Microblink aces IDNet identity document verification evaluation

Put to the test against the IDNet dataset, Microblink’s identity verification platform detected 100 percent of fake documents and had zero false acceptances in end-to-end analysis across documents and territories.

In a release, President and COO of Microblink Hartley Thompson says that “as generative AI makes it easier to create convincing fake identities, independent testing has never been more important.” There is currently no standardized metric for deepfake detection, meaning there is no norm against which to check self-reported metrics.

On the other hand, Thompson says, “our results on the IDNet dataset show just how effective the Microblink Platform is at catching advanced synthetic fraud. We saw zero false acceptances in our comprehensive testing. This is a clear sign our AI is delivering real protection for businesses and consumers.”

The DHS S&T sponsored the IDNet dataset in response to a rise in identity fraud spurred by generative AI tools. A release calls it “the first large-scale, independently created dataset specifically designed for benchmarking document verification solutions against synthetic identities,” intended to provide an objective standard for evaluating vendor performance. It includes over 837,000 images from 10 U.S. states and 10 European countries, covering biometric fraud techniques such as face morphing and portrait swaps, as well as text field alterations.

The Brooklyn, New York-based computer vision and digital identity verification company evaluated its identity document verification capability for 10 U.S. states: Arizona, California, Nevada, North Carolina, Pennsylvania, South Dakota, Utah, Washington D.C., West Virginia, and Wisconsin, using 83,652 synthetically generated images within the IDNet dataset.

Microblink Platform launched in March 2025. But, says Thompson, “we’ve been using our in-house Fraud Lab for years to generate synthetic IDs and simulate real-world fraud. With over 280,000 AI-generated documents, we’re constantly training and testing our models to stay ahead of both digital and physical tampering.”

“The release of the IDNet dataset gave us the perfect opportunity to validate our technology against a rigorous, third-party benchmark. The results speak for themselves.”

Article Topics

DHS S&T  |  document verification  |  generative AI  |  identity document  |  IDNet dataset  |  Microblink  |  synthetic identity fraud