Product1 publisher3 min readPublished
MIT Technology Review counted more than 1,050 deaths inside the range of CBP's surveillance towers
The investigation found deaths inside the advertised range of nearly two-thirds of the towers it analysed, and most of them were not in blind spots. In the New Mexico case it follows closely, landfill workers made the call to Border Patrol.
The Product Desk · Product desk
What happened
- MIT Technology Review matched nearly 4,000 locations of human remains against nearly 600 towers catalogued by the Electronic Frontier Foundation and counted more than 1,050 deaths within tower range since 2015.
- Its topographical analysis found some towers can see as little as 10% of their advertised surveillance area, and that most of the deaths did not fall in the towers' blind spots.
- More than 110 of those deaths were within range of Anduril's autonomous towers, which CBP has deployed since 2021.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint Advertised surveillance radius is the unit these tower deployments are described in, and the topographical finding means some of those radii deliver far less usable coverage than the figure implies.
- decision Anyone deploying a detection system has to choose whether acceptance turns on detections logged or on a named responder reaching the person inside a stated time.
- capability Remains records kept by No More Deaths and Humane Borders, plus EFF's tower list, let outsiders audit outcomes around a government sensor network without any agency telemetry.
- exposure CBP has credited the virtual wall with saving lives, and the 1,050 figure gives Congress and litigants a countable number to set against that claim.
What US Customs and Border Protection bought is a notification chain. Live video goes to a nearby control room. An alert can go to the government-issued smartphone in an agent's pocket, prompting the closest available agent to respond [3].
In southern New Mexico on April 8, 2024, the step that worked was a phone call from a landfill. Workers spotted Jose Morales Bernal that morning, saw him lying in the sand at 1 p.m., and called Border Patrol at 4 p.m. when he still had not moved [4][5]. Three hours sat between the sighting that mattered and the call [3]. Agents arrived 45 minutes after it, and he was dead [6]. One agent then called 911 and said the body was "probably one of the migrants crossing through there" [7]. Morales died 360 feet from the closest tower [8].
MIT Technology Review matched nearly 4,000 locations of human remains against nearly 600 towers catalogued by the Electronic Frontier Foundation and found more than 1,050 deaths inside tower range between 2015 and early 2026 [12][13]. Across about eleven years, that averages roughly 95 a year [1]. More than 110 of those deaths were within range of Anduril's autonomous towers, deployed since 2021 [16], about 22 a year [2].
The investigation's topographical analysis found towers with sight of as little as 10% of their advertised surveillance area, and it also found that most of the deaths did not happen in blind spots [15].
MIT Technology Review does not report whether any of the three towers detected Morales, or whether an alert reached a phone [4]. The record here does not convict the model, but it does separate advertised coverage from recorded outcome: the outcome inside the advertised coverage did not move, while CBP describes the virtual wall as persistent surveillance and situational awareness and has credited it with saving lives [11].
Nearly two-thirds of the towers analysed had a death inside their advertised range [14]. In the one morning reported in detail, the only recorded contact came from landfill employees rather than from the agency watching the ground [4].
Acceptance turns on two things. One is whether the sensor detects the event. The other is whether a named responder acts on it inside a stated time. Detection demos prove the first, and acceptance tests usually stop there, on a border stack with 25 years and billions of dollars behind it [11]. The only outcome measure available to reporters from outside was deaths inside advertised range.
An acceptance test written with a person in it would set a detection time, a time by which contact has to be made, and a recorded outcome. The tradeoff is real, because a response-time clause lets the sensor pass while the program fails its own test, and somebody inside the agency has to own that number.
What to watch
- Whether CBP publishes alert-to-dispatch data for the tower network, which would show where in the chain the notification stops.
- Whether any future CBP tower award sets acceptance criteria on response time and outcome, not detection performance alone.
- Whether the remaining Dying on Camera reporting documents specific missed detections or alerts that agents never actioned.