Veryfi Launches Real‑Time Fraud Detection Suite as AI‑Generated Document Fraud Surges

Veryfi has announced the general availability of its Fraud Detection & Prevention Suite, an API‑first platform designed to detect fraudulent documents, checks, receipts, and invoices in real time across accounts payable, expense management, and identity verification workflows.

The launch comes amid a sharp rise in document fraud. Nasdaq’s Global Financial Crime Report estimates check fraud losses in the Americas reached nearly $21 billion, representing 80% of global losses. FinCEN has also warned of a surge in AI‑generated documents used to bypass financial controls—an area where traditional rule‑based systems are increasingly ineffective. In expense management, AI‑generated receipts now account for 70% of fraud cases.

In a sample of 200,000 checks processed through Veryfi’s suite, the system flagged 16,664 items for review, representing $7.38 million in protected value. AI‑generated checks made up only 1% of volume but carried the highest per‑incident risk, averaging $1,026 per flagged check. Digitally altered checks accounted for 9,493 flags ($4.48M protected), LCD photo fraud for 2,709 flags ($795K protected), AI‑generated checks for 1,983 flags ($2.03M protected), and duplicate resubmissions for 2,212 flags.

“Fraud has gone from a compliance checkbox to an existential risk for any company processing documents at scale,” said Ernest Semerda, Co‑Founder of Veryfi. He noted that generative AI has made fraudulent documents indistinguishable to the human eye, making real‑time AI‑powered detection essential.

Veryfi’s suite uses a three‑layer detection architecture. Visual fraud detection identifies AI‑generated documents, tampered fields, handwritten edits, and LCD screen captures using purpose‑built vision models. Metadata and document forensics analyze hidden text layers, font mismatches, embedded edits, EXIF data, and C2PA provenance signals. Behavioral and data signals detect duplicate submissions, high‑velocity device behavior, emulated devices, blocked histories, and field‑level anomalies across checks, receipts, and invoices. Each document receives a fraud score with signal‑level attribution, enabling teams to act or configure automated rejection rules.

The suite is fully configurable by workflow. AP teams can enable duplicate detection and PDF forensics; expense platforms can tune handwriting and velocity thresholds; loyalty and CPG programs can add device fingerprinting and AI‑generation detection. Each signal can be independently weighted and tested against live data before deployment.

Fraud detection is delivered through a single API call with results returned in real time, adding under 0.5 seconds to standard processing latency. An asynchronous option is available for latency‑sensitive workflows.