Landlords, lenders, employers and online platforms ask for documents: payslips, bank statements, utility bills, ID cards. For years, checking them meant looking for obvious edits. Generative AI ended that. Convincing fake statements and ID images can be produced quickly, and services selling them operate openly online. Fraud prevention firms have reported steep growth in AI-assisted document fraud.
Here is how verification is changing, from both sides of the desk.
What AI forgery is good at
- Visually perfect documents: correct fonts, logos, layout and realistic numbers.
- Consistent sets: a payslip, a bank statement and a reference letter that agree with each other.
- Variations at scale: different names and amounts for many fake applicants.
- Photographs of ID documents with realistic lighting and wear.
What it still struggles with
- Data that must match the outside world. A bank statement can look perfect, but the transactions will not exist in the real bank's systems.
- File-level traces. PDFs generated or edited by certain tools leave patterns in metadata, fonts and object structure. See hidden data in PDFs explained.
- Liveness. A live selfie video with movement is harder to fake than a static photo, though deepfake video is improving too.
- Cryptographic signatures. A digitally signed statement from a bank cannot be altered without breaking the signature.
How verification is shifting
From documents to data sources. Instead of uploading a bank statement, applicants increasingly connect their bank account through open banking, so the lender reads transactions directly. Payroll and tax data connections do the same for income.
From photos to chips and wallets. Reading the NFC chip in a passport, or using digital identity wallets such as the EU Digital Identity Wallet, gives cryptographically signed identity data. See the EU Digital Identity Wallet.
From one check to layered checks: document analysis, metadata forensics, liveness, device signals and data consistency together.
From visual review to signature verification where documents are digitally signed. See how to verify a digital signature in a PDF.
For businesses that still accept PDFs
- Prefer documents you can verify at the source: ask for a digitally signed statement, or confirm by contacting the issuer.
- Check metadata and producer fields as one signal, not a verdict. See how to detect tampered PDFs.
- Cross-check figures: does the income on the payslip match the deposits on the statement?
- Use specialist verification services for high-risk decisions.
For applicants
Expect more requests to connect a bank account or scan a passport chip, and fewer requests to email PDFs. When you do send documents:
- Send originals as downloaded from the issuer. Editing a genuine statement, even just to hide unrelated transactions, can make it look tampered. If you need to hide information, ask whether redaction is acceptable and use proper redaction. See how to redact text in a PDF.
- Use secure channels, not plain email, for identity documents. See how to share a PDF securely.
Takeaway
AI made forged documents look genuine, so verification is moving from "does it look right?" to "can the source confirm it?". Expect data connections, chips and signatures to replace screenshots and PDFs for high-stakes checks. For related fraud, see AI-generated fake receipts.
