CBP still working toward anomaly detection algorithms at the border

Customs and Border Protection is continuing to work on developing algorithmic systems to conduct automated inspections of vehicles at the border, with the providers of these systems still in the testing phase and an agency goal of transitioning them to “operational readiness.”

A privacy threshold analysis for automated detection algorithms from 2022 — but only made public on the Department of Homeland Security’s website earlier this month — reveals some details about the government’s aspirations for the technology. Through a pilot, originally planned to occur between 2023 and 2024, the agency hoped to develop an “automated threat recognition” algorithm, with the help of DHS’s AI Center of Innovation and its office of the chief information officer. As part of the testing, the agency said it would make around one million images of commercial trucks and cars available for companies to develop their systems. 

Eventually, the technology is supposed to help the agency identify contraband and threats through the Non-Instrusive Inspection program, which uses screening technologies like X-ray machines and other detectors deployed at the border. 

One anticipated algorithm described in the document aimed to evaluate whether trucks or trailers were empty, while other use cases could be assigned to the firms developing the technology as the pilot progressed. The technology was also supposed to be stored in a sandbox, a separate secure environment for testing new systems, according to the analysis. 

It’s not the first time DHS has expressed interest in automated detection algorithms. In April 2023, the agency released a solicitation and a statement of work for the project and then later hosted an industry day focused on that project. At least one company, Pangiam — which was acquired last year by a firm called BigBear.ai — has won a contract to develop the technology for CBP. 

“NII systems such as vehicle X-ray, radiation portal monitors (RPM), chemical detectors, etc., are tools that provide a more efficient and effective means to screen conveyances and individuals for contraband and identify threats, compared to manual search techniques,” a solicitation document from 2023 stated. “The NII Systems Program is working to recapitalize existing assets and expand coverage of NII scanning technology with modern, integrated Commercial Off-the-Shelf (COTS) systems capable of greatly increasing scanning throughput.”

The goal with these algorithmic systems is to increase the number of vehicles, individuals, containers and packages scanned, while also reducing the time required by officers.  

When asked about the status of the program, a spokesperson told FedScoop that “CBP has been working closely with the algorithm providers to train the algorithms so they can transition from testing to operational readiness.”  

The spokesperson added: “The goal of implementing Anomaly Detection Algorithm (ADA) solutions is to provide computer-assisted analysis of nonintrusive inspection image images and other data that increase the thoroughness of inspections, reduce the time needed to conduct review of low-risk trade and travel, and support the analysis of complex inspections by U.S. Customs and Border Protection officers.” 

Other privacy threshold analyses have also been published in DHS in recent weeks.

One filing discusses the use of artificial intelligence from Matroid, which provides computer vision technologies, and works with CBP’s Office of Field Operations. The software is supposed to help the agency mine through images for signs of potential anomalies and threats. A pilot of the system is supposed to end this month, according to the documentation. 

Another analysis discusses CBP’s work with Clarifai, a different artificial intelligence platform. Similarly, the application was supposed to help the agency mine through images. One PTA references its work with Saildrone, which includes deploying autonomous vessels for maritime monitoring. The boats are supposed to identify “items of interest.” 

A combined version of a privacy threshold analysis for the agency’s Simplified Arrival Program, which is hosted on Amazon Cloud, was also published.