U.S. border agents are seeking help to take pictures of everyone entering the country by car

U.S. Customs and Border Protection are asking tech companies to send tone to real-time facial recognition tools to take pictures of everyone in the vehicle at a border crossing, including anyone in the back seat, and match them to travel documents, according to documents released last week in the federal register.

The request for information or RIF says the CBP already has a facial recognition tool that takes a person’s photo in an entry port and compares it with travel or identity documents given to border officials and other documents that are already “in government holding”.

“Entries that confirm entry into the United States on biometrics are added to the traveler's cross-record,” the document said.

CBP, an agency under the Department of Homeland Security, said its facial recognition tools “are currently operating in aerial, ocean and land pedestrian environments.” The agency’s goal is to bring it into a “land vehicle environment.” According to a page updated on the CBP website last week, the agency is currently “testing” how to do it. The tests show that while such facial recognition tools have been “improved,” it is not always possible to get photos of every vehicle passenger, especially when they are in the second or third row.

“Human behavior, multi-passenger car rides and environmental obstacles are all challenges facing the vehicle environment,” the document said. The CBP said it hopes private suppliers will provide it with a tool that will “increase passenger images” and “capture 100% of vehicle passengers.”

Dave Maass, director of the Electronic Border Foundation investigation, received a CBP document through a public record request that revealed the agency's 152-day test results from late 2021 to early 2022.

Maas said what stands out for him is the error rate. The cameras at the intersection of Anzalduas border between Mexico and McAllen in Texas took pictures of everyone in the car, only 76%, and of these people, only 81% met the “verification requirement” to match their faces with their identification documents.

The current iteration of the system matches a person’s photos with their travel files, which is one-to-one facial recognition. The main risk here is the inability to recognize a system that someone matches their own files, Maas said. This is unlike one-to-many facial recognition, where police may identify the suspect based on surveillance photos, where the main risk is someone obtaining a false positive match and being mistakenly identified as the suspect.

It is not clear whether the error rate of CBP is mainly related to the camera or the matching system itself, Maas said. “We don't know that racial differences, gender differences, etc. are these systems,” he said.

According to the Intercept, the DHS’s Science and Technology Bureau released a message last August, similar to the information released by the CBP last week. However, the DHS documentation seems to be unavailable at the moment.