Surveillance AI just became your problem, not just legal's

For a while, the ethics of surveillance tech lived in someone else's inbox. Legal handled it. Policy handled it. You built features. That gap is closing fast. The systems making news right now are the same kinds of systems your team touches, and the public is watching who builds them.

Let me catch you up on what shifted this month and what it means for the calls you own.

The build-out is real, and the money is committed

This is not a pilot anymore. The Trump administration plans to grow the border surveillance network to 2,300 AI-powered towers by 2034, a roughly $1 billion expansion up from around 830 towers today. The 2027 budget alone steers $97 million toward 95 new towers.

Private money is moving the same direction. Anduril raised $5 billion in May at a $61 billion valuation and is building autonomous ships, drones, and AI software for the military. Flock Safety hit a $7.5 billion valuation selling cameras and license-plate readers to police. The capital says this category is a growth market, not a fringe.

The backlash is organized, and it costs deals

Here is the part that lands on your roadmap. Public pushback is no longer just angry posts. It shows up in contracts. Amazon's Ring ended its partnership with Flock, and the LAPD let its Flock deal lapse. People have gone as far as destroying Flock's cameras in the street.

Anduril is drawing protests at its Seattle office, with organizers calling the company a threat to migrant and working-class communities. The company answered that it finds protest against a firm supporting the military "perplexing." Whatever you think of that exchange, note the pattern: the position a company takes in public now shapes whether partners stay and cities renew.

The tools you already ship can be turned into weapons

You do not need to build a border tower to be exposed. Ordinary tech gets repurposed. Iran abused SS7, a decades-old flaw in how cell networks route calls, to locate U.S. troops in the Middle East and strike them. It also used ordinary ad-targeting tech to track phones. The surveillance came from everyday plumbing, not spy gear.

Data integrity is the same story. A single weather station at Paris CDG airport got tampered with, earning one gambler $20,000. The bigger risk is coordinated nudges across many inputs, small enough to look real, feeding AI models that then act on their own. When you strip humans out of the loop for speed, you also strip out the person who catches the fake.

Quantifying people has a cost you can't measure

Historian Jill Lepore makes the argument underneath all of this. In her new book, she says turning human concerns into data flattens them, and that private systems keep taking over jobs the public used to do. Her example: political campaigns swapped door-knocking for data mining, and lost the neighborhood conversations that held civic life together.

That framing helps you at work. Every time your team replaces a human judgment with an automated score, you gain speed and lose a check. Sometimes that trade is fine. Sometimes it removes the only person who would have said "that reading looks wrong" or "that match is a person's freedom." Name the trade out loud when you make it.

The deep cut

The thing easy to miss: your public position is now a product input, not a PR afterthought. Flock lost Ring and the LAPD. Anduril's stance shapes who protests and who partners. Your team will get asked, by recruits, by customers, by press, what you build and who you sell it to. If you have no answer, silence becomes your answer, and it will read as the worst-case version.

So decide before you are cornered. Write down which uses you will build for, which data you will collect, which buyers you will take, and where a human stays in the loop no matter the speed cost. Put it where your team can see it. A stated position you can defend beats a scramble after a story breaks.

Three questions for your team

1. If a reporter asked tomorrow who can buy our product and what they can do with it, do we have a written answer we would stand behind?
2. Where have we already pulled a human out of the loop for speed, and what fake or abuse would that human have caught?
3. Which parts of our stack could be repurposed for tracking or targeting the way SS7 and ad tech were, and who owns closing those holes?