Soviet constructivist propaganda poster: a watchtower camera lens scanning a grid of file-folder silhouettes, a cracked padlock below

A Legal Observer Got a Warning From Border Patrol. It Knew Her Name and Her Face. Her TSA PreCheck Was Gone Three Days Later.

Newly unsealed federal court records from Minnesota show Homeland Security Investigations used Clearview AI facial recognition, undercover agents, and Signal monitoring to build surveillance files on immigration protesters, legal observers, and union members who were never charged with anything.

The Filing Cabinet Opens

Newly unsealed federal court records out of Minnesota show what Homeland Security Investigations, the investigative wing of ICE, actually builds when it says it's "monitoring public safety" around immigration protests. The documents, tied to operations with names like Operation Metro Surge, Operation Puppet Master, and — one presumes an intern won a naming contest — Project Whipple Shield, describe undercover agents infiltrating community meetings, secretly recording conversations, and slipping into encrypted Signal chats. One agent did this under the username "peanutbutterjelly," which is either an elaborate operational security choice or someone's actual go-to alias for everything, including his bank login.

Running underneath all of it: Clearview AI, the facial recognition company whose reported 70-billion-image database was built by scraping the public internet without asking anyone's permission, least of all the people in the photos.

Fifteen defendants now face federal charges for interfering with immigration officers. Far more than fifteen people got swept into the files.

The Search Results Don't Stay in Their Lane

At least two Clearview AI reports were produced in the case of defendant Isaac Sant, cross-referencing facial scans against vehicle records, social media, and location data. In a detail that undercuts any claim this was surgical, one of those searches returned photographs of an entirely different man — along with Facebook pictures of his wife and child, who were, to be clear, not suspected of interfering with anyone.

Sources within the Civil Liberties Community note that this is the recurring failure mode of dragnet facial recognition: it doesn't just misidentify strangers, it drags their families in with them. And yet — as if this were not enough — the surveillance didn't stop at the defendants. Court records show it swept in legal observers, union members from the Service Employees International Union and the Communications Workers of America, and volunteers with groups like the Sunrise Movement and Whipple Watch, none of whom were charged with anything.

A wiretap doesn't know the difference between a suspect and a bystander. The Algorithm, in this case, didn't either.

The Body Camera Remembered Her Name

One legal observer told reporters that a Border Patrol agent addressed her by name in the field and mentioned "facial recognition and a body camera" while doing it. Three days later, her Global Entry and TSA PreCheck privileges were revoked. Homeland Security has offered no public explanation connecting the two events, and in a development that will surprise no one who has been paying attention, the timing has not been described by the agency as coincidental, non-coincidental, or anything at all.

This all arrives months after ICE signed a $3.75 million contract with Clearview AI — its largest facial-recognition purchase on record — the same season the agency deleted its own internal oversight policy on facial recognition from its public website. Authorities confirm the deletion. They have been considerably quieter about the reasoning.

The policy is gone. The cameras, notably, are not.

What Happens to a Legal Observer's Face

A legal observer's entire job is to stand at the edge of a protest, watch, and report what happens — a role built on the assumption that watching is a protected, unglamorous act of civic hygiene. Court records now suggest that in Minnesota, the watchers were also being watched, scanned, and cross-referenced against a database of 70 billion faces none of them consented to join.

Nobody has explained why a facial recognition system meant to identify suspects apparently has such a difficult time telling suspects apart from spouses, children, union organizers, and the people whose actual job is to hold a clipboard and take notes. Sources confirm the technology is very good at finding a face. What it does with that face afterward remains, as ever, somebody else's problem.

Sources: Biometric Update — Court records detail Clearview AI's role in ICE protest surveillance · Washington Post — The powerful tools in ICE's arsenal to track suspects — and protesters