2026 Research Briefing

Global Document Fraud Report 2026

AI, screenshots, and the new fraud supply chain — how document fraud became industrialized, and what teams should change before approving the next applicant, loan, vendor, or claim.

TrueDoc Research June 2026 15 min read

PDF · ~1 MB

Key findings

Document fraud went industrial in 2025–2026

$12.5B
U.S. consumer fraud losses in 2024 (+25% YoY)
FTC Consumer Sentinel Network Data Book 2024
$16.6B
Record losses to FBI IC3 in 2024 (+33% YoY)
FBI IC3 2024 Internet Crime Report
50%
Forged documents were the #1 fraud type in 2024
Sumsub Identity Fraud Report
$40B
Projected U.S. GenAI fraud losses by 2027 (from $12.3B in 2023)
Deloitte Center for Financial Services

Sources: FTC Consumer Sentinel Network Data Book 2024 (released March 2025); FBI IC3 2024 Internet Crime Report; Sumsub Identity Fraud Report; Deloitte Center for Financial Services. Full citations in the "Sources & methodology" section at the end of this report.

The new fraud supply chain

From $30 template to full application package

Old fraud changed one number in a PDF. Modern fraud assembles a whole story — ID, pay stub, bank statement, address, employer letter — across multiple tools and formats. The danger isn't a single fake. It's the package.

  1. 1
    Template / generator
    Editable pay stub, invoice, statement template
  2. 2
    AI edit
    Numbers, names, balances polished with AI tools
  3. 3
    Metadata stripping
    Authoring tool, timestamps, layers removed
  4. 4
    Format hopping
    PDF → screenshot → photo to erase evidence
  5. 5
    Application package
    ID + pay stub + bank statement + address bundled
Most targeted documents

The documents fraudsters fake most

Two views on the same problem, both from the Sumsub Identity Fraud Report: the share of each identity-document type inside all forged identity documents, and the share of each fraud category inside all fraud attempts in 2024.

Forged identity documents — share by type (2025)

Source: Sumsub Identity Fraud Report 2025–2026. ID cards were ~70% of forged ID documents in 2024 and 72% in 2025.

  • ID cards72%
  • Passports13%
  • Driver's licenses10%

Percentages are share of fraudulent identity documents, not share of all fraud. Remaining ~5% is split across visas, residence permits, and other IDs.

2024 fraud-type split — share of all attempts

Source: Sumsub Identity Fraud Report. Forged or altered documents were the #1 fraud type in 2024.

  • Forged documents50%
  • Chargebacks15%
  • Account takeovers12%
  • Deepfakes7%
  • Fraudulent networks4%

Bars are scaled 2× the underlying share for visual readability — numeric values shown are the real Sumsub percentages. Remaining ~12% is distributed across smaller categories.

Global identity-fraud rate moved from 1.1% of verifications in 2021 to 2.6% in 2024, and eased to 2.2% in 2025 (Sumsub) — still roughly double the early-decade baseline.

Format hopping

Screenshots aren't fraud — they erase evidence

When a native PDF becomes a screenshot, scan, or photo, structural and metadata signals disappear: PDF objects, embedded fonts, creation timestamps, layer history, and source-app indicators. Some users genuinely don't know how to download the original — but from a risk perspective, screenshots of payroll, banking, tax, and address documents should be treated as elevated risk.

Simple rule: if a bank statement, pay stub, tax form, or utility bill normally exists as a downloadable PDF and the applicant submits a screenshot, the review process should ask why.
Original PDF
Fonts · objects · metadata
Screenshot
Image only
Lost signals
Structure · metadata · layers
  • Embedded fonts and PDF objects removed
  • Creation & modification timestamps gone
  • Source application indicators erased
  • Manipulation traces flattened into pixels
Metadata trap

Metadata is a signal, not a verdict

A document may show traces of Adobe, Preview, Photoshop, online PDF tools, scanning apps, mobile capture, or payroll systems. Some of those are legitimate. Some hint at manipulation. The tool alone does not prove fraud — and a clean metadata profile does not prove authenticity. Professional fraudsters strip or normalize metadata before submission.

Useful as a signal
  • Inconsistent authoring tools across an application
  • Edit traces that contradict the document's claimed origin
  • Software fingerprints known for templated fraud
Misleading on its own
  • "Clean" metadata after intentional stripping
  • Legit tools (Acrobat) used to commit fraud
  • Mobile scans of authentic paper documents
AI-generated document fraud

Manual fake vs AI-assisted fake

The bigger risk isn't one person editing one PDF. It's automation — generating dozens of variations, scoring them with another model, packaging them with stolen identity data, and submitting the best version across many platforms. This is document fraud at software speed.

Manual fake

2015–2022
  • One document, one editor (Photoshop / PDF tool)
  • Visible artifacts, font mismatches
  • Inconsistent math and YTD totals
  • Limited scale — one applicant at a time
  • Often caught by an attentive human reviewer

AI-assisted fake

2024–2026
  • Generated and refined at scale across many variants
  • Clean layout, plausible fonts, realistic math
  • Bundled with synthetic IDs and addresses
  • Format-hopped (PDF → screenshot) to erase evidence
  • Submitted across many platforms in parallel
AI surge — by the numbers

The growth curve is no longer theoretical

The shift from "AI could fake documents" to "AI is faking documents at scale" happened inside a single 12-month window. Each figure below is published by a named source — Sumsub, Deloitte — not modeled by us.

+1,100%
Deepfake fraud growth, Q1 2024 → Q1 2025
Source: Sumsub
+300%
Synthetic identity-document fraud Q1 2024 → Q1 2025 (+311% in North America)
Source: Sumsub
Deepfakes' share of all fraud rose 4× vs. 2023 — now 7% of attempts
Source: Sumsub
1 in 50
Forged documents that are AI-generated (ChatGPT, Gemini, Grok)
Source: Sumsub 2025–2026
+700%
Deepfake incidents in fintech in 2023
Source: Deloitte
67%
Firms reporting an increase in fraud in 2024
Source: Sumsub
Anatomy of a modern attack

The Arup deepfake: $25M wired after a video call full of fakes

In January 2024, an employee in the Hong Kong office of global engineering firm Arup transferred roughly $25 million (HK$200 million) in 15 transactions after joining a video call he believed included the company's UK-based CFO and several colleagues. Investigators confirmed that every other "executive" on the call was an AI-generated deepfake, built from publicly available footage. The employee acted only after the live call appeared to confirm a written request he had initially flagged as suspicious.

Stage 1 — pretext
Phishing email requesting a confidential transfer.
Stage 2 — synthetic trust
Live deepfake video call impersonating multiple executives.
Stage 3 — execution
15 wires totalling ~$25M before the fraud was discovered.

Source: widely reported (Hong Kong Police; CNN, Reuters, FT — February 2024). The Arup case is the largest publicly confirmed single-incident deepfake loss to date and is consistent with Regula's finding that deepfake incidents in 2024 averaged ~$500,000 in losses, with financial-sector companies losing ~$603,000 per affected company.

TrueDoc detection stack

Layered evidence beats a single red flag

TrueDoc combines structural, mathematical, visual, and AI-detection signals into one explainable report — so reviewers see why a document was flagged, not just a black-box score.

L01

OCR extraction

Read names, dates, employers, balances, totals, and document identifiers.

L02

Math consistency

Net vs gross pay, YTD totals, transaction balances, invoice math.

L03

File structure

Inspect PDF objects, fonts, layers, and signs of reconstruction.

L04

Metadata

Authoring app, timestamps, and edit traces — as one signal, not a verdict.

L05

Visual forensics

Spacing, alignment, font swaps, text replacement, and image artifacts.

L06

AI-generation signals

Indicators of generative authoring, paired with other evidence.

L07

Explainable report

Reviewer-readable evidence, not a black-box score.

Industry risk

Where document fraud hurts most

The higher the volume and the faster the decision, the more attractive the workflow becomes to fraudsters.

2026 checklist

What companies should change in 2026

  • 1
    Prefer original PDFs — request the native file when one should exist.
  • 2
    Treat screenshots and scans as elevated risk for income, identity, banking, tax, and address.
  • 3
    Run math checks on pay stubs, statements, invoices, and tax forms.
  • 4
    Combine metadata with structural, visual, and contextual signals.
  • 5
    Cross-document consistency: name, address, employer, income story.
  • 6
    Define clear risk policies before fraud happens — approve / review / request / decline.
  • 7
    Retain evidence and explainable reports for every decision.
FAQ

Document fraud in 2026 — common questions

What is document fraud?

+

Document fraud is the creation, editing, manipulation, or misuse of a document to misrepresent identity, income, address, employment, finances, business status, or eligibility.

What documents are most commonly faked?

+

Pay stubs, bank statements, IDs, tax forms, proof-of-address documents, invoices, receipts, employer letters, and business documents.

Can AI create fake documents?

+

Yes. AI tools can generate or edit realistic-looking documents, especially when combined with online templates, PDF editors, and stolen identity data.

Are screenshots safe for verification?

+

Screenshots are not always fraudulent, but they are weaker evidence than original PDFs because file structure and metadata signals are lost in conversion.

How does TrueDoc detect fake documents?

+

TrueDoc uses layered analysis across OCR, math consistency, file structure, metadata, visual forensics, AI-generation indicators, and explainable risk reporting.

Should companies automatically reject suspicious documents?

+

Not always. The right action depends on risk policy. Some documents should be declined, some should be reviewed, and some should trigger a request for the original file.

Key statistics — cite this report

Headline numbers, with sources, ready to quote

For journalists, analysts, and bloggers: the figures below are the most-cited statistics from this report, each attributed to its original public source. When referencing, please link back to this page — truedoc.io/blog/global-document-fraud-report-2026.

  1. 1

    $12.5B in U.S. consumer fraud losses in 2024, up 25% year-over-year.

    Source: FTC Consumer Sentinel Network Data Book 2024 (released March 2025)

  2. 2

    $16.6B record losses reported to the FBI IC3 in 2024, up 33% YoY; cyber-enabled fraud accounted for 83% of losses ($13.7B).

    Source: FBI IC3 2024 Internet Crime Report

  3. 3

    $2.77B in business email compromise (BEC) losses in 2024.

    Source: FBI IC3 2024

  4. 4

    Fraud-related Suspicious Activity Reports rose 110% — from 552,920 (2020) to 1,165,642 (2024).

    Source: FinCEN, via U.S. Federal Reserve Consumer Compliance Outlook (2025)

  5. 5

    Forged or altered documents were the #1 fraud type in 2024, accounting for 50% of all fraud attempts.

    Source: Sumsub Identity Fraud Report

  6. 6

    Among fraudulent identity documents: ID cards ~70% (2024; 72% in 2025), passports 13%, driver's licenses 10%.

    Source: Sumsub Identity Fraud Report 2025–2026

  7. 7

    2024 fraud-type split: forged documents 50%, chargebacks 15%, account takeovers 12%, deepfakes 7%, fraudulent networks 4%.

    Source: Sumsub

  8. 8

    Global identity-fraud rate: 1.1% of verifications (2021) → 2.6% (2024) → 2.2% (2025).

    Source: Sumsub

  9. 9

    67% of firms reported an increase in fraud in 2024.

    Source: Sumsub

  10. 10

    Deepfakes were 7% of all fraud attempts in 2024 — a 4× increase from 2023.

    Source: Sumsub

  11. 11

    Q1 2024 → Q1 2025: deepfake fraud +1,100%; synthetic identity-document fraud +300% (+311% in North America).

    Source: Sumsub

  12. 12

    Roughly 1 in 50 forged documents is now AI-generated, produced with tools like ChatGPT, Gemini, and Grok.

    Source: Sumsub 2025–2026

  13. 13

    Deepfake incidents in fintech rose 700% in 2023.

    Source: Deloitte

  14. 14

    U.S. generative-AI-enabled fraud losses are projected to reach $40B by 2027, up from $12.3B in 2023 — a 32% CAGR (conservative scenario ~$22B).

    Source: Deloitte Center for Financial Services, 'Deepfake Banking Fraud Risk on the Rise'

  15. 15

    Average loss per fraud event in 2024 was ~$300,000; deepfake-related incidents averaged ~$500,000; financial-sector companies lost ~$603,000 per affected company.

    Source: Regula, Deepfake Trends 2024 (and Sumsub)

  16. 16

    January 2024: an Arup employee in Hong Kong wired $25M (HK$200M) after a video call in which every 'executive' on the screen was a deepfake.

    Source: Widely reported

Sources & methodology

Built on the latest public data from named authorities

This report is a synthesis of the most recent publicly available statistics on consumer fraud, identity fraud, document forgery, deepfakes, and AI-assisted attacks, drawn from U.S. federal agencies, leading identity-verification platforms, and major consulting and research firms. Every figure cited inline is attributed to a named source below — no number in this report is modeled, extrapolated, or produced by TrueDoc. We treat sourcing as a feature, not a footnote: every claim should be independently verifiable.

  • FTC Consumer Sentinel Network Data Book 2024
    Released March 2025.
  • FBI Internet Crime Complaint Center (IC3) — 2024 Internet Crime Report
    Annual federal report on cyber-enabled crime losses.
  • FinCEN — Suspicious Activity Report (SAR) data
    Referenced via the U.S. Federal Reserve Consumer Compliance Outlook, 2025.
  • Sumsub Identity Fraud Report — 2024 and 2025–2026 editions
    Global identity-verification provider; aggregated platform-wide fraud signals.
  • Deloitte Center for Financial Services — 'Deepfake Banking Fraud Risk on the Rise'
    Forward projection of GenAI-enabled fraud losses.
  • Regula — Deepfake Trends 2024
    Survey of average per-incident loss across industries.

Note on time-frames: 2024 figures are full-year totals released in 2025 (FTC, FBI IC3, Sumsub). 2025 figures, where shown, are interim updates from the Sumsub 2025–2026 edition. The Deloitte $40B 2027 projection is the firm's central scenario; their conservative scenario is ~$22B. All losses are expressed in U.S. dollars unless otherwise noted.

See the evidence before you decide

Upload a document to TrueDoc and get layered evidence — math, structure, metadata, visual forensics, and AI-generation signals — in a reviewer-ready report.

PDF · ~1 MB · free to cite

Methodology: Every statistic in this report is published by a named public authority — FTC, FBI IC3, FinCEN, Sumsub, Deloitte, or Regula — and attributed inline. TrueDoc did not model, estimate, or extrapolate any of these figures. See the "Sources & methodology" section above for full citations. Last updated June 2026.