Where does AI-generated document fraud show up?
AI-generated document fraud shows up wherever a business accepts a file as evidence: expense claims, insurance claims, onboarding, lending and supplier payments. Image and text generators make a convincing fake cheaper to produce than before.
- Expense receipts generated as images, with plausible merchants, totals and creases.
- Insurance claim photos of damage that never happened, or real damage made worse.
- Invoices, bank statements and payslips submitted as PDFs for lending or onboarding.
- ID photos or selfies created or altered with AI to open accounts.
- Video onboarding sessions where the face on screen may be synthetic.
- Voice messages or calls from a “supplier” or “executive” announcing changed bank details.
In November 2024, the US Financial Crimes Enforcement Network (FinCEN) issued an alert describing fraud schemes that use deepfake media to get around financial institutions’ identity verification, and listed red flags for reporting.
Is AnyFact a KYC or ID verification tool?
No. AnyFact answers one question: does this file show signs of AI generation? It is a useful extra signal for fraud and KYC deepfake detection, but it does not replace an identity verification provider.
- It does not check ID documents against government or issuer registries.
- It does not run liveness checks or match a selfie to an ID photo.
- It does not validate security features such as holograms or machine-readable zones.
- It is built to spot AI generation, not every manual edit: a total changed by hand in a PDF editor is a different forensic problem.
What can go wrong in document fraud AI checks?
Missing a fake means paying a false claim or opening an account for a fraudster. Flagging a genuine customer is costly too: a denied claim, a frozen account or a complaint. Honest customers often submit compressed phone photos, filtered images and re-scanned documents, which are exactly the files detectors find hardest.
Treat a detector flag the way FinCEN treats a single red flag: a reason to look at the surrounding facts, not a conclusion. Read how accurate AI detectors are before setting review thresholds.
How to detect AI-generated documents and images for fraud: a review workflow
The workflow below checks the data, then the file, then escalates to a person. Most fakes fail at step one.
- 01Check the data. Does the merchant exist at that address? Do line items add up, and does the tax math work? Does the policy cover that item on that date? Is the invoice number format right for that supplier?
- 02Request originals. Ask for the original photo from the phone or the PDF as issued, not a screenshot or a re-saved copy.
- 03Read provenance. A content credentials checker reports C2PA and EXIF/XMP data, which may name an AI generator.
- 04Run the file through the right detector: the AI image detector for receipts and claim photos, the AI document detector for PDFs.
- 05Look for duplicates. Reverse image search claim photos; compare receipts against earlier submissions from the same person.
- 06Escalate flagged cases to a trained reviewer who can contact the customer, inspect the item or ask the merchant.
- 07Record the decision and the evidence, and file regulatory reports where your obligations require them.
AI Document Detector · Free plan
Upload an invoice or statement and get a page-by-page reading of its text, embedded images and metadata.
Fake receipt detection: red flags to check by hand
Generated receipts and documents often fail on details a model does not understand. Train reviewers to check these before running any tool:
- Merchant name, address, phone and tax number that do not match the real business, or do not exist.
- Line items, tax and totals that do not add up, or prices that are wrong for that merchant.
- Lettering that is slightly warped, inconsistent fonts within one line, or characters that turn into shapes when you zoom in.
- Paper texture, creases and shadows that look too even, or a receipt photographed on a background with no real lighting direction.
- File metadata showing an image generator or editor, or a creation date after the claimed purchase.
- Several claims from different people with near-identical layouts or backgrounds.
For claim photos, the guide on how to tell if an image is AI-generated lists the visual anomalies that hold up best.
Onboarding video and voice: KYC deepfake detection signals
For recorded onboarding sessions, watch for edges that blur around the face when the person turns, lighting on the face that does not match the room, and lip movements that drift from the audio. The guide on how to spot a deepfake video explains each one. Calls that announce changed bank details deserve the same suspicion; see our page on voice clone scams.
Which AnyFact detectors fit fraud teams, and what do they return?
Four detectors cover the evidence fraud teams handle. Each returns a verdict (probably human, uncertain or probably AI), the signals behind it, a plain-language explanation, the method’s limits and a PDF report you can attach to a case file.
| Evidence | Detector | What it returns |
|---|---|---|
| Receipt, claim photo, ID selfie | Photo | Provenance findings, a classifier verdict and visible anomalies such as lettering and textures |
| Invoice, statement, payslip (PDF, DOCX) | Document | A page-by-page result: text blocks, up to 30 embedded images, metadata, OCR on scanned pages |
| Onboarding or claim video | Video | Suspicious moments on a timeline and a face-swap check when a frontal face is clear |
| Call or voice message | Voice | A score per segment, a transcript and flags for urgency, secrecy, money requests and changed bank details |
AnyFact works through its web app; there is no public API yet. That suits investigators reviewing escalated cases rather than screening every transaction.
AI Photo Detector · Free plan
Check a receipt or claim photo for AI-generation signals and get a PDF report for the case file.
Limits to keep in mind
Detection is weakest on heavily compressed images, small crops and scanned copies of printed fakes. A fraudster who prints a generated receipt and photographs it removes much of the digital trace. That is why data checks and human review stay at the center of the process.
Frequently asked questions
Can AnyFact verify that an ID document is genuine?
No. AnyFact is not a KYC or ID verification provider. It does not check documents against registries, validate security features, run liveness checks or match faces. It only looks for signs that an image or file was generated by AI.
Can you detect an AI-generated ID photo?
Often there are signals, such as provenance data naming a generator or visual anomalies, and AnyFact’s photo detector looks for them. Results are indications, and they are weaker on compressed or cropped images. Use them alongside your identity verification provider, not instead of it.
How do I spot insurance claim photo fraud?
Ask for the original photos, check their metadata and dates, reverse image search them, and compare them with the policy details and any inspection. An AI image check adds a signal for photos that may be generated.
Can AnyFact check scanned documents?
Yes. The document detector reads scanned pages with OCR (the first 20 scanned pages) and also checks embedded images and metadata, for PDF, DOCX or PPTX files up to 20 MB and 50 pages.
Is there an API for automated screening?
Not yet. AnyFact is available through its web app on every plan, including the free one, which fits manual review of escalated cases.