Trump's AI Towers Can See Into Your Neighborhood
Trump's AI Towers Can See Into Your Neighborhood
Daniel Rao | October 24, 2026
The wall was supposed to run along the border. Instead, according to a New York Times investigation published on October 24, 2026, it watches you walk out of your front door. The administration's AI surveillance towers can see people leaving their homes, children on playgrounds, and, the Times found, inside a bedroom window. More than a million people in the United States and Mexico live within potential view. Nobody in that million gets to say no.
That last fact changes the story. Border surveillance has always been sold as something that happens out there, in remote desert, to other people. The Times investigation moves it inland, into streets where Americans buy groceries and walk dogs. The cameras, residents told the paper, are "always watching". This is no longer a debate about a line on a map. It is a debate about living under a machine that never blinks.
"It follows you"
Eloisa Cavazos lives in Mission, Texas, in a house across from a tower. She did not choose the view. "It can see when I leave, who I'm with, who's coming and going from our house," she told the Times. "It follows you."
The tower does not merely record a crossing. As Cavazos describes it, it assembles a pattern of a life: departures, companions, visitors, routines. That is the difference between a camera and a dossier. A camera sees an event. A pattern sees a person.
To show what that means, Times reporters used drone footage to approximate what the towers can see: children on an elementary school playground, people entering and leaving homes, vehicles moving through town. None of those people were crossing a border. They were living.
Nor does stepping out of one tower's sightline end the matter. The system tracks people as they pass between overlapping coverage areas, so leaving one camera's range does not mean leaving the system's view. The towers hand you off to each other, like relay runners with no finish line. A security tool starts to feel like something else: not a fence, but a net.
What October 24 is — and what it is not
Some necessary precision first, because this story is already being garbled online. October 24 is the date of the Times investigation, not the date 1,500 towers were switched on. No fleet appeared overnight. What the administration has secured, the Times reported, is more than $1 billion to build 1,500 new AI towers as part of a multi-year programme. Some aggregators have flattened that into "the government has deployed 1,500 new towers". That is wrong, and it matters: a build-out you can still scrutinise is a different political fact from one presented as finished.
The programme is documented in a June 2026 report from the Government Accountability Office (GAO), the independent auditor that reports to Congress. The GAO report, as summarised by the Electronic Frontier Foundation (EFF), describes the Integrated Surveillance Towers programme as currently comprising 803 fielded towers, with a plan to grow from around 830 towers today to 2,300 by March 2034. MIT Technology Review's September investigation gives a closely matching figure: CBP plans to spend $1 billion for 1,497 more towers by 2034. The small differences reflect different tallies and rounding. All describe the same thing: a near-tripling over eight years.
We cross-checked those numbers against each other before writing this article, and they line up. The build-out is real and funded, but not complete — which is why the privacy argument is live now, while towers are being sited.
How the towers watch
The modern towers combine radar, optical cameras, and thermal imaging, and the AI layer sets them apart. Anduril Industries, founded by Oculus creator Palmer Luckey, builds most of them. Its Sentry system — a 33-foot solar-powered tower, with an 80-foot extended-range version advertised as able to detect objects up to 7.5 miles away — uses computer vision to classify what it sees and alert agents. The software behind it, Lattice, fuses sensor feeds into one picture. Anduril's tower contract is now worth up to $1.1 billion, according to MIT Technology Review.
Anduril is no longer alone. General Dynamics, the defence giant behind the older Remote Video Surveillance System, has released AI towers of its own and received a CBP order in June worth up to $115 million. The EFF, which has tracked the towers since 2022, says some have already been spotted at the border.
The programme also folds together four legacy systems: Anduril's autonomous towers, Elbit's fixed towers, and General Dynamics' remote video systems. In 2023, the government estimated its tower plans, then numbering 803, would cost $6.2 billion over their lifespan. The new $1 billion sits on top of one of the most expensive surveillance projects in American history — and one never properly graded on whether it works.
There is an irony in the timing. By the government's own figures, the border is far quieter than it was: MIT Technology Review cites Department of Homeland Security figures showing just 128,009 enforcement encounters from January to August 2026, against more than 2.4 million in the same period in 2024. Encounters count Border Patrol interactions, not individuals, but the direction is unmistakable. Crossings have collapsed while the camera network triples.
The official answer — and the opt-out that isn't one
Customs and Border Protection says the technology is "essential to border security" and that it respects privacy and civil liberties. That is the agency's position, and it deserves recording rather than caricature.
But the Times pointed out the gap: people cannot opt out. Anduril disputed the analysis, saying customers can block designated areas from monitoring. CBP did not explain how often it uses that safeguard. A privacy protection nobody will say is switched on is, for the people underneath it, not a protection at all.
This is where the politics gets uncomfortable for everyone. "We don't want to be surveilled by our government," former Laredo councilman Gene Belmares, a Republican, told the Times. Surveillance that follows citizens home does not sort neatly into red and blue. The border towns living with it hold ranchers, teachers, retirees, and Border Patrol families — people who disagree about almost everything else and agree that they did not sign up to be watched eating breakfast.
Dave Maass, investigations director at the Electronic Frontier Foundation, called the programme a "costly boondoggle". "Every few years, they make these amazing claims about the technology," he told the Times. "And then an oversight body looks at it and finds it's a big, wasteful program that is not delivering results." The EFF has mapped these towers since 2022, and its July 2026 analysis called the expansion "a digital dumpster fire for human rights and civil liberties".
A separate investigation, a separate warning
Keep two investigations apart — they are already being merged online. The Times investigation is about neighbourhood privacy. MIT Technology Review's September 2026 investigation asked a different question: when people die near the towers, did the system see them in time?
MIT Technology Review's answer was grim: between 500 and 700 people have died since 2015 near at least one tower, with more than 300 of those deaths in rural areas. Nearly 350 died within range of Elbit's fixed towers alone. Its method was deliberately conservative: only deaths with GPS coordinates or precise locations, from more than 1,500 cases obtained from 14 Texas counties.
The two investigations do not corroborate each other, and should not be quoted as if they do. One is about the living being watched; the other is about the dying being missed. Side by side, they frame an awkward question neither CBP nor the tower companies has answered well: a system powerful enough to follow a resident across her neighbourhood, yet apparently unable to reliably spot people in distress in the desert it was built to watch.
CBP has a reply. Hilton Beckham, CBP's assistant commissioner for public affairs, told MIT Technology Review that CBP judges the towers on detection, response coordination, agent safety, and mission outcomes. Anduril added that once delivered, a tower is operated by CBP, and an incident nearby does not mean it missed a detection. They also cut the other way: if boundaries limit what towers see, the privacy question turns on where those boundaries are drawn.
Why neighborhoods are the story
Step back and the shift is clear. The first wall debate was about concrete: how high, how long, who pays. The second is about software: how far it sees, what it remembers, who watches the watchers. Concrete stops at the border. Software does not, because a camera does not know where the border ends and a cul-de-sac begins.
That geometry is why the EFF's July analysis warned that remote video systems sit "very close to the border fence in Arizona, including residential neighbourhoods where cameras are capable of spying on homes on both sides of the border". The border is not a desert abstraction. It runs past towns, schools, farms, and ordinary American streets. A tower sited to watch a crossing will, as a matter of physics, also watch the houses behind it.
The funding trail explains why the problem will get bigger. The EFF notes that the $1 billion comes from the One Big Beautiful Act, the tax and spending law President Trump signed in 2025. Money appropriated for a named purpose tends to get spent on it, and towers, once installed, tend to stay installed. The eight years to 2034 are not a distant horizon. They are a procurement schedule, and procurement schedules become facts on the ground.
There is a wider pattern here. This administration has moved aggressively on AI across the board — from the White House's AI policy build-out, tracked in Trump's AI policy, explained, to the Department of Homeland Security's widening technology controls, including its recent purge of DeepSeek from federal systems. The towers are not an outlier in that programme; they are its physical expression. AI, in this administration's hands, is not only a chatbot. It is a pole in a field, watching a street.
Update, Oct. 24: the San Clemente precedent — a city that asked for a ban and got 'avoid' instead
If the neighbourhood argument in this article sounds abstract, San Clemente, California, makes it concrete — and the post's first version missed it. In January, the San Clemente City Council voted 3-1, with one abstention, to let Customs and Border Protection install a surveillance tower on city-owned land at the Avenida Salvador Reservoir: a $1 million system of radar, cameras, sensors, and communications equipment, meant to watch the coastline for suspected human smuggling. The city asked for what residents actually wanted — contract language prohibiting surveillance of residential neighbourhoods. CBP refused it. The agency's answer, recorded in the city staff report, was that the system would be configured to "avoid" scanning residential areas, but that if an active smuggling event detected at sea continued onto land, "the system could continue to monitor the event" — and that restricting that capability "would be contrary to the spirit and intent of the deployment." So the safeguard is a configuration preference, not a prohibition, and it yields exactly when it matters most. Then there is retention. The Electronic Frontier Foundation's account of the same documents finds a 30-day imagery retention period contradicted in the same paperwork by the line that training data "should not be deleted" — the footage, in other words, doubles as fuel for Anduril's product. Oakland Privacy, which helped organise local opposition, put the wider point bluntly: a tower on an Orange County coastline 75 miles from the border shows "the militarization of the border region is rapidly moving northwards and across the entire state." San Clemente is not a border town in the Mission, Texas, sense — it is a coastal city of ordinary streets — and that is precisely why it belongs in this story: it is the proof that "neighbourhoods are the story" is not a prediction. It already happened, in a council vote, in January. (Council vote, $1 million system, rejected ban language, and "avoid" wording per the San Clemente staff report via the EFF, April 2026, and a San Clemente-area councilmember's account; retention contradiction and Oakland Privacy quote per the EFF.)
Frequently asked questions
Can residents opt out of tower surveillance?
No. Customs and Border Protection says the technology is essential to border security and respects privacy and civil liberties, but the Times found no opt-out exists for the more than a million people within potential view. Anduril says customers can block designated areas from monitoring; CBP has not explained how often it uses that safeguard — and San Clemente's experience shows a city that asked for a contractual ban was offered "avoid" instead.
Are the 1,500 new towers already watching?
No — October 24 is the date of the Times investigation, not the date 1,500 towers were switched on. What the administration has secured is more than $1 billion to build 1,500 new AI towers as part of a multi-year programme, growing from around 830 towers today toward 2,300 by March 2034. The build-out is funded and real, but not complete — which is why the privacy argument is live now, while towers are still being sited.
Can a city refuse a surveillance tower?
San Clemente, California, voted 3-1 in January to let CBP install a tower on city land, but CBP refused contract language banning residential surveillance, offering only to "avoid" scanning homes — and to keep monitoring if a smuggling event moved onto land. The safeguard is a configuration preference, not a prohibition.
How many people live under this surveillance?
The Times counts more than a million people in the US and Mexico within potential view of the towers; the ACLU puts 15 million Southwest border residents under CBP's wider surveillance umbrella (ACLU figure via Reason, July 30, 2026). The two numbers measure different things — tower view vs the full CBP surveillance footprint — and both are cited here so readers do not confuse them.
Our view: a border tool must not become a neighborhood habit
Here this article stops reporting and starts judging, because some stories do not permit neutrality without becoming evasive. A government that can watch citizens in their neighbourhoods, track them between cameras, and offer no opt-out has crossed a line that should trouble people regardless of how they vote on immigration.
The test is simple, and CBP could pass it tomorrow. Publish, tower by tower, which areas are masked and which are not, how often the masking safeguard is used, and the effectiveness data — detections, rescues, deaths nearby — audited by someone other than the vendor. A system that cannot survive those disclosures has a legitimacy problem, not a communications problem.
Until then, the burden of proof sits with the watchers, not the watched. Cavazos did not ask to be a test case for the smart wall. Belmares did not ask his city to become a demonstration site. The million-plus people within potential view never consented, because consent was never on offer.
What happens next will be decided in unglamorous places: GAO follow-up reports, oversight hearings, privacy assessments, and possibly courtrooms. Texas has shown it will take AI companies to court, as in its lawsuit against Sam Altman. Watch for whether any member of Congress asks the simplest question: if the towers are essential, show us the masked areas.
The smart wall was promised as a cheaper, cleverer alternative to concrete. What the Times found is cheaper only in the sense that concrete at least stayed where it was put. Cameras wander. They follow you home, hand you to the next tower, and never tire. That is a remarkable capability — and, for the people underneath it, a remarkable imposition. Impositions, in a democracy, are supposed to come with an argument, an expiry date, and a way out. This one comes with none of the three.
Sources: A New York Times investigation published October 24, 2026, as relayed by DNYUZ/Raw Story; MIT Technology Review, "The US spent billions on border surveillance. Why can't it catch people before they die?" (September 21, 2026); the Government Accountability Office's June 2026 report on the Integrated Surveillance Towers programme, as summarised by the Electronic Frontier Foundation (July 2026) and relayed by Activist Post; Department of Homeland Security enforcement-encounter figures via MIT Technology Review. Quotations from the Times investigation are as relayed by DNYUZ/Raw Story; the Times article itself is paywalled and was not read directly.
Daniel Rao covers technology and politics for TechPulse USA, with a focus on AI policy, surveillance, and the companies reshaping American infrastructure.
Comments
Post a Comment