agencies check an applicant's immigration status against DHS records using details from their immigration documents.

    What it does not prove: Available to registered agencies, not to private businesses.

United States-specific documents

Each has its own page covering what a genuine copy contains and how forgeries of it are made.

Universal documents

These document types are used worldwide and TamperCheck runs country-aware forensic checks for United States submissions.

Document fraud in the United States: recent cases

  1. Feb 2026

    OnlyFake creator pleads guilty in New York

    Prosecutors say the AI fake-ID site produced at least 10,000 digital fake IDs, including passports for 56 countries, and took about $1.2 million in cryptocurrency.

    US Attorney's Office, SDNY
  2. Nov 2024

    FinCEN issues alert on deepfake media targeting financial institutions

    The alert describes rising reports of fraudulent identity documents made with generative AI and lists red flags for onboarding teams.

    FinCEN

Regulators and issuing authorities

Frequently asked questions

Can AI-generated driver's licences pass online ID checks?

They are designed to. FinCEN reported a rise in accounts opened with AI-altered or AI-generated licences and passports1, and OnlyFake sold images of IDs photographed on a surface for exactly this purpose2. Image-level forensic analysis, live verification and cross-checks between documents are what catch them.

How do US lenders verify income documents?

For prior years, by pulling IRS tax transcripts through IVES with the borrower's consent on Form 4506-C and comparing them with the W-2s and returns supplied3. For current income, pay stubs are checked against payroll deposits on bank statements, and the employer is contacted to confirm employment.

What is the difference between a fake identity and a synthetic identity?

A fake identity impersonates a real person. A synthetic identity is a new one built around a real SSN with an invented name and date of birth, so there is no victim to notice and complain. SSA's eCBSV lets permitted institutions check whether a name, date of birth and SSN belong together, which is aimed at exactly this4.

What red flags does FinCEN list for deepfake identity documents?

Among them: a photo that reverse image search links to galleries of AI-generated faces, inconsistencies between a customer's identity documents, and a customer who avoids or cannot complete live verification1. FinCEN asks institutions to reference the alert in suspicious activity reports.

Sources

  1. 1.FinCEN Issues Alert on Fraud Schemes Involving Deepfake Media Targeting Financial Institutions (FIN-2024-Alert004). Financial Crimes Enforcement Network, 13 Nov 2024.
  2. 2.Creator of OnlyFake charged and pleads guilty to selling more than 10,000 digital fake IDs. US Attorney's Office, Southern District of New York, Feb 2026.
  3. 3.Income Verification Express Service. Internal Revenue Service, accessed Oct 2026.
  4. 4.Social Security's synthetic identity fraud service costs more than projected, GAO finds. FedScoop, accessed Oct 2026.

Figures are quoted as each source published them, with the source's own date. Spot an error or a newer figure? Tell our support team.

Check a document

Upload a PDF or a photo from the United States. TamperCheck returns a risk score from 0 to 100 and marks the region that was edited, usually within seconds.

For developers: see the API and MCP documentation.

All country hubs