Clearview built an AI that turns one face into a dossier

Clearview AI spent six years turning faces into names for the police. WIRED has now found the next step sitting in the company's own website code: an unreleased assistant called InquiryIQ, designed to take a face-recognition hit and go looking for everything else about the person by itself. Clearview says it is a prototype that no officer has ever used.

Found in the front door

Nothing was hacked. When you open Clearview's login page, your browser downloads the files it needs to draw the interface, and those files were more talkative than intended: code plus thousands of lines of interface text, including instructions, warnings and feature descriptions. WIRED had used the same approach a month earlier on Flock Safety, reconstructing a mock-up of an unreleased search tool from publicly served files.

What the tool is built to do

1. An investigator runs an ordinary Clearview face search and marks the details they consider relevant.
2. Those details go into a profile alongside age, gender, race, hair colour and eye colour. The interface states these help the AI make smarter decisions when running its searches.
3. The assistant searches the web and image sources, opens pages, and runs Clearview's face recognition on photographs it encounters along the way.
4. It returns a Candidate Graph of possible identities and associates, and fills the profile with possible addresses, phone numbers, employers, social media accounts, arrest history and aliases.
5. The officer accepts or rejects each finding. Clearview's own warning says the generated demographic, social media and arrest data may or may not be accurate, and the officer must attest to verifying it independently.

In the prototype WIRED reviewed, the interface included a control for choosing which model does the reasoning, listing xAI and Amazon Bedrock. Chief executive Amos Kyler says that selector existed so Clearview's engineers could compare models during testing, not so police could pick one. Amazon told WIRED that AWS is not involved in developing InquiryIQ.

Why researchers call it digital rummaging

Andrew Guthrie Ferguson, a law professor at George Washington University who studies AI and policing, uses that phrase for automated investigation of this kind. The point is not that any single fact is secret. It is that facts which used to sit in separate places, each requiring somebody to go and look, get pulled together into one profile at machine speed.

Woodrow Hartzog, a privacy scholar at Boston University, makes the sharper version of the argument: the effort involved in investigating a person was itself a form of protection, and nobody wrote rules against surveillance that was simply too laborious to do at scale.

The privacy protections we have in place right now were mainly built in a world that assumed a certain amount of friction in the ability of governments to collect information about people.

Woodrow Hartzog, Boston University, to WIRED

The practical consequence is not that police solve more crimes, though they might. It is that looking into someone becomes cheap enough to do on a hunch. A search that would have cost a detective a week is worth running on a maybe.

The human in the loop

Clearview's answer to all of this is that a person reviews everything, and that the tool surfaces leads rather than conclusions. Hartzog's response is that a reviewer who is handed confident output all day eventually becomes a rubber stamp.

There is already a documented example of what that looks like, and it does not involve InquiryIQ at all. In a Minnesota case, United States v. Sant, involving undercover Homeland Security Investigations agents and surveillance of political activists, defence lawyers obtained a Clearview report built from roughly fifteen years of protest photography. Every result in it was stamped as accepted by one named user, including a match that Clearview itself had labelled as a less likely result, which can only be exported if somebody accepts it. The defence allege the report swept in photographs of an entirely different man, along with his pregnant wife and young daughter.

Michael Price, litigation director at the National Association of Criminal Defense Lawyers' Fourth Amendment Center, puts the reliability question in the terms a court would use, since this material can end up supporting probable cause.

A hallucination-prone chatbot would not be trusted as an informant under any other regular circumstances.

Michael Price, NACDL Fourth Amendment Center, to WIRED

This direction is not new

A Clearview patent filed in August 2020 and granted in February 2022 already described a crawler that collects facial images together with the personal information attached to them, then returns a name, address, telephone number, email and links to social, professional and employer pages once it finds a likely match. It listed face-based background checks among the applications, and contemplated spotting second faces in a photograph to establish a relationship between the people in it.

The company's own pitch to intelligence and defence customers puts it more plainly than any critic could: traditional collection starts from known selectors such as a phone number or an email address, and with Clearview's tools the selector is a person's face.

1. 2020The New York Times reveals Clearview scraped more than 3 billion images. Lawsuits and regulatory investigations follow; the scraping continues.
2. August 2020Clearview files the patent describing a crawler that ties faces to names, addresses and employer links. Granted February 2022.
3. February 2026CBP seeks 15 Clearview licences for Border Patrol intelligence staff, citing network development and strategic counter-network analysis.
4. 10 September 2026WIRED publishes InquiryIQ, found in Clearview's own publicly served login files.

Clearview is also not alone in automating this work. Intelligence platforms sold to law enforcement, including ShadowDragon's SocialNet, Penlink's Tangles and Fivecast, already help investigators surface aliases and associates and map a person's digital footprint.

One honest upside

Ferguson points out something the critics usually skip. Detectives already search databases, discard leads and follow hunches without recording most of it. A system that preserves its prompts, its searches, the model it used and the paths it took would leave a more auditable trail than a human investigator does. He calls it a silver lining, then adds that in sixteen years of watching police technology arrive he has seen the same playbook every time: deploy first, work out the governance later.

What it means if you are not under investigation

The uncomfortable part of this story is that none of the raw material is stolen. It is the photograph a friend tagged, the old forum profile, the company page listing your team, the local news item with your name in it. Individually harmless, collectively a dossier, and the only thing that ever kept them apart was that assembling them took a person several days.

A VPN is worth naming here only to rule it out. It changes the route your traffic takes and the address services see. It does nothing about material you or other people published years ago under your own name and face. The controls that actually bite are the boring ones: what you publish, what you let others tag, and how much of your face is on public pages in the first place.