Clearview AI is testing an internal tool called InquiryIQ that allows police to automatically search the web for details about a person they are investigating.
The company, which previously gained notoriety for scraping over 3 billion photos to identify faces, is now experimenting with using artificial intelligence to connect those identities to names, addresses, and associates.
According to code reviewed by WIRED, the prototype tool accepts input on age, gender, and race to guide its search logic. One of the AI models available for this task comes from SpaceXAI, the division behind the Grok chatbot developed by Elon Musk.
Musk has marketed Grok as an alternative to systems he claims are biased. The model has faced criticism for generating racist and extremist content. It is unclear how Clearview intends to use demographic data to influence the search, and the company declined to comment on the specifics.
Experts warn that such tools could compress weeks of detective work into minutes. This speed might help solve crimes faster but also lowers the barrier to scrutinizing individuals who might otherwise escape investigation.
Because generative AI can produce different results from the same starting point, it may become difficult to explain why a system followed one lead over another.
Clearview states that InquiryIQ is a prototype never pitched to customers and not planned for release in its current form. The company disputes the idea that the tool acts as an automated investigator.
CEO Amos Kyler told WIRED that the interface allows engineers to compare how different models perform during testing. He said no law enforcement user has ever used the tool.
WIRED found the interface in files sent to browsers when visiting Clearview’s login page. This is the same method used last month to uncover a similar tool called OS Investigate developed by Flock Safety.
The files contain thousands of lines of text describing the tool’s features, including instructions on how to “automatically discover and enrich personal data from web sources.” They do not reveal how well the tool works or who has tested it.
Clearview argues that modern AI makes it faster to build sophisticated prototypes. The company says its engineers are directed to move prototypes along rapidly.
Founded in 2017, Clearview spent its early years largely out of the public eye. Peter Thiel invested $200,000 in the first year. The company soon began offering free trials to police departments.
A 2020 HuffPost investigation detailed founder Hoan Ton-That’s ties to the far right, including a 2016 Republican National Convention dinner with white nationalist Richard Spencer. Ton-That later apologized for his past writings.
In 2020, The New York Times revealed that Clearview had scraped more than 3 billion images from Facebook, YouTube, Venmo, and millions of other websites. The report transformed the company into one of the country’s most controversial surveillance firms.
Tech companies demanded the harvesting stop. Lawsuits and regulatory investigations followed. Clearview continued scraping. Its database grew from over 3 billion images in 2020 to what the company now says is well over 70 billion.
The company states its technology is used by more than 2,000 law enforcement agencies nationwide.
Ton-That stepped down as CEO in December 2024 and left the board the following spring. Kyler, who became CEO in October last year, joined Clearview as an engineer in 2019.
Where the early years were defined by boundary-pushing, Kyler talks instead about controls, auditing, and oversight.
“The mission today is the same,” Kyler says. He describes the company’s recent focus as “refinement” and “ensuring that the product hits the kind of expectation of integrity that our customers expect.”
Signals and Noise
For years, Clearview said its job stopped at surfacing possible leads for law enforcement using face recognition. In a 2022 post, Ton-That wrote that it was “up to the investigator to follow those links and do more research to find additional information.”
InquiryIQ appears designed to take on some of that work. As Clearview describes it, the tool begins after an investigator has already run a face-recognition search and identified details they consider relevant.
Kyler says InquiryIQ was conceived as a way to test pieces of information an investigator had already identified as potentially relevant. “We looked at it in the form of trying to put together a proposition and then invalidating a proposition,” he says.
The system would take a “factoid,” run searches around it, and ask, as Kyler puts it, “Is this related or not?”
The code reviewed describes InquiryIQ as able to run web and image searches, browse web pages, and use face recognition on photographs it encounters. As the system searches from information supplied by the investigator, it is designed to build what Clearview calls a “Candidate Graph” of possible identities and associates.
It fills out the subject’s profile with possible addresses, phone numbers, employers, social media accounts, arrest history, and aliases.
Clearview is not the first company to automate the process of trawling publicly available information. Other intelligence platforms sold to law enforcement, including ShadowDragon’s SocialNet, Penlink’s Tangles, and Fivecast, help investigators uncover aliases and associates and map a person’s digital footprint.
Andrew Guthrie Ferguson, a George Washington University law professor who studies AI and policing, describes this kind of automated investigation as “digital rummaging.” “They’re basically going to create a profile of you based on all of the random digital clues you left on the internet,” he explains.
Woodrow Hartzog, a Boston University privacy scholar, argues that the labor of investigative work once served as a practical check on surveillance. By vastly reducing the labor required to investigate a person, tools like InquiryIQ also eliminate those checks.
“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,” he says. “There are a lot of rules we never had just because we never needed them—because there were these practical barriers to following everyone around.”
When an InquiryIQ search is finished, the interface is designed to present the officer with the identities, connections, and other details the system has surfaced. The officer can accept or reject them before they are added to the profile.
Clearview warns that automatically generated demographic, social media, and arrest data “may or may not be accurate.” Before accepting any finding, the officer must attest that they independently verified it.
Kyler says that human review is central to Clearview’s design. The system is meant to surface possible leads, not determine what is true, he says. “That’s the job of an analyst—to evaluate what’s true, what’s not; what’s noise, what’s reality.”
“A human in the loop is a little bit of a cold comfort,” Hartzog says. Over time, investigators can defer to automated systems until the person checking the machine becomes “a sort of rubber stamp.”
For example, in United States v. Sant, a Minnesota case involving undercover Homeland Security Investigations agents and surveillance of political activists, defense lawyers obtained a Clearview report drawing matches from roughly 15 years of protest photography.
Every result was stamped “Accepted by Guy Gino,” including one Clearview itself labeled “A Less Likely Result”—which, the report’s footnote notes, can only be exported if a user accepts it. Defense attorneys allege the report swept in photos of an entirely different man, his pregnant wife, and young daughter.
There is no evidence InquiryIQ was used in the case.
Pitfalls and Silver Linings
In the prototype reviewed, what InquiryIQ returned could vary depending on which model was selected. Its interface includes a control for choosing the model that runs the research and lists xAI and Amazon Bedrock, a platform for accessing models built by other companies.
In 2025, what xAI said was an unauthorized change to Grok’s system prompt caused the chatbot to inject claims about a supposed “white genocide” in South Africa into unrelated conversations. Less than a month later, after another change to its instructions, Grok produced antisemitic posts and praise for Adolf Hitler.
SpaceXAI did not respond to a request for comment.
“A hallucination-prone chatbot would not be trusted as an informant under any other regular circumstances,” says Michael Price, litigation director of the National Association of Criminal Defense Lawyers’ Fourth Amendment Center. He refers specifically to information police rely on to establish probable cause in court.
“This person just told me to go eat rocks and drink bleach. And they’re going to be the basis for probable cause?”
Price, who helped litigate Chatrie v. United States, the landmark Supreme Court case over geofence warrants, argues that concerns around the reliability of generative AI tools extends beyond Grok.
AI models are trained on material pulled from the internet, including, he says, “uninformed posts, conspiracy theories, all of the prejudices that sadly seem to go along with the internet.”
