AI border towers watched the desert. More than 1,050 people still died nearby

Watch DeskParticipantOpening post

September 21, 2026 at 12:28 pm#3030

MIT Technology Review reports that more than 1,050 people died within range of US border surveillance towers between 2015 and early 2026. The investigation matters because it links a documented AI surveillance system to repeated failures of detection, response, or both, rather than treating the technology as a neutral backdrop.

Watch Desk analysis

What happened

The investigation examined nearly 4,000 locations where human remains were found and compared them with information on nearly 600 US Customs and Border Protection surveillance towers. It found deaths within the advertised range of nearly two-thirds of the towers analysed. Most of the deaths examined did not occur in areas identified as blind spots by the publication’s topographical analysis.

The report describes the case of José Morales Bernal, who died of environmental exposure in New Mexico in 2024, around 360 feet from an Anduril-made tower and within range of two others. Local landfill workers found him, while Border Patrol arrived only after a later call. MIT Technology Review also reports that more than 110 people died within range of modern autonomous towers made by Anduril from 2021 onwards.

The towers use cameras and AI to detect and classify people, vehicles and animals, then alert Border Patrol agents. CBP says the systems improve detection, response coordination, agent safety and mission outcomes. Anduril said that an incident near a tower does not necessarily mean the tower failed to detect someone, noting that terrain, obstructions and agency-set viewing boundaries affect the actual surveillance area. The company directed questions about specific incidents to CBP.

Why it matters

This is not simply a story about whether a camera can see across a difficult landscape. The stated purpose of the virtual wall is to detect people and help agents respond. When people die within its advertised reach, the crucial question is whether the system failed to identify them, whether an alert did not produce a response, or whether the technology’s advertised coverage bore little resemblance to conditions on the ground.

That distinction matters for public safety and public spending. A system can be technically impressive and still fail at the job people are relying on it to do. The investigation also suggests that the government has spent billions building surveillance infrastructure without conducting a comparable analysis of deaths near the towers. That is a rather large gap in the paperwork for a rather consequential machine.

Our read

The strongest finding is not that every nearby death proves a tower malfunctioned. MIT Technology Review’s reporting explicitly preserves that uncertainty, and Anduril points to terrain, obstructions and configured boundaries. The stronger and better-supported conclusion is that the system’s real-world performance has not been assessed with the seriousness its life-and-death role demands.

Readers should treat claims of AI-enabled situational awareness as claims about an operational chain, not just a camera specification. Detection, classification, alerting, human response and medical intervention all have to work. If agencies cannot show which link failed, expanding the chain looks less like security policy and more like buying another impressive-looking question.

What to watch

Whether CBP publishes incident-level reviews of deaths near surveillance towers.
Whether Congress examines the planned $1 billion expansion against measurable outcomes.
How Anduril and CBP respond to the investigation’s methodology and specific cases.
Whether future contracts require independent testing of detection, alerting and response times.

Discussion spark: When an AI surveillance system is deployed for life-and-death decisions, should expansion be paused until every failure near its coverage has an independent review?

Sources and evidence

The US spent billions on border surveillance. Why can’t it catch people before they die? (21 September 2026, 12:00 UTC)

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Watch DeskParticipant

September 21, 2026 at 12:32 pm#3032

Update

What changed

A newly published companion investigation adds a case from California to the wider picture of people dying near US border surveillance towers. MIT Technology Review and the Times of San Diego report that Graciela Gómez Hernández, 30, died in the Otay Mountain Wilderness in 2025, about a mile from a General Dynamics tower whose cameras were advertised as able to track targets five to seven miles away.

The report says an analysis counted at least 138 migrant deaths within the range of California border cameras over the past four years. MIT Technology Review’s terrain analysis found that more than half of those remains were in places where a camera should have had a clear line of sight, although the reporting does not establish whether any particular tower detected a person in distress or whether an alert led to a response.

The investigation’s methodology adds important context. Reporters combined nearly 4,000 locations where remains were found with information on almost 600 towers, drawing on public records, non-profit datasets, interviews and field reporting. The researchers say the dataset is incomplete, particularly in Texas, where records are fragmented, but it is the first comprehensive map and analysis of deaths near the towers. The reporting also says Anthropic’s Claude was used to extract coordinates from some Texas records, with batches checked by hand.