Each death should have alerted U.S. Customs and Border Protection (CBP) to flaws in the agency’s surveillance network. That is especially true in the case near the latest tower built by defense company Anduril. Anduril continues to win large government contracts and is positioned to benefit from the federal funding planned for border security. The United States plans to spend $1 billion to triple the size of its virtual wall by 2034.

Here are four steps CBP should take to address these failures in AI border surveillance.

1. Conduct a comprehensive audit of deaths near virtual walls

“If I were the federal government, I would do an audit because I feel like this is not working,” New Mexico Congresswoman Sara Silva said.

MIT Technology Review found that more than 18 people died near watchtowers in 2024 alone in a strip of desert near Silva’s district.

An audit is necessary because, despite a review of thousands of pages of police reports and other records, there is limited public information about how many watchtowers exist and where they are located.

CBP does not publish a comprehensive map of the virtual wall. Instead, researchers relied on a mapping effort led by the nonprofit Electronic Frontier Foundation, which used satellite imagery to identify specific towers and estimate when they were erected.

Our analysis identified nearly 600 towers—far fewer than the 800 towers the government says it currently operates. Estimates of when individual towers became operational are also imprecise. As a result, our tally of more than 1,050 people who died near the towers is an undercount, and there is no complete picture of where along the border the problem is most severe.

An internal CBP audit would not have those limitations. In theory, the agency should track when each tower was installed and activated. Since at least 2015, it has had the ability to assess the system’s effectiveness while quantifying the number of people who died nearby. However, officials say no such audit has ever been conducted. These findings reinforce the need for one.

2. Fix how authorities track migrant deaths and recovered bodies

When CBP analyzes deaths near its surveillance technology, it needs accurate information about when and where those deaths occurred. The current system for tracking that data is deeply flawed.

As detailed in our methodology, no national agency collects and publishes all available records on migrant deaths. Although we reviewed thousands of pages of records from dozens of government agencies in Texas alone, the lack of a common process for recording deaths in public records leaves the overall picture incomplete.

Since 2007, the Border Patrol has operated an internal system called the Border Security Initiative Tracking System, or BSITS. National rules for how different departments should enter cases into the system were not established until 2021. The following year, the Government Accountability Office (GAO) found that many cases—including deaths that went undetected by Border Patrol—were still missing from the system. In 2023, GAO reported that the Border Patrol had taken steps to address the problem, but outside researchers continue to say the agency’s estimates are too low.

CBP also launched the Missing Migrant Program in 2017 to help prevent migrant deaths. In April 2025, the GAO found that the program was improving how it collected and tracked data about 911 calls, rescues, and related events. However, the agency had no plan to use that data to determine whether the program was increasing rescues or reducing deaths—its primary goals. We provided two recommendations to address the problem, but neither has been implemented.

3. Track when surveillance technology leads to encounters and arrests

In 2014, when virtual walls were much smaller, the GAO recommended that Border Patrol agents record when technology helped them identify and apprehend someone attempting to enter the United States without authorization.

Although documenting these encounters would take agents additional time, the data would help Congress understand how often different surveillance technologies lead to actual arrests. Once the information was collected, agencies would also need to measure whether the technology was effective. CBP agreed and began requiring agents to enter this information.

Glenn Fawcett/CBP Photography

In 2017, the GAO reported that the data being collected was inaccurate and unreliable. For example, investigators in Texas flagged a particular type of tower as unsafe hundreds of times, even though no such tower existed anywhere in the state. The Border Patrol said it would improve the process and train staff on the importance of accurate data collection.

Several years later, however, the agency told us that measuring the technology’s effectiveness was not yet a priority. One former CBP official described the effort as “like spitting in the wind, to be honest,” saying that the messy reality of border security does not easily translate into reliable metrics. The official, who requested anonymity, worked during the Biden administration but was not authorized to discuss sensitive issues.

In August 2021—more than seven years after the agreement—CBP still had not conducted an analysis of the towers’ effectiveness. The GAO completed an analysis in 2022, but its results have not been published.

This means the virtual wall is expanding faster than officials are measuring its impact.

4. Investigate deaths as potential surveillance failures

CBP should investigate new fatalities as they occur and determine whether they happened within view of towers or other surveillance technology. Authorities also need to establish why people were not found sooner.

Multiple officials told us that when Border Patrol agents find the body of someone who appears to have died while attempting to enter the United States, they are not required to determine whether the person died near surveillance infrastructure. There is also no established process for doing so.

At a minimum, this means potentially important evidence about the death may never be examined. It also means migrant deaths are not routinely treated as possible failures of the virtual wall.

CBP should automatically compare records of all migrant deaths with the internal databases of nearby towers. When a death occurs within surveillance range, the agency should investigate what the system detected and whether agents responded appropriately. Agents already record GPS coordinates when they capture a group or encounter footprints.

Without this comparison, law enforcement, medical investigators, and family members may never know that someone died near a watchtower—even if footage of the moments leading up to the death exists. CBP typically deletes tower surveillance footage after 30 days unless an internal review is underway, which usually happens only in cases involving an in-custody death. Flagging these incidents could preserve footage for investigators and give families an opportunity to request it.

These post-mortem reviews could also reveal flaws in the virtual wall itself. Consider one location in New Mexico that we identified. Several people died there even though advanced AI towers were clearly visible. The deaths occurred over a period of months, sometimes only a few hundred feet apart. Investigations could reveal that cameras repeatedly failed to detect or identify people in certain areas, or that agents failed to respond to alerts.

Such reviews are the only way to answer the question most often raised by families, researchers, and volunteers in our investigation of the hundreds of migrants who died near the towers: “Did the cameras see them, or did no one come?”