The US spent billions on border surveillance. Why can’t it catch people before they die?

More than 1,050 migrants died within range of US border surveillance towers between 2015 and early 2026, according to a first-of-its-kind investigation by MIT Technology Review and Times of San Diego. The finding challenges claims by the Department of Homeland Security and Customs and Border Protection that AI-enabled surveillance systems improve detection and save lives, revealing instead a pattern of undetected deaths occurring within camera range—sometimes just a few hundred feet away—even as the government plans to spend $1 billion deploying 1,497 more towers by 2034.

José Morales Bernal, 31, died on April 8, 2024, in the desert near Sunland Park, New Mexico, within sight of three surveillance towers built by defense contractor Anduril Industries. His body lay undetected for hours despite being within 360 feet of the closest tower—equipped with AI-powered cameras and radar designed to automatically detect and track people crossing the border. Landfill workers, not Border Patrol, discovered him that afternoon. When agents arrived, a responding officer's 911 call suggested uncertainty about whether Morales had passed through the agency's surveillance systems earlier that day, despite his proximity to the towers.

The investigation cross-referenced nearly 4,000 locations where human remains were found—drawn from nonprofit records and hundreds obtained through public records requests in Texas—with information on nearly 600 towers identified by the Electronic Frontier Foundation. The researchers found more than 1,050 deaths within tower range since 2015, including more than 110 deaths within range of Anduril's autonomous towers since 2021, often in clear terrain where camera views were unobstructed. Nearly two-thirds of all towers analyzed were within range of at least one death. The analysis identified two types of failures: towers failing to detect, and agents failing to respond—each a system failure in its own right.

The findings expose long-standing technological and institutional shortcomings. Earlier Remote Video Surveillance System (RVSS) towers, operated by contractors like General Dynamics, suffered from information overload: a single agent monitoring 30 screens watched up to 850 square miles, making detection of activity near-impossible. By 2024, more than 66% of first-generation towers were unusable, despite annual operating budgets of $50 million to $100 million. Anduril's newer autonomous systems promised to solve this via AI classification, but agents reported algorithm failures: groups covered with tarps or blankets evaded detection, large groups fragmented and were missed during handoffs between towers, and false negatives—people the systems failed to spot—persisted despite retraining efforts. One agent described an incident where a car crashed into the Rio Grande less than 100 feet from an Anduril tower; the camera never captured it.

CBP has conducted no formal investigation into whether surveillance failures contributed to individual deaths, and footage is overwritten every 30 days unless retained as part of an active investigation—which begins only if someone dies in custody. Officials across four presidential administrations told researchers they believed deaths near the virtual wall were rare or nonexistent; none could point to comparable government analysis. The agency did not address the investigation's detailed questions about specific incidents or broader evaluation methods. The government has spent 25 years and billions building this system, yet lacks basic accountability mechanisms for measuring whether it works or why people die within its view. With new federal funding approved and less oversight in place following DHS restructuring in October 2025, CBP is expanding aggressively despite unanswered questions about the lives already lost.