US border patrol seeks AI to ‘see through walls’ in urban areas

The US Customs and Border Protection (CBP) is seeking “advanced artificial intelligence” (AI) technologies, increasingly sophisticated autonomous systems, and even the ability to see through walls to surveil urban residential areas.

A CBP presentation for an “Industry Day” summit with private sector representatives, obtained by The Intercept, reveals a detailed list of technologies the agency hopes to acquire. This list includes technologies such as satellite connectivity for surveillance towers along the border and improved radio communications.

The list also indicates that state-of-the-art, AI-powered surveillance technologies will be central to the Trump administration’s anti-immigrant campaign, which will extend deep into the North American continent.

The recent passage of Trump’s comprehensive “big, beautiful law” funnels tens of billions of dollars to the Department of Homeland Security. While a large portion of this funding will go to Immigration and Customs Enforcement (ICE) to support the administration’s arrest and deportation operations, a significant part has been allocated for purchasing new technology and equipment for the federal agencies tasked with preventing immigrants from entering the country: the CBP, which manages the nation’s border surveillance system, and its subsidiary, the US Border Patrol.

One page of the presentation outlines the wish list for the Border Patrol’s Law Enforcement Operations Division, stating that the agency needs “advanced artificial intelligence to detect and track suspicious activity in the urban environment” due to the “challenges” posed by “densely populated areas.” What constitutes “suspicious activity” is not specified.

The section mentioning AI-powered urban surveillance appears on a page dedicated to the operational needs of the Border Patrol’s “Coastal AOR” (area of responsibility). This area covers the entire southeastern US, from Kentucky to Florida. A page describing the “Southern AOR,” which includes the entire interior of Nevada and Oklahoma, similarly notes the need for “advanced intelligence to detect suspicious patterns” and “long-range surveillance” because “it is difficult to distinguish normal activities from suspicious activities in urban environments.”

Although the Fourth Amendment of the US Constitution provides protection against arbitrary police searches, federal law grants immigration agencies the authority to conduct detentions and searches without a warrant within 100 miles of the Canadian, Mexican, or US coastline borders. This zone encompasses many of the largest US cities, including Los Angeles, New York, and the entire state of Florida.

The document does not specify which surveillance methods or “advanced artificial intelligence” tools will be used in urban environments. In the Southwest region, residents of towns like Nogales and Calexico are already being monitored by surveillance towers placed in their neighborhoods.

A 2014 border surveillance privacy impact assessment by the DHS warned that these towers “may collect information about individuals or activities beyond CBP’s authority.” Since the border also includes populated areas, video cameras can capture individuals as they go about their daily lives, entering places or engaging in activities. For example, this could include “videos of an individual entering a doctor’s office, attending public rallies, social events, or meetings, or associating with other individuals.”

Last year, the Government Accountability Office found that the DHS tower surveillance program failed on all six of its privacy policies designed to prevent such overreach.

CBP is also known to use “artificial intelligence” tools to uncover “suspicious activities,” according to agency documents. Among the AI applications listed in DHS’s 2024 inventory is the Rapid Tactical Operational Reconnaissance program (RAPTOR), which “uses artificial intelligence to enhance border security through real-time surveillance and reconnaissance.” The AI system processes data from radar, infrared sensors, and video surveillance systems to detect and track suspicious activities along US borders.

Spencer Reynolds, a former lawyer at the Department of Homeland Security specializing in intelligence matters, says, “The Border Patrol’s increasing immigrant raids and harsh responses to protests show that this agency operates heavily not just in remote deserts, but also in cities. Every day, its activities are based less on suspicion and more on racial and ethnic profiling. References to operations in ‘densely populated areas’ are alarming signals that expanded operations or tracking plans are being made in American neighborhoods.”

The automation of immigration enforcement has been a priority for the Department of Homeland Security for years. An example of this is the bipartisan effort to expand the use of machine learning-based surveillance towers, like those sold by arms manufacturer Anduril on the southern border.

The agency’s 2024-2028 strategy document states, “Autonomous technologies will enhance USBP’s ability to detect, identify, and classify potential threats in the operational environment. Once a threat is detected and classified, autonomous technology will enable USBP to monitor threats in near real-time through an integrated network.”

The automation desired by the Border Patrol appears to rely heavily on computer vision, a type of machine learning that excels at pattern matching to find objects like humans, cars, or other “objects of interest.” This system does not require teams of human agents to monitor camera feeds and other sensors around the clock. The Border Patrol’s presentation includes numerous requests for small unmanned aerial vehicles that incorporate AI technologies to assist in the “detection, tracking, and classification” of targets.

A computer system that analyzes numerous photos of trucks driving in the desert can become effective at recognizing similar vehicles in the future. However, efforts to algorithmically label human behavior as “suspicious” based solely on appearance have been criticized by some AI academics and civil liberties advocates as error-prone, overly subjective, even pseudoscientific, and often based on ethnic and religious stereotypes.

Efforts to apply predictive techniques based on surveillance data from entire urban or residential areas could further increase the risks of bias and inaccuracy.