Customs and Border Protection (CBP) has overcome initial problems it had with a biometric-related feature of a mobile service application that the Biden administration has begun asking asylum seekers to use, but the facial comparison technology used on the backend of the CBP One mobile app and other biometric processing done by the agency is showing high rates of accuracy regardless of what countries people are coming from, a CBP official said on Wednesday.

The facial recognition match rates of persons seeking entry into the U.S. or people departing the country are 99.6 percent for Middle Eastern countries, 99.5 percent for African countries and 98.9 percent for North American countries, Diane Sabatino, deputy executive assistant commissioner for CBP’s Office of Field Operations, told a House panel.

The facial recognition algorithm that CBP uses in its Traveler Verification System is “high performing,” she told the House Oversight and Accountability Subcommittee on National Security, the Border, and Foreign Affairs. She added that “we have some very significant statistics and high match rates with respect to countries of citizenship.”

CBP uses a facial matching algorithm provided by NEC Corp.

Sabatino was responding to concerns raised by Rep. Alexandria Ocasio-Cortez (D-N.Y.), who mentioned that facial recognition technology “has been of extraordinary concern to us in the committee” and highlighted previous testimonies and reports of “baked in biases withing facial recognition algorithms in certain technologies.”

Ocasio-Cortez also highlighted a recent decision by the Department of Homeland Security to begin using the CBP One app as a new path for foreign nationals seeking asylum in the U.S. She said early reports indicate that the app is presenting facial recognition difficulties for migrants from Africa and Haiti.

Sabatino replied that it wasn’t the facial recognition piece of the CBP One processing that was an issue but rather problems with the liveness detection algorithm used to sort out whether the photo taken by the mobile app is of an actual person rather than say a picture of a person. The liveness detection feature is meant to prevent, in part, bad actors scheduling appointments with CBP for other people.

CBP has changed its procedures for using the CBP One app and that has fixed the liveness detection issues, Sabatino said. The app is being used to streamline the entry application process to reduce the administrative burden on CBP officers.

Rep. Maxwell Frost (D-Fla.) later in the hearing picked up on Ocasio-Cortez’s comments, saying “we are still receiving” information from non-governmental organizations that the CBP One app is presenting difficulties for people with darker skin to “be processed in a timely matter.”

Sabatino said that changes CBP has made in recent weeks to how the app is used have “significantly cut down on those data errors,” which aren’t related to facial matching technology. Since the enhancements have been made to the app, CBP has seen an increase of family units and groups traveling to the border.