As generative AI makes forged financial documents harder to spot, Veryfi has made its Fraud Detection & Prevention Suite generally available. The configurable, API-first platform identifies fraudulent cheques, receipts, invoices, and other documents in real time across accounts payable, expense management, and identity verification workflows. Each document is assessed at the point of collection, before it enters downstream processes.
Cheque fraud and AI-generated documents drive demand
According to Nasdaq's Global Financial Crime Report, cheque fraud losses across the Americas reached nearly USD 21 billion, representing 80% of global cheque fraud losses. In the US, FinCEN has warned of a rise in AI-generated documents used to bypass financial controls, a threat that rule-based systems struggle to detect. Veryfi also reports that AI-generated receipts now account for 70% of fraud cases in expense management.
In a real-world sample of 200,000 cheques, the suite flagged 16,664 for review, representing USD 7.38 million in value. Digitally altered cheques formed the largest group, with 9,493 flagged and USD 4.48 million in value, while photos of screens accounted for 2,709 cheques worth USD 795,000, and 2,212 duplicates were identified. AI-generated cheques made up about 1% of volume, but the 1,983 flagged were worth USD 2.03 million, an average of USD 1,026 each.
Three detection layers configured by workflow
The suite analyses documents across three layers at the same time. Visual detection identifies AI-generated documents, tampered fields, handwritten changes, and photos of screens. Document forensics checks for hidden text layers, font mismatches, embedded edits, and EXIF and C2PA provenance data. Behavioural signals cover duplicate submissions, high-velocity or emulated devices, blocked device histories, and data anomalies such as amount mismatches. Each document receives a fraud score showing which signals were triggered, and compliance and operations teams can review these scores or set automated rejection rules.
Signals can be enabled, weighted, and tested against live data before deployment. Accounts payable teams can activate duplicate detection and PDF forensics, while expense platforms can adjust handwriting and velocity thresholds. Loyalty and consumer packaged goods programmes can add device fingerprinting and AI-generation detection. Results are returned through a single API call, adding less than 0.5 seconds to processing time, with an asynchronous option for latency-sensitive workflows. Ernest Semerda, Co-Founder of Veryfi, said detection needs to run at the point of collection, before any funds move.
