Clearview AI’s InquiryIQ Prototype Would Turn Face Searches Into Digital Profiles

Clearview AI’s InquiryIQ was described as an unreleased prototype designed to extend a facial-recognition lead into a much wider online investigation. The concept would search the web and images, analyze photographs, follow possible relationships and assemble a “Candidate Graph” of people and connections. It was not presented as a public product or a demonstrated police deployment.

That distinction matters. InquiryIQ’s proposed workflow could reduce the amount of time and effort needed to move from “Who is this person?” to “What else can be connected to them?” It could also make unreliable clues look more coherent than they really are.

The prototype Clearview says police have not used

Clearview AI CEO Amos Kyler said that no law-enforcement user had used InquiryIQ. Clearview also said the company did not plan to release the prototype in its present form.

The concept surfaced through interface code and materials delivered before authentication. Those materials described intended functions and warnings, not a live customer workflow or a public product launch. The result is a reported prototype with a clear design ambition—but no demonstrated record of police operation.

InquiryIQ is therefore best understood as a proposal for extending Clearview AI’s existing facial-recognition workflow, not as a tool that police departments can simply sign into today.

How InquiryIQ would expand a facial-recognition lead

The proposed sequence is straightforward, and that is precisely what makes it significant:

1. An investigator begins with information from a Clearview AI facial-recognition search.
2. The investigator adds details considered relevant, such as possible age, gender, race, hair color or eye color.
3. InquiryIQ searches webpages and images, browses pages and analyzes photographs it encounters.
4. Facial recognition can be applied to photographs found during that wider search.
5. The system organizes possible identities, associates and related facts into a Candidate Graph.
6. A human investigator reviews the surfaced information before accepting it into a profile.

The interface described demographic details as inputs that could help make “smarter decisions.” It did not establish how age, gender or race would affect the results. That is an important boundary: the presence of an input is not proof that it improves accuracy, or even a clear explanation of how the system uses it.

The interface also listed model-related options associated with xAI and Amazon Bedrock. Clearview said the selector was used for internal engineering comparisons rather than as a model choice offered to police users.

From a lead to a possible profile

The comparison describes a change in workflow, not a proven performance advantage. InquiryIQ was designed to connect more steps in one process; no accuracy, speed or real-world investigative result is established for the prototype.

What could end up in the profile

The proposed Candidate Graph could contain possible identities and associates, along with possible:

• addresses;
• phone numbers;
• employers;
• social-media accounts;
• arrest histories;
• aliases; and
• physical characteristics.

That list is not a collection of verified facts. It is a set of possible connections that the system might surface for human review. The interface warned that generated demographic, social-media and arrest information “may or may not be accurate.”

This is the classic generative-AI problem wearing a police badge: a system can produce a tidy-looking narrative from messy clues. A tidy profile is not automatically a true profile. InquiryIQ’s proposed safeguard was independent human verification before information entered the record.

The surveillance-friction problem

Andrew Guthrie Ferguson, a law professor specializing in AI and policing, warned that a system like this could create a profile from the scattered digital clues people leave online. The concern is not merely that the software might find more information. It is that it could make broad searches cheap enough to pursue more often.

Historically, investigative labor acted as a practical brake. Searching across websites, images, aliases, phone numbers and social accounts takes time. Automating those steps could lower that barrier and make broad “fishing expeditions” easier to launch.

Woodrow Hartzog, a privacy scholar, tied the concern to the assumptions behind existing privacy protections: many rules were developed in a world where governments faced practical limits on following everyone’s digital trail. When software reduces those limits, the old safeguards may no longer provide the same friction.

The risk is especially sharp when possible matches are treated as a chain of confirmation. One uncertain clue can lead to another, and the resulting profile can appear persuasive simply because it contains many connected details.

Human verification is a safeguard—and a point of dispute

Clearview’s design placed a person in the review process. Investigators could accept or reject surfaced details, and the interface required independent verification before accepting them.

That is a meaningful control, but it is not a guarantee of accuracy or due process. Hartzog described a human reviewer as “a little bit of a cold comfort” if the person eventually becomes a rubber stamp for automated output. Michael Price, litigation director of the National Association of Criminal Defense Lawyers’ Fourth Amendment Center, made a similar point by comparing reliance on a hallucination-prone chatbot with relying on an informant who would not otherwise be trusted.

The practical question is not simply whether a human clicks “approve.” It is whether the investigator checks each lead against independent facts, records why it was pursued and remains willing to reject a compelling-looking but unsupported connection.

The auditability argument

There is one possible benefit in the design. Ferguson argued that preserving prompts, searches, model choices and investigative paths could make an AI-assisted investigation easier to audit than an undocumented human search.

That benefit depends on what the system records and how investigators use those records. A searchable trail could show which assumptions shaped an inquiry and where a questionable connection entered the process. It could also make later review more concrete.

But an audit trail does not make an initial search lawful, accurate or proportionate by itself. It tells reviewers what happened; it does not automatically justify why it happened.

What InquiryIQ would change if it were released

The important shift is not simply “more AI.” It is the movement from facial recognition as a way to generate a lead toward software that could assemble a broader digital portrait around that lead.

Clearview says its broader facial-recognition technology is used by more than 2,000 law-enforcement agencies nationwide, and the company says its database contains well over 70 billion images. Those are figures about Clearview AI’s broader platform, not measurements of InquiryIQ or proof that the prototype was used by those agencies.

InquiryIQ’s significance lies in the workflow it sketches: identify a face, search outward, connect fragments and present a possible profile for review. That could make investigations faster to start and harder for ordinary people to escape once their scattered online traces are pulled into one place.

For now, the meaningful fact is the design ambition—not a demonstrated police capability. InquiryIQ points toward a future in which the hardest part of an investigation may no longer be finding one face, but deciding whether the many digital clues attached to it deserve to be believed.