Global fraud rate: the headline number
Across every document type submitted to the Turing Verify platform in the first quarter of 2026 (diplomas, transcripts, government IDs, professional certificates, award letters), about 11% were flagged as forgeries or unverifiable. The rate was 7% in 2024. Most of the jump sits in academic credentials.
That number hides huge variance. A US passport submitted by a Fortune 500 employee runs a fraud rate close to zero. A master's degree submitted by a remote candidate applying for a senior role from a high-risk source country can run above 30%. The 11% is an average across that whole spectrum. Useful as a benchmark. Dangerous as a planning number.
Top document types faked
Not every document is a forgery target in equal measure. Forgers concentrate where the payoff is highest and the verification is weakest. The result: the distribution below. Academic credentials dominate.
Top source countries
Source-country distribution is the hardest statistic to publish responsibly. The same country may be over-represented because of legitimate immigration patterns, not because its residents are more likely to forge documents. We publish ranked source countries internally for risk modeling and report only directional findings publicly.
The directional finding for 2026: forgeries originate disproportionately from regions where the official credential registry is weak, or where the gap between in-country wages and cross-border opportunity is widest. Both correlations are economic, not cultural. Both are dropping as registries improve and AI verification closes the in-country gap that used to shield fraud.
Fraud technique breakdown
The single biggest shift between 2024 and 2026 is the technique mix. Two years ago, the common forgery was a Photoshop edit of a real diploma: a name swap, a date change, a signature paste. Today, AI-generated forgeries are the plurality of fakes in the academic category.
- AI-generated (≈42%): end-to-end synthetic documents produced by generative image and layout models. Visually flawless; fail forensic checks at the seal, signature, and registry layers.
- Photoshop edits (≈26%): real templates with fields modified. Detected through edit-boundary artifacts, font inconsistency, and metadata mismatch.
- Diploma-mill issuance (≈19%): documents genuinely issued by an entity, but the entity is unaccredited or non-existent. Detected through registry cross-check.
- Stolen-identity submissions (≈8%): a real document submitted by someone other than the credential holder. Detected through liveness checks and identity cross-verification.
- Other (≈5%): printed-and-scanned reproductions, manually altered originals, and edge cases.
In two years, AI-generated forgeries went from a curiosity to the plurality of fakes in academic credentials.
Cost to victims
The fully loaded cost of a single credential-driven bad hire, counting onboarding, salary, downstream errors, project delays, and replacement cost, ranges from $240,000 to $850,000 depending on role seniority and industry. For specialized roles (engineering leads, medical staff, finance officers), the upper bound climbs much higher once you factor in regulatory exposure.
At the institutional level, university admissions fraud has a different cost profile: lost tuition revenue when the fraud is caught late, accreditation risk if patterns emerge, and opportunity cost from displacing a legitimate applicant. Insurance KYC fraud, covered in our KYC Fraud Patterns in Insurance: 2026 Field Report, has a cost profile closer to direct claim payouts.
Year-over-year trend
The trend is clear. Comparing the first quarter of 2026 with the same period in 2024:
- Overall fraud rate: about +57% (from ~7% to ~11% of submissions).
- AI-generated forgeries: roughly +400% in absolute volume, and now the plurality of academic fakes.
- Diploma-mill issuance: about +90% in volume, driven by the productization of mill kits on dark-web marketplaces.
- Photoshop edits: down as a share of the total, roughly flat in absolute volume.
- Detection rate: up sharply on AI-generated forgeries after forensic engines added registry depth and signature biomechanics models.
Frequently asked questions
Where do these numbers come from?
From the Turing Verify production platform. Every document submitted is forensically evaluated, and the aggregated, anonymized results feed the Wall of Forgeries dashboard. We publish methodology notes alongside each metric so external analysts can reconstruct our definitions.
Are these numbers representative of the whole world?
They are representative of the document mix submitted to our platform, which skews toward HR, admissions, and KYC use cases in North America, Europe, and parts of Asia. Government and law-enforcement document mixes are not represented and would shift the distribution.
What counts as a confirmed forgery in your data?
A document that fails the registry cross-check, or fails three or more independent forensic signals at high confidence. We do not count borderline cases as forgeries. Those get reported separately as “suspicious: pending human review.”
Can I cite these statistics in my own report?
Yes. Please cite as “Turing Verify Wall of Forgeries, Q1 2026” and link to this page. For raw data access or custom slices, enterprise customers can request export through their account team.
How often are these numbers updated?
The dashboard is updated continuously; this page is refreshed quarterly with rolling 12-month figures. Methodology changes are versioned and noted in the dashboard's changelog.