4 ways to address the failures we found along the US border’s “virtual wall”

The story is not abstract policy. MIT Technology Review, with Times of San Diego in the Dying on Camera project, documented people who walked through areas watched by advanced, AI-enabled towers, died nearby, and whose bodies went unnoticed for weeks or months. Broken towers, algorithms that failed to detect people, and agents who did not answer alerts sit at the center of that record.

Congress and the agencies are still pouring money into the same architecture. The U.S. is set to spend $1 billion to triple the size of the virtual wall by 2034, and Anduril’s newer towers keep winning contracts. If the network expands without an honest accounting of who dies in its blind spots, “AI surveillance” becomes a brand for coverage that does not cover.

We’re naming the harm plainly: people die in range of cameras that were sold as protection. Families lose the chance to know whether anyone saw them. That is a technology failure and a governance failure at once.

We’re reading this as a accountability brief, not a gadget review.

The four recommendations are boring in the best way—audits, death tracking, apprehension logging, postmortem reviews—and that is why they matter. New Mexico state representative Sarah Silva, shown that more than 18 people died near towers in one small stretch of desert near her district in 2024 alone, put it bluntly: this feels like it’s not working. The reporters analyzed nearly 600 towers against the roughly 800 the government says operate today, using Electronic Frontier Foundation mapping and satellite checks because CBP does not publish a comprehensive map. Their tally of more than 1,050 deaths near towers is already an undercount. An internal audit would not have those holes.

Useful AI, when it earns the label, detects a person in distress and triggers a rescue. That is the promise border tech sold. What the investigation describes is the opposite pattern: detection that fails, alerts that go unanswered, and no routine check of whether a recovered body fell inside a tower’s view. Footage typically vanishes after 30 days unless an internal review opens—usually for deaths in custody, not deaths near the cameras. So the question families and volunteers keep asking—did the cameras see them, and nobody came, or did nobody see them at all?—often cannot be answered after the fact.

The GAO trail is damning on process. Border Patrol’s death-tracking system has existed since 2007; national entry rules arrived only in 2021; outside researchers still find undercounts. The Missing Migrant Program, launched in 2017, was improving data collection by an April 2025 GAO look—but still lacked a plan to measure whether rescues rose or deaths fell. On effectiveness, CBP agreed in 2014 to log when technology helped an apprehension; by 2017 the data was unreliable enough that agents in Texas were crediting tower types that did not exist in the state; an effectiveness analysis finished in 2022 and stayed unpublished. A former official called the mandate “spitting in the wind.” Meanwhile the wall triples.

We’re for tools that actually save lives. We are against AI theater that expands faster than anyone measures its misses. Investigating every death against the tower map, preserving footage when someone dies in range, and publishing whether the system worked would turn “virtual wall” from a slogan into something that can be falsified. Without that, scaling Anduril-class towers is not safety engineering—it’s procurement momentum with a body count in the footnotes.

I’m watching whether CBP or Congress answers with an audit and death-as-failure reviews, or with another contract announcement.

This piece follows MIT Technology Review’s broader investigation into deaths near the southern border’s surveillance towers. One New Mexico site showed several deaths within months, sometimes only hundreds of feet apart, in clear sight of an advanced AI tower—exactly the pattern postmortem surveillance reviews are meant to catch.