A major investigation into the United States’ technology-driven border security system has raised serious questions about the effectiveness of surveillance towers, artificial intelligence, and automated detection systems deployed along the US–Mexico border. The investigation found that more than 1,050 people died within the advertised surveillance range of border towers between 2015 and early 2026, highlighting a troubling gap between the capabilities promised by the technology and the outcomes observed on the ground.
Published by MIT Technology Review in collaboration with Times of San Diego on September 21, 2026, the investigation examined thousands of records documenting migrant deaths alongside information about nearly 600 border surveillance towers. The findings suggest that the United States’ so-called virtual wall has repeatedly failed to ensure that people in distress are identified and located in time for authorities to intervene. <Cite refs={[“turn396576view0”]}/>
The findings arrive as the federal government continues investing billions of dollars in advanced surveillance infrastructure. Although officials describe these systems as important tools for improving border security and coordinating responses, the investigation raises concerns about technical limitations, staffing challenges, inadequate oversight, and the absence of systematic reviews when people die near monitored areas.
What the Investigation Revealed
The investigation compared nearly 4,000 locations where human remains were discovered with the locations, operating periods, and estimated viewing ranges of surveillance towers identified by the Electronic Frontier Foundation.
Researchers examined cases dating back to 2015, combining records from humanitarian organizations, medical examiners, local authorities, and other sources. They also conducted interviews with more than 45 people, including current and former Border Patrol personnel, government officials, medical examiners, humanitarian workers, and representatives of technology companies.
The resulting analysis identified more than 1,050 deaths within the advertised range of border surveillance towers. The researchers found deaths within range of nearly two-thirds of the towers they analyzed. Their estimates also indicated that more than 110 people had died within the surveillance range of modern autonomous towers developed by Anduril since 2021. <Cite refs={[“turn396576view0″,”turn319691search1”]}/>
These figures do not establish that every death could have been prevented or that a surveillance system necessarily detected each person before death. They do, however, raise important questions about whether the technology is performing its intended functions and whether authorities are responding effectively when people are identified.
The distinction matters because the virtual wall is intended to do more than record activity. Its purpose includes detecting people crossing the border, helping agents locate them, and supporting timely law enforcement or emergency responses.
A Fatal Case Raises Questions About AI Surveillance
One case highlighted by the investigation involved José Morales Bernal, who crossed into the United States near Sunland Park, New Mexico, on April 8, 2024.
According to the report, Morales was within the advertised range of three surveillance towers operated by US Customs and Border Protection. The towers, developed by Anduril, use cameras and artificial intelligence to identify and track people and can transmit alerts to agents in the surrounding area.
Despite the nearby surveillance infrastructure, local landfill workers—not Border Patrol agents—were the first to notice Morales. Workers reportedly saw him lying in the desert later that day and contacted Border Patrol after realizing he had not moved. Agents arrived approximately 45 minutes after the call and found him dead.
Morales was discovered approximately 360 feet from the closest surveillance tower. An autopsy attributed his death to environmental exposure. The investigation questioned why the surrounding surveillance systems had not resulted in an earlier intervention. <Cite refs={[“turn396576view0”]}/>
The case illustrates the difference between having a camera pointed toward an area and having a fully functioning response system. Even if a surveillance device can observe a location, the system must still identify relevant activity, communicate accurate information, and enable personnel to reach the person before a situation becomes fatal.
The investigation documented other deaths near surveillance infrastructure, including cases involving people who reportedly died in areas where researchers estimated that cameras had a clear line of sight. These findings raise concerns about both detection failures and delays in responding to people in distress.
Technical Limitations Undermine the Virtual Wall
The virtual wall includes several generations of technology, ranging from conventional remote video surveillance systems to integrated fixed towers and newer autonomous platforms.
These systems use combinations of cameras, radar, thermal imaging, motion sensors, and automated software to monitor remote terrain. Their capabilities vary, and their actual performance depends on the equipment installed, environmental conditions, terrain, and the way each system is configured.
The investigation found that some towers had visibility across only a small fraction of their advertised surveillance areas. Researchers estimated that certain towers could see as little as 10 percent of the area they were supposed to cover. Nevertheless, their analysis also indicated that most of the recorded deaths did not occur in locations where terrain alone appeared to block the towers’ view.
This suggests that geographical blind spots are only part of the problem.
Even when a person is technically visible, successful detection may depend on the camera’s orientation, image quality, operating condition, software performance, and the ability of personnel to recognize what is happening. A system may also detect movement without determining that someone needs urgent medical assistance.
Environmental exposure, dehydration, injury, and drowning can become fatal before authorities arrive. In such circumstances, effective surveillance must be linked to a response process that can quickly assess the situation and dispatch appropriate assistance.
The investigation also found longstanding reliability problems involving older camera systems. Former agents described equipment that frequently malfunctioned, offered limited functionality, or required operators to monitor numerous screens simultaneously.
These issues illustrate a broader limitation of surveillance technology: increasing the number of cameras does not automatically guarantee that the information they produce will be acted upon effectively.
Human Oversight and Delayed Responses
Another central finding concerns the role of Border Patrol personnel in monitoring surveillance systems and responding to alerts.
Earlier generations of the virtual wall relied heavily on agents watching multiple camera feeds while managing other responsibilities. Personnel interviewed by the investigators described situations in which operators were expected to monitor numerous screens alongside sensor alerts, radio communications, record checks, and coordination duties.
This workload can make it difficult to identify every significant event, particularly across large areas of remote terrain. Even a functioning camera system may provide limited protection if no one notices a person in distress or if the information does not reach agents who can respond promptly.
Automation was introduced partly to address this problem. Newer systems can use radar and computer vision to identify movement, classify objects, track people, and generate alerts without requiring an operator to continuously inspect every camera feed.
However, automation does not eliminate the need for effective human oversight. Alerts must be accurate, sufficiently informative, and delivered to personnel who are available to act. False alarms, missed detections, unclear responsibilities, and competing operational demands can all undermine the effectiveness of an automated system.
The investigation concluded that both technological failures and failures to respond were occurring. It also reported that former agents and officials could not identify a comparable government-led analysis of deaths near surveillance towers, raising questions about whether agencies have adequately evaluated the systems’ humanitarian outcomes.
Billions of Dollars in Border Technology Spending
The findings have broader implications because the United States continues to devote substantial resources to border surveillance infrastructure.
The investigation reported that the government estimated the lifetime cost of its surveillance tower plans at approximately $6.2 billion in 2023. It also reported plans to spend another $1 billion on 1,497 additional towers by 2034.
These investments reflect the government’s interest in using technology to improve coverage across remote sections of the border. Supporters argue that automated systems can help agents identify activity more efficiently, improve situational awareness, and reduce the need for personnel to continuously watch camera feeds.
Nevertheless, large infrastructure budgets do not guarantee effective outcomes. Public agencies must demonstrate that the equipment they purchase works as intended, that operating costs are justified, and that failures are identified and corrected.
The investigation’s findings therefore raise questions about how surveillance programs are evaluated. Assessments that focus primarily on detecting border crossings may not fully measure whether the systems are functioning reliably, how often equipment fails, or whether alerts result in timely responses.
A more comprehensive evaluation would consider detection accuracy, equipment availability, response times, false alarms, and documented cases in which surveillance helped prevent harm. It would also examine incidents in which people died within monitored areas and determine whether changes to technology, staffing, or operational procedures could have improved the outcome.
Accountability and Government Oversight
The investigation also highlighted gaps in accountability. According to the reporting, former agents and officials said that the discovery of a body did not routinely trigger a formal review of whether surveillance equipment should have detected the person or why agents had not reached them sooner.
Without consistent incident reviews, agencies may struggle to identify recurring problems and determine which corrective measures are necessary.
The US Government Accountability Office has separately called for improvements in how Border Patrol evaluates its Missing Migrant Program. In an April 2025 report, the GAO recommended that Customs and Border Protection develop an evaluation framework and collect information clearly connected to the program’s objectives. The recommendations were intended to improve transparency and assess progress toward reducing migrant deaths and identifying remains. <Cite refs={[“turn319691search2”]}/>
Although the GAO report addresses the Missing Migrant Program rather than the specific surveillance-tower findings, it reinforces the importance of systematic evaluation and reliable data when government agencies are responsible for preventing deaths and responding to emergencies.
Greater transparency could help establish whether failures arise from equipment limitations, staffing shortages, communication problems, or deficiencies in operational policy. It could also help policymakers compare the performance of different surveillance technologies before approving further spending.
The Limits of AI in Border Security
The investigation highlights a wider issue confronting governments that increasingly rely on artificial intelligence for public safety and security.
AI systems can process visual information, identify patterns, and direct human attention toward potentially significant events. These capabilities can be valuable across large areas where continuous manual monitoring would be difficult.
However, identifying a person is not the same as understanding their circumstances. A system that detects movement may not reliably recognize whether someone is injured, lost, dehydrated, or in immediate danger. Even when software correctly identifies an event, an effective response depends on communication, staffing, access to the location, and the availability of emergency resources.
This means that technological performance should not be measured solely by how many cameras are installed or how many alerts are generated. Agencies must also evaluate whether the entire system achieves its intended operational and humanitarian objectives.
For border surveillance, this includes examining whether technology supports timely interventions, whether emergency alerts reach the appropriate personnel, and whether failures are investigated transparently.
Conclusion
The investigation into the US virtual border wall raises serious questions about the effectiveness, accountability, and humanitarian consequences of technology-driven border surveillance.
The finding that more than 1,050 people died within the advertised range of surveillance towers between 2015 and early 2026 does not prove that every death was preventable. It does, however, demonstrate the need to investigate whether surveillance systems detected people in distress, whether authorities received relevant information, and whether timely assistance was possible.
As the federal government continues investing in automated towers, cameras, sensors, and AI-powered monitoring, reliable performance assessments and transparent incident reviews will become increasingly important.
Ultimately, the effectiveness of a virtual border wall cannot be judged by its technological sophistication alone. Its success must also be measured by whether it performs its stated functions, whether personnel can respond effectively to the information it provides, and whether avoidable failures are identified and addressed.
Without meaningful oversight and evidence of real-world effectiveness, expanding surveillance infrastructure risks increasing public expenditure without adequately resolving the operational and humanitarian problems it was intended to address.
