The surveillance vendors you have never heard of

The contract is a year long and it costs $225,000. That is roughly what a mid-size police department spends on patrol cars in a quarter. For that sum, according to a procurement record dated February 2026, U.S. Customs and Border Protection bought access to a database of more than 60 billion facial images scraped from the public internet, and the right to run them against the faces of people it picks up at the border. The vendor is Clearview AI. The use is described, in the agency's own language, as "tactical targeting" and "counter-network analysis." The document does not say whether the faces of U.S. citizens will be searched, what agents are allowed to upload, or how long a result is kept. It says, in effect, that the search will happen and the rules will be worked out later.

This is how the American surveillance market works now. Not as a single all-seeing system but as a chain of vendors, each selling one capability — a face, a license plate, a keystroke — to a buyer who assembles them into something larger than any one contract admits. The companies are not household names. The buyers are police departments, federal agencies and, increasingly, your employer. The paper trail is public, if you know which filing cabinet to open.

The face: a database that keeps growing through every lawsuit

Clearview AI built its product by scraping photographs from social media and the open web without consent. That practice has cost it. In March 2025, a federal judge in the Northern District of Illinois approved a settlement in the consolidated biometric-privacy litigation against the company — an unusual deal that hands the plaintiff class a roughly 23 percent equity stake in Clearview rather than cash the company did not have. A bipartisan group of state attorneys general objected. The settlement stood. The database kept growing.

It kept selling, too. The February 2026 CBP agreement extends face-search access to Border Patrol's headquarters intelligence division and to the National Targeting Center. The company has separately reported contracting with the FBI and, by its own count, with thousands of state and local police departments. The marketing language is about catching fugitives. The documented outcomes include people who were not fugitives at all.

In March 2026, the police department in Fargo, North Dakota arrested a Tennessee woman on bank-fraud charges and had her extradited across the country. The match that led to the arrest came from facial-recognition software; the woman had never set foot in North Dakota, and records placed her shopping in Tennessee at the time of the crime. She is not alone. By public count, at least eight people in the United States have been wrongfully arrested on the strength of a face-recognition false positive. The error rate is not the scandal. The scandal is that a 60-billion-image dossier assembled without anyone's permission is now treated as probable cause.

The error rate is not the scandal. The scandal is that a dossier assembled without anyone's permission is now treated as probable cause.

The plate reader: 80,000 cameras and a single audit log

The most widely deployed surveillance hardware in the country right now is not a face scanner. It is the license-plate reader, and the dominant vendor is Flock Safety, a company that operates more than 80,000 cameras nationwide feeding a single searchable database. Police can query it with minimal justification. The competing system, Vigilant Solutions, is owned by Motorola Solutions. Both are bought by city councils, often without a public hearing, and the cameras turn up in places that have nothing to do with municipal policing — including the parking lots of Home Depot and Lowe's.

Flock's public position is unambiguous: the company says it does not work with U.S. Immigration and Customs Enforcement. The audit logs say something more complicated. Reviews by journalists, researchers and local officials found local police running searches that list "ICE" or "immigration" as the stated reason — queries performed, in effect, on behalf of a federal agency that has no contract of its own. The outlet 404 Media documented more than 4,000 such lookups conducted at federal request, and a separate search by Texas deputies for the car of a woman who had had an abortion. The architecture lets a department with a Flock account become a search terminal for anyone it chooses to help.

That architecture is now in court. A class action filed in California in 2026 alleges Flock illegally shared the state's driver-location data with federal agencies including ICE, CBP, the FBI and ATF. According to the complaint, cameras operated by the San Francisco Police Department were searched by out-of-state agencies more than 1.6 million times across a seven-month window. Cities have started pulling out: in February 2026, NPR reported a wave of cancelled Flock contracts over exactly these data-sharing concerns. Flock's response was a software setting — a January 2026 toggle that lets an agency switch off all federal sharing at once. The default, until an administrator finds the switch, is to share.

The intermediary problem

The recurring move in every one of these arrangements is the intermediary — the local agency, the reseller, the "real-time crime center" that aggregates feeds so the ultimate buyer never signs a contract anyone can FOIA. By 2026, roughly 300 U.S. law-enforcement agencies operate a centralized technology hub of this kind, pulling cameras, plate readers, face search and analytics into one room. The systems are sold by Axon, Motorola, Genetec and Flock, among others. The proprietary algorithms inside them are shielded as trade secrets. The contracts are negotiated outside public view. Oversight boards that try to look in are told the workings are confidential.

The fragmentation is not a side effect. It is the design. ICE does not need to buy Clearview directly if Border Patrol already has it and shares. ICE does not need a Flock account if a sheriff's deputy will run the plate. Each handoff is individually defensible — a local agency exercising local discretion, a vendor honoring a lawful request — and the sum is a federal surveillance capability that no single document authorizes. When you cannot find the buyer in the contract, that is frequently because the buyer was never on it.

The same tools, pointed at workers

The market does not stop at the police station. The fastest-growing branch of this economy points the same techniques — behavioral analytics, biometric monitoring, automated flagging — at people who are simply at work. Industry analysts put the employee-monitoring software market at $587 million in 2024 and project it past $1.4 billion by 2031. By the available surveys, roughly 78 percent of employers now run some form of monitoring, and 61 percent use AI-driven analytics to score performance — the keystroke logs, screenshots, and "unusual behavior" flags that the trade press has taken to calling bossware.

What that monitoring is for is the open question regulators have begun to ask. In February 2026, the Electronic Privacy Information Center and the National Institute for Workers' Rights launched a joint project to investigate how employers use monitoring tools to detect and disrupt worker organizing — surveillance aimed not at productivity but at the legally protected act of forming a union. State legislatures are circling the same conduct: California's proposed No Robot Bosses Act would require a human to review automated discipline decisions, and a Massachusetts bill would bar certain biometric monitoring outright and demand 30 days' notice before an algorithm's flag can become a firing.

The numbers from inside monitored workplaces are not subtle. In the surveys cited by labor researchers, 72 percent of monitored workers say the tracking does not improve their output, 59 percent say it erodes trust, and 42 percent of monitored employees plan to leave within a year — against 23 percent of their unmonitored peers. The technology is sold as efficiency. The measured result is attrition.

What the record supports

Here is what the documents establish, and only that. A federal agency bought face-search access to a database built without consent, and wrote the rules loosely. A plate-reader network with 80,000 cameras became, through local intermediaries, a search tool for agencies that never signed for it — and is now defending that in a class action over 1.6 million queries. A worker-monitoring industry heading toward $1.4 billion has begun to be used against organizing, which is why two regulators opened a file. None of this required a new law. All of it ran on existing contracts, default settings and the safe assumption that no one would read them together.

The fix is not technological, because the technology is doing exactly what it was sold to do. It is procedural. It is the public hearing before the city signs. It is the warrant requirement before the plate is searched, the audit log that names the requester, the default that shares nothing until someone with authority decides otherwise. The surveillance economy depends on fragmentation — on the belief that the data is too scattered and the entities too obscure to assemble into a picture. The contracts are public. Someone has to read them together. Until then, the market will keep selling one capability at a time, and the buyer will keep staying off the page.