They published the number. Almost three thousand. I printed all 653 pages of the disclosure table. Carol asked why. I said: evidence.
The Washington Post published something this week that should have been the biggest story in the country. Twenty-nine federal agencies disclosed 2,987 active government AI surveillance programs by the end of 2025. Up from 1,684 the year before. Nearly doubled. In one year. And the article ran on a Sunday morning like it was a recipe for banana bread.
When I was running networks at MedStar Regional, I could tell you what every one of 340 employees was doing on their work machine at any given moment. Imagine that capability multiplied by the entire federal government, powered by artificial intelligence that can read your handwriting.
They Put It In Writing
That’s the part that got me. A defense contractor called Palantir built a system for ICE called ELITE—Enhanced Leads Identification and Targeting for Enforcement—that uses AI to read handwritten documents. Rap sheets. Warrants. Handwriting. The stuff you put on paper specifically because you thought it couldn’t be digitized. Nothing is safe anymore. I’ve been writing my notes in a personal shorthand since 2016 for exactly this reason. Carol says it looks like “angry hieroglyphics.” She’ll understand eventually.
And then there’s the facial recognition. DHS deployed something called “Mobile Fortify”—an app that lets agents scan your face in the field. Not at a checkpoint. Not at an airport. In the field. Meaning anywhere. I told you about AI tracking back in January. The shopping chatbots were phase one. This is phase two. Your face is now searchable by an agent with a phone app.
The Part Nobody Is Talking About
Here’s what froze me. The Pentagon—which is exempt from even disclosing its AI programs—put out a request for “agentic” AI with “decision-making capabilities” and “human-like agency” for Special Operations forces. They want AI that weighs rules about when to start combat and assesses civilian casualty risk. These tools should “adapt and learn in real time” but would be prohibited from learning during “kinetic fires”—live ammunition—because it “may lead to undesired behavior.”
They’re building AI that learns, but they’ll turn off the learning when the shooting starts. That’s their safeguard. A software toggle. I spent 22 years in IT. You know what the off switch was for most of my career? Pulling the plug out of the wall.
What I’ve Done About Government AI Surveillance
My setup has been hardened against this for years. Flip phone. Cash. Tape on the webcam. No smart devices. I print everything important because digital files can be altered. I have 14 binders of printed evidence in my office. Carol asked last week if I “needed all the binders.” I told her the binders are the only thing between us and total digital exposure. She made a face and went back to her iPad.
Her iPad. Connected to WiFi. Signed into her Apple account. Location services on. She says she “only uses it for recipes and Wordle.” I’ve explained the problem eleven times.
The article quoted a Brown University professor saying the government removed guardrails. The Defense Secretary wrote that they should approach AI “as if we were at war.” They’re approaching it like a man who bought a table saw and skipped the safety manual because he’s in a hurry.
Meanwhile, I’m here with my flip phone and my binders and my shorthand notes, and 2,987 government AI surveillance programs are humming along in federal buildings across the country. My daughter says I “sound like a man yelling at the ocean.” Maybe. But the ocean just published its own inventory, and it’s bigger than anyone imagined.
About the Author
