Our 15-month investigation into death and surveillance along the US-Mexico border began with a simple yet haunting question: Why did so many people die near government surveillance towers specifically meant to help track and apprehend them?
This groundbreaking story is part of "Dying on Camera," a collaborative reporting project between MIT Technology Review and the Times of San Diego. Journalists across both newsrooms spent the past year examining the systemic failures of border surveillance technology while uncovering the deeply human stories of those who lose their lives in the unforgiving borderlands.
To find answers, our team combed through thousands of pages of government documents, traveled to the border multiple times, and interviewed more than 45 individuals. Our sources included current and former White House advisors, presidential appointees, Border Patrol agents, medical examiners, local sheriffs, humanitarian volunteers, and employees of tech companies.
The culmination of this exhaustive reporting is the first comprehensive map and analysis tracking deaths near border surveillance towers. Understanding how this project came together requires looking at the massive data architecture, the difficult decisions made along the way, and what the data can—and cannot—tell us about the modern border.
Our Data
The core of our investigation relied on synchronizing two massive sets of information: knowing precisely where migrants have died, and mapping the exact locations and installation timelines of US Customs and Border Protection (CBP) towers. To capture a holistic view of modern border enforcement, we analyzed cases dating back to 2015, a timeframe spanning multiple presidential administrations, shifting federal policies, and evolving tower technologies.
Migrant Deaths
The US-Mexico border is widely recognized as the deadliest land border in the world. Border Patrol estimates indicate that more than 10,000 people have died during unauthorized crossings since 2000, though experts agree this figure is a severe undercount. Publicly available information regarding where, when, and who these individuals were varies wildly. Many remains are never recovered, and for those that are, there is no unified national protocol for recording or handling the records.
We established a strict threshold for inclusion. A case was considered for our analysis only if we could verify three foundational details: whether the individual was believed to have been crossing the border, where their remains were discovered, and a reasonable approximation of when they died.
Our dataset was constructed by merging existing databases compiled by humanitarian organizations and civil liberties groups—such as No More Deaths, Humane Borders, and the Electronic Frontier Foundation—with thousands of new records we independently obtained from more than a dozen government agencies.
Public Records Requests in Texas
Texas proved to be the most significant missing link in existing databases of migrant deaths. Unlike Arizona, which maintains an online public portal for border death records, or New Mexico, which centralizes investigations, Texas leaves death investigations entirely up to individual counties. While a handful of dedicated independent researchers had compiled fragments of local data, no comprehensive or current public record existed.
We identified 17 Texas counties relevant to our geographic scope: Culberson, Jeff Davis, Presidio, Terrell, Val Verde, Kinney, Maverick, Dimmit, Webb, Zapata, Jim Hogg, Starr, Hidalgo, Cameron, Kenedy, Brooks, and Duval. Other border counties exist, but because they lack areas covered by surveillance towers, they were excluded.
In most of these jurisdictions, locally elected administrative judges known as justices of the peace oversee death investigations. However, records requests submitted to these judges can stall for years, with some outstanding requests remaining unanswered more than a year later. Furthermore, justices of the peace occasionally omit crucial scene details, with some even declaring individuals deceased remotely over FaceTime rather than visiting the scene in person.
To secure reliable data, MIT Technology Review began filing public records requests with the sheriffs of these 17 counties in July 2025. Because sheriff’s departments are frequently the first responders when remains are discovered, their incident reports often contain granular details regarding recovery locations. However, relying on sheriff records inherently undercounts total migrant deaths in Texas, as it excludes cases handled exclusively by municipal or state police.
Securing these records required near-daily phone calls to local offices. While some agencies provided documents free of charge, others charged between $300 and $1,300 to fulfill requests. Several counties warned that their records were incomplete, noting that unknown numbers of files had been lost, damaged, or misplaced during administrative transitions between sheriffs.
Compounding these administrative hurdles, most agencies do not track whether a deceased individual was believed to be an unauthorized border crosser. While some counties maintain highly granular tracking systems for other crimes—such as Zapata County tracking whether jewelry was stolen during a burglary, or Cameron County noting if an intruder entered through a chimney—migrant deaths lack a dedicated classification. Consequently, local offices had to manually search archives, making the process slow, expensive, and vulnerable to human error.
Ultimately, we received usable records from 14 of the 17 counties, while Dimmit and Duval counties have yet to respond, and Maverick County imposed a cost-prohibitive per-record fee. In total, agencies provided more than 4,000 pages of police reports, some of which were handwritten.
For the vast majority of these cases, reporters extracted relevant location and circumstance data by hand. For Kenedy, Webb, and Hidalgo counties, which delivered massive volumes of documents, we utilized Anthropic’s Claude via API to parse case reports and extract geographical coordinates of where remains were found, subsequently verifying batches of data manually for accuracy. Agencies delivered records on a rolling basis between July 2025 and February 2026.
After vetting the data to exclude irrelevant incidents, such as individuals who passed away inside Border Patrol stations or died in car crashes during pursuits, we incorporated more than 1,500 verified cases from Texas into our final analysis.
Accessing Arizona Records Directly
In Arizona, the Pima County Office of the Medical Examiner (PCOME) manages death investigations for all border counties except Yuma. PCOME maintains a robust data portal tracking migrant deaths that is updated monthly and shared with Humane Borders, which publishes an interactive map and database. We integrated data from this portal covering the period from 2015 through April 2026.
To ensure high geographic accuracy, we filtered for cases with GPS coordinates precise to within 300 feet—a standard data collection practice implemented around 2012. These records also included estimated post-mortem intervals, ranging from less than a day to several months. In total, we incorporated more than 1,700 cases from Humane Borders into our study.
Data Collected by No More Deaths
To evaluate deaths in California, New Mexico, Yuma County in Arizona, and El Paso and Hudspeth counties in Texas, we partnered with the nonprofit organization No More Deaths, which has meticulously documented migrant fatalities since 2004. We restricted our inclusion criteria to cases where the recovery location was verified within a quarter-mile radius.
Because No More Deaths records often included notes on the degree of decomposition, MIT Technology Review consulted with Dr. Greg Hess, director of the Pima County Office of the Medical Examiner, to establish a standardized framework for translating physical descriptions into reliable estimates of when an individual died. Combining these sources allowed us to integrate nearly 1,000 additional cases into our master dataset, bringing our total reviewed pool to nearly 4,000 cases.
Surveillance Towers
Evaluating whether surveillance infrastructure could have detected a migrant in distress required mapping three core variables: the physical location of each tower, the specific tower type—which determines its maximum optical and thermal range—and the precise date of its installation.
The Electronic Frontier Foundation (EFF) has tracked CBP surveillance towers since 2023 using satellite imagery, ground-level reporting, and government records. While government estimates suggest roughly 800 towers are deployed nationwide, EFF has cataloged approximately 600, meaning our analysis inevitably misses deaths occurring near uncatalogged towers. Working closely with EFF director of investigations Dave Maass, our team verified coordinates using satellite imagery and established historical timelines for when every tower was erected or removed.
We focused our study on three primary tower types, excluding checkpoint and station-mounted systems due to a lack of publicly available range data. To determine the operational parameters of these towers, we reviewed technical specifications from manufacturers, examined EFF research, analyzed redacted government documents, and interviewed former Border Patrol agents.
Determining whether a tower was active when a death occurred involved meticulous cross-referencing of historical satellite imagery, Google Street View archives, and photographs. While historical imagery is not captured daily, most border towers are semi-permanent or permanent structures. Multiple Border Patrol agents confirmed that even mobile autonomous towers are rarely relocated in practice. However, our analysis establishes the physical presence of a tower, not whether it was actively powered and transmitting data at any given micro-moment.
Our Analysis
With our databases established, every case of human remains was evaluated against two strict criteria: distance and timing. The death had to occur within the published surveillance range of a nearby camera, and the tower had to be physically present at the time of the person’s death.
Dr. Minho Kim, a postdoctoral researcher at Stanford University specializing in geospatial and wildfire mapping, led our distance and topographical analysis. Using specialized software, Kim’s team mapped the spatial relationship between human remains and surveillance towers, calculating exact distances and conducting viewshed analyses.
A topographical viewshed analysis models whether line-of-sight between a tower and a set of remains was blocked by natural terrain features, such as ridges or hills. By calculating the elevation of a 98-foot grid against an estimated human height of five feet, six inches, the software flagged whether a tower’s perspective was obstructed. While this model cannot account for minor micro-topography or localized vegetation, it provides a crucial baseline for understanding the physical limitations of camera sightlines in rugged desert terrain.
Beyond the Data
The most profound and agonizing question for the families of the deceased is whether the United States government saw someone attempting to cross the border and failed to intervene—or never saw them at all.
Answering this definitively through static records remains nearly impossible. Border Patrol does not routinely disclose which digital alerts were triggered around the time of a migrant death, and camera footage is not permanently retained unless an internal investigation is launched, a protocol triggered solely by deaths occurring directly in custody.
To bridge this investigative gap, MIT Technology Review conducted dozens of interviews with former Department of Homeland Security officials, presidential advisors, and Border Patrol agents. These discussions revealed unpublicized details regarding how agents utilize surveillance feeds, highlighted operational shortcomings, and exposed how rarely the effectiveness of these multi-million-dollar virtual walls is internally evaluated.
We also sought out friends and family members of individuals who died within sight of surveillance towers, approaching interviews with strict adherence to trauma-informed journalistic practices. Requests for site visits and interviews with active Border Patrol leadership, initiated in September 2025, ultimately stalled despite initial local approvals in Big Bend, Texas, before being formally denied by DHS headquarters prior to publication.
When approached for comment prior to publication, CBP and manufacturing firm Anduril responded without substantively addressing our specific findings. General Dynamics directed inquiries back to CBP, while Elbit did not respond to requests for comment.