CBP Signs Clearview AI Deal to Use Face Recognition for ‘Tactical Targeting’

United States Customs and Border Protection plans to walk $225,000 for a twelvemonth of entree to Clearview AI, a face recognition instrumentality that compares photos against billions of images scraped from the internet.

The woody extends entree to Clearview tools to Border Patrol’s office quality part (INTEL) and the National Targeting Center, units that cod and analyse information arsenic portion of what CBP calls a coordinated effort to “disrupt, degrade, and dismantle” radical and networks viewed arsenic information threats.

The declaration states that Clearview provides entree to “over 60+ cardinal publically disposable images” and volition beryllium utilized for “tactical targeting” and “strategic counter-network analysis,” indicating the work is intended to beryllium embedded successful analysts’ day-to-day quality enactment alternatively than reserved for isolated investigations. CBP says its quality units gully from a “variety of sources,” including commercially disposable tools and publically disposable data, to place radical and representation their connections for nationalist information and migration operations.

The statement anticipates analysts handling delicate idiosyncratic data, including biometric identifiers specified arsenic look images, and requires nondisclosure agreements for contractors who person access. It does not specify what kinds of photos agents volition upload, whether searches whitethorn see US citizens, oregon however agelong uploaded images oregon hunt results volition beryllium retained.

The Clearview declaration lands arsenic the Department of Homeland Security faces mounting scrutiny implicit however look designation is utilized successful national enforcement operations acold beyond the border, including large-scale actions successful US cities that person swept up US citizens. Civil liberties groups and lawmakers person questioned whether face-search tools are being deployed arsenic regular quality infrastructure, alternatively than constricted investigative aids, and whether safeguards person kept gait with expansion.

Last week, Senator Ed Markey introduced legislation that would barroom ICE and CBP from utilizing look designation exertion altogether, citing concerns that biometric surveillance is being embedded without wide limits, transparency, oregon nationalist consent.

CBP did not instantly respond to questions astir however Clearview would beryllium integrated into its systems, what types of images agents are authorized to upload, and whether searches whitethorn see US citizens.

Clearview’s concern exemplary has drawn scrutiny due to the fact that it relies connected scraping photos from nationalist websites astatine scale. Those images are converted into biometric templates without the cognition oregon consent of the radical photographed.

Clearview besides appears successful DHS’s precocious released artificial quality inventory, linked to a CBP aviator initiated successful October 2025. The inventory introduction ties the aviator to CBP’s Traveler Verification System, which conducts look comparisons astatine ports of introduction and different border-related screenings.

CBP states successful its nationalist privateness documentation that the Traveler Verification System does not usage accusation from “commercial sources oregon publically disposable data.” It is much likely, astatine launch, that Clearview entree would alternatively beryllium tied to CBP’s Automated Targeting System, which links biometric galleries, watchlists, and enforcement records, including files tied to caller Immigration and Customs Enforcement operations successful areas of the US acold from immoderate border.

Clearview AI did not instantly respond to a petition for comment.

Recent investigating by the National Institute of Standards and Technology, which evaluated Clearview AI among different vendors, recovered that face-search systems tin execute good connected “high prime visa-like photos,” but falter successful little controlled settings. Images captured astatine borderline crossings that were “not primitively intended for automated look recognition” produced mistake rates that were “much higher, often successful excess of 20 percent, adjacent with the much close algorithms,” national scientists say.

The investigating underscores a cardinal regulation of the technology: NIST recovered that face-search systems cannot trim mendacious matches without besides expanding the hazard that the systems neglect to admit the close person.

As a result, NIST says agencies whitethorn run the bundle successful an “investigative” mounting that returns a ranked database of candidates for quality reappraisal alternatively than a azygous confirmed match. When systems are configured to ever instrumentality candidates, however, searches for radical not already successful the database volition inactive make “matches” for review. In those cases, the results volition ever beryllium 100 percent wrong.