A 15-month investigation by MIT Technology Review and Times of San Diego examined why so many migrants have died near government surveillance towers along the U.S.–Mexico border. The project produced the first comprehensive map and analysis of deaths near these towers, detailing how the dataset was constructed, which methodological choices were made, and what the resulting data can and cannot reveal.
Methods — sources and datasets
The investigation required knowing where migrants died and where and when U.S. Customs and Border Protection (CBP) towers were installed. The team analyzed cases dating back to 2015 so the dataset would encompass different border policies, presidential administrations, and tower technologies.
The final dataset combined previously compiled public databases from nonprofit organizations (No More Deaths, Humane Borders, and the Electronic Frontier Foundation) with new records obtained directly from more than a dozen agencies through public-records requests.
Public-records requests in Texas
Texas presented a major gap: unlike Arizona, Texas has no statewide initiative to publish border-death records online, and unlike New Mexico, it leaves death investigations to individual counties. The investigation identified 17 Texas counties relevant to the border corridor under study: Culberson, Jeff Davis, Presidio, Terrell, Val Verde, Kinney, Maverick, Dimmit, Webb, Zapata, Jim Hogg, Starr, Hidalgo, Cameron, Kenedy, Brooks, and Duval.
Starting in July 2025, MIT Technology Review filed records requests with sheriffs’ offices in these counties, since sheriffs are often first on scene and their reports tend to include detailed location information. Obtaining records was slow and sometimes costly: some counties provided records for free, while others charged between $300 and $1,300. Several requests stalled for months or remain unfilled more than a year after filing. Counties warned some records could be incomplete, lost, damaged, or misplaced during staff transitions.
Many agencies do not record whether a death is believed to be a border crossing, forcing manual review of records. The team received usable records from 14 of the 17 counties (records were still outstanding from Dimmit and Duval; Maverick’s per-record fees proved prohibitive). Agencies supplied more than 4,000 pages of case reports; some were handwritten.
For Kenedy, Webb, and Hidalgo counties — which supplied especially large volumes — MIT Technology Review used Anthropic’s Claude (via API) to extract coordinates from each case, then spot-checked batches by hand to verify accuracy. Agencies provided records on a rolling basis from July 2025 to February 2026; most did not specify how current their records were.
After excluding cases that clearly were not migrant-crossing deaths (for example, people who died at home) and several hundred cases lacking sufficient location detail, the Texas records contributed more than 1,500 cases to the final analysis.
Arizona records via Pima County and Humane Borders
The Pima County Office of the Medical Examiner (PCOME) investigates deaths across Arizona’s border counties (except Yuma) and publishes a monthly data portal that it shares with Humane Borders, which maintains a map and database. MIT Technology Review included PCOME/Humane Borders data from 2015 through April 2026. Humane Borders notes the precision of location descriptions; the investigation included only cases with GPS coordinates precise to within 300 feet. PCOME/Humane Borders data added more than 1,700 cases to the analysis.
No More Deaths data
To cover California, New Mexico, Yuma County (Arizona), and El Paso and Hudspeth counties in Texas, the team used the No More Deaths database. The organization has tracked migrant deaths since 2004 and documents its methodology. The investigation included only No More Deaths cases with location accuracy within a quarter-mile and used decomposition-level notes where available to estimate death windows. No More Deaths contributed nearly 1,000 cases to the analysis.
Aggregate case count
After merging sources and excluding deaths without sufficient location or timing information, the team analyzed nearly 4,000 cases in total.
Tower data — which towers were mapped and why
For every tower included, the team needed its GPS location, its type (to estimate how far it could see), and when it was installed.
-
The Electronic Frontier Foundation (EFF) has been tracking and mapping CBP surveillance towers since 2023 using field reporting, satellite imagery, and public records. EFF’s tally is an undercount: government records indicate roughly 800 towers are deployed, while EFF has logged about 600. The analysis therefore misses deaths near towers not yet catalogued by EFF.
-
MIT Technology Review began with the EFF database (April 2026), supplemented EFF’s records with information about towers that have since been removed, and verified tower GPS coordinates via satellite imagery. EFF reference numbers were generally used, with changes to ensure unique identifiers.
-
The study focused on three main tower types; other EFF-mapped sites that did not fit these categories (for example, installations at checkpoints or on buildings) were excluded because it was difficult to determine their advertised surveillance ranges.
-
To determine when towers were present, the team inspected historical satellite imagery to log the first observed appearance and, when relevant, the last observed presence. For some towers EFF had already established timelines; MIT Technology Review verified and extended those checks. Where imagery was insufficient, they used Google Street View, government records, or EFF’s photographic evidence.
Limitations: historical imagery is not continuous for every location, so short-term removals or re-installations might be missed; and visible presence does not prove a tower was operational (powered, transmitting data, or integrated into monitoring systems).
The researchers also assigned minimum and maximum heights to towers (based on EFF research and imagery) because height affects how far a tower can see.
Analysis: matching deaths to towers (distance, timing, viewshed)
For a death to be considered within range of a given tower, two criteria had to be met:
- Distance: the location of the remains had to be within the tower’s published surveillance range.
- Timing: the tower had to be present during the estimated window when the person died.
Distance calculations were performed by software that paired each human-remains record with nearby towers and computed distances.
Timing estimates came from explicit time-of-death notes in medical examiner reports where available. When not provided, the team used decomposition level, cause of death, photographs, police reports, news accounts, and expert consultation to build approximate death windows. The team worked with Greg Hess, director of the Pima County Office of the Medical Examiner, to translate scene descriptions and decomposition notes into rough timing estimates. If an individual’s entire estimated death window fell after a tower’s first observed appearance, the match was considered confident; if any part of the window predated the tower, the case was excluded.
For many skeletal remains without a specific estimate, the team conservatively assumed at least six months of presence and used a three-year cutoff: if a tower had been at the site for at least three years before discovery, the case was included; if not, it was excluded.
Viewshed (line-of-sight) analysis used topographical data to estimate whether terrain would have blocked a tower’s view of the location where remains were found. The terrain model uses 30-meter cells (about 98 feet) and the analysis drew a line from the tower’s estimated height to an assumed 5 ft 6 in (1.68 m) person at the remains site; if intervening terrain rose high enough to block that line, the view was marked obstructed. The method provides a rough estimate, not a perfect reconstruction; smaller terrain features, vegetation density, and building heights are imperfectly represented. The researchers flagged towers in or near urban areas where buildings might have impeded sightlines.
What the data can and cannot answer
A central unanswered question for families is whether the U.S. government ever saw a person attempting a crossing and failed to intercept them, or whether the person was never observed at all. Records and data generally cannot answer this: Border Patrol does not disclose which alerts corresponded to any given death, and footage is not retained unless an internal investigation is triggered (which typically occurs only if a migrant dies while in custody).
To supplement records, the investigation conducted dozens of interviews with current and former Border Patrol agents, CBP and DHS staff, presidential advisors, medical examiners, sheriffs, humanitarian volunteers, and technology company employees. Reporters also identified and interviewed friends and family of people who died near towers, following trauma-informed reporting practices.
The MIT Technology Review requested interviews and border visits starting in September 2025. Local-level approval for a Big Bend visit was granted in July 2026 but later denied by DHS headquarters. The team sent questions to CBP, Anduril, General Dynamics, and Elbit; CBP and Anduril responded but did not substantively address the questions, General Dynamics redirected to CBP, and Elbit did not respond.
Key figures and timelines
- Investigation length: 15 months.
- Case window analyzed: 2015 through April 2026.
- Data received from agencies on a rolling basis: July 2025–February 2026.
- Texas cases included from records requests: more than 1,500.
- Humane Borders (Pima County) cases included: more than 1,700.
- No More Deaths cases included: nearly 1,000.
- Total cases analyzed after filtering: nearly 4,000.
- Estimated number of CBP towers deployed (government figures): roughly 800; EFF has logged about 600 towers.
Why this matters
By matching nearly 4,000 deaths to mapped surveillance infrastructure and estimating whether towers were present and could have had a line of sight, the investigation sheds light on gaps in surveillance effectiveness, data availability, and record-keeping across jurisdictions. The analysis highlights how limited public data, inconsistent local practices, and operational ambiguity about whether towers were functional complicate efforts to assess whether surveillance technology has reduced deaths or failed those it was intended to detect.
Acknowledgments: Minho Kim (Stanford University), Greg Hess (Pima County Office of the Medical Examiner), Stephanie Leutert (University of Texas at Austin), and Don White (Brooks County Sheriff’s Office) assisted with the project.
