Clearview AI has built and tested an unreleased AI assistant called InquiryIQ, designed to take a face-recognition lead and fan out across the public web to assemble a profile of the person behind it: possible employers, aliases, associates, addresses, social media accounts and arrest history. WIRED reported the prototype on the morning of 10 September 2026, after finding it in files that Clearview AI’s login page sends to every visitor before anyone signs in. One of the models the company tested came from xAI, the maker of Grok, which merged with SpaceX in February.
Clearview AI says InquiryIQ is a prototype that has never been pitched or shipped to customers and is not planned for release in its present form. “No law enforcement user has ever used it, period,” chief executive Amos Kyler told WIRED. The company describes the tool as a limited way to automate web searches that detectives already run, not as an automated investigator.
We downloaded the same public login files on the same day to see what they still say. The InquiryIQ name, its model menu and its demographic search fields are all still there, while two phrases WIRED quoted are not. Below, we set that code against the scale of Clearview AI’s image database, its federal contracts, €95.7 million in European penalties, and the practical steps UK organisations can take to shrink what an automated profiler could find about their staff.
Table of contents
- What Clearview AI’s InquiryIQ Prototype Is Built to Do
- How InquiryIQ Was Found in Code Sent Before Sign-In
- What We Found in Clearview AI’s Login Files on 10 September
- The Scale Behind Clearview AI’s Search
- Why a Model Layer Changes the Risk
- Clearview AI’s Defence: A Prototype No Officer Has Used
- Clearview AI’s Federal Customers and xAI’s Government Push
- Clearview AI’s Legal Record in Europe, the UK and the US
- Where This Leaves UK Police Face Search
- What an Automated Profile Could Assemble About Your Staff
- Six Checks Before You Buy or Build AI That Profiles People
- Clearview AI and InquiryIQ: Frequently Asked Questions
- References
What Clearview AI's InquiryIQ Prototype Is Built to Do
Clearview AI’s core product is a face search engine. An investigator uploads a probe photo, and the system returns public web pages where similar faces appear. For years the company said its job stopped there. In a 2022 post, founder Hoan Ton-That wrote that it was “up to the investigator to follow those links and do more research to find additional information.” InquiryIQ is designed to take over part of that follow-up work.
From a face match to a list of possible people
According to the interface text WIRED analysed, InquiryIQ starts after an investigator has run a face search and picked out details they consider relevant. The system then runs web searches and image searches, reads web pages and compares faces in the photographs it encounters. As it goes, it builds what Clearview AI called a “Candidate Graph” of possible identities and associates, while filling a profile with possible addresses, phone numbers, employers, social media accounts, arrest history and aliases.
Kyler describes the logic as testing a proposition. The system takes a “factoid”, runs searches around it, and asks, in his words, “Is this related or not?” That is a reasonable description of how an analyst works. The difference is that software can repeat the loop hundreds of times without getting tired, and without the investigator seeing every page it read along the way.
The model menu, and why xAI is on it
The prototype’s interface includes a control for choosing which model runs the research. WIRED found it listed xAI and Amazon Bedrock, a service that gives access to models built by other companies. Kyler says the options exist so Clearview AI engineers can compare model performance during testing, not so police can pick a model. SpaceXAI did not respond to WIRED’s request for comment, and Amazon said AWS is not involved in developing InquiryIQ.
The age, gender and race fields
The detail that drew the most attention is a set of optional search fields. InquiryIQ’s interface said supplying age, gender and race could help the system make “smarter decisions” as it searches. It is unclear how any model would use those inputs or how they would shape the output, and Clearview AI declined to speculate. Race is special category data under UK data protection law, which matters for any UK organisation tempted to build something similar.
The table below compares the product police already buy with the prototype, using only what Clearview AI has said publicly and what the interface text shows.
| Factor | Core Clearview AI face search | InquiryIQ prototype |
|---|---|---|
| Starting point | One uploaded probe photo | A persona built from a face search, plus fields the investigator enters |
| What it runs | A face match against scraped web images | Web searches, image searches, page reading and face comparison |
| Output | Links to pages with similar faces | Possible people, associates, aliases, arrests, social media links and a PDF export |
| Engine | Clearview AI’s face-matching algorithm | A selectable model, with xAI and Bedrock in the provider menu |
| Human step | The investigator follows the links | The officer accepts or rejects each finding and attests to verifying it |
| Status | Used by more than 2,000 US agencies, according to Clearview AI | Labelled an experimental prototype for employee use only |
How InquiryIQ Was Found in Code Sent Before Sign-In
Modern web applications ship much of their interface to the browser before a user logs in. That code often includes every label, warning and menu option for features the visitor cannot reach. Clearview AI’s files contain code and thousands of lines of user interface text, and that is where WIRED found InquiryIQ, described in its own words as built to “automatically discover and enrich personal data from web sources.”
The same method exposed Flock’s OS Investigate
WIRED used the same technique in August to report on OS Investigate, an AI search tool being developed by Flock Safety, reconstructing a mock-up of its interface from publicly accessible files. Interface text shows how a company has designed and described a feature. It does not show what happens on the company’s servers, how well the feature works, or who has used it, and WIRED was explicit about that limit.
What the web address says about the headline
The story’s web address reads “clearview-ai-is-testing-an-ai-tool-that-lets-cops-instantly-unearth-your-online-activity”, and the lead image file carries the same “lets cops instantly” wording. The published headline says the tool “would let” cops unearth your life online. Web addresses are usually fixed early, so the gap suggests the framing was softened before publication, which tracks Clearview AI’s central point that this is a prototype rather than a product.
Why a public login page is a disclosure channel
For any software supplier, this is a lesson in what a front end leaks. Feature flags hide buttons, but they do not remove the text bundled for those buttons. Unreleased product names, internal warnings and third-party model integrations can sit in a public file for anyone who opens a browser’s developer tools. A review of shipped bundles belongs in the scope of penetration testing for exactly this reason.
What We Found in Clearview AI's Login Files on 10 September
On 10 September 2026, hours after the story ran, we loaded Clearview AI’s public login page at app.clearview.ai and saved every script file it referenced, following each import until no new files appeared. That produced 22 files totalling 5.9 MB, including seven interface translation files. We searched them for InquiryIQ and for the phrases WIRED quoted. We did not sign in or call any service behind the page.
InquiryIQ is still described, as an employee-only prototype
The United States English translation file, 282,588 bytes long, names InquiryIQ three times. Its description reads: “InquiryIQ is an AI assistant that runs follow-on web searches and gathers publicly available information for human review as part of valid, legal searches. This is an experimental prototype for Clearview employee use only.” The model menu still offers “All providers”, “xAI” and “Bedrock”, and the optional search fields still include gender, age, race, hair colour and eye colour.
Two quoted phrases are gone
Two phrases WIRED quoted do not appear in any of the 22 files. “Smarter decisions”, the wording attached to the demographic fields, returns zero matches, and so does “enrich personal data”. The label “Candidate Graph” is absent too, although a graph panel titled “Possible people found” remains. From outside, we cannot tell whether that text was edited after WIRED’s questions or sits in files the login page no longer loads.
How the prototype sorts people
The graph panel labels each possible person “Active”, “Validated”, “Eliminated (associate)”, “Eliminated (no face match)” or “Unresolved”. Face comparisons return verdicts such as “Validated”, “Associate”, “Group match” and “Image quality check failed”. Counters report “Web searches run”, “Image searches run”, “Pages read” and “Possible people found”, and a message warns when the “image search budget” for a persona is reached. Findings can be exported as a PDF.
The persona sits at the centre
The wider interface is built around a “persona”, the profile object that face searches, notes, uploads and links attach to. A separate extractor offers an “LLM extractor mode” with “Top choice”, “Top 3 choices” and “All values” options, and data fields can be marked “Agent-generated” or “Extracted data with agent inference”, with a “System reasoning” note. Arrest records can be added from what the interface calls “our mugshots collection”.
The counts below come from a plain text search across all 22 files. A zero means the phrase is absent from everything the login page loaded on 10 September 2026.
| Interface text | Matches | What it indicates |
|---|---|---|
| InquiryIQ | 3 | The feature is still named in the shipped interface |
| xAI provider option | 5 | The model menu still lists xAI |
| Grok or SpaceX | 0 | No model or brand name beyond the provider label |
| experimental prototype | 2 | The employee-only caveat is present |
| may or may not be accurate | 1 | The accuracy warning WIRED reported is present |
| independently verified | 8 | Verification attestation wording is present |
| smarter decisions | 0 | The quoted reason for demographic fields is absent |
| enrich personal data | 0 | The quoted purpose statement is absent |
| Candidate Graph | 0 | The quoted label is absent; “Possible people found” is used |
| persona | 525 | The profile object the interface is built around |
| deconfliction | 227 | Checks for overlapping investigations on shared personas |
The Scale Behind Clearview AI's Search
Clearview AI’s own figures show why an automated follow-up layer matters. In 2020, The New York Times revealed that the company had scraped more than 3 billion images from Facebook, YouTube, Venmo and millions of other websites. The Dutch data protection authority put the database at more than 30 billion photos when it fined the company in 2024. Reporting on the February 2026 border contract cited more than 60 billion images, and Clearview AI now tells WIRED the figure is well over 70 billion.
On the company’s own stated figures, the database is at least 23 times larger than it was in 2020 (70 ÷ 3 = 23.3).
Leads from a face match, not identities
A face search returns similar faces, not confirmed identities. Face matching is one of the most heavily studied problems in computer vision, and Clearview AI says its algorithm scores above 99% accuracy across demographic groups in testing by the US National Institute of Standards and Technology. That figure measures one step. InquiryIQ adds several more, each built on web pages, image search results and model output, and no published benchmark covers how often that chain attaches the wrong associate, employer or address to a person.
More than 2,000 agencies
Clearview AI says more than 2,000 US law enforcement agencies use its technology. Under its 2022 settlement with the ACLU, it is permanently barred from giving most private companies and individuals in the US access to its database, so its growth depends on government buyers. Any feature added to the platform therefore lands in a customer base of police, border and national security agencies.
Digital rummaging
Andrew Guthrie Ferguson, a George Washington University law professor who studies AI and policing, calls automated investigation “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 told WIRED. The clues were always public. What changes is that scattered pieces, once hard to connect, become cheap to join, just as the Clarifai OkCupid photos case showed how easily shared photographs become training material nobody agreed to.
Why a Model Layer Changes the Risk
Other intelligence platforms sold to police already automate parts of this work. WIRED names ShadowDragon’s SocialNet, Penlink’s Tangles and Fivecast, which help investigators uncover aliases and associates and map a person’s digital footprint. What InquiryIQ adds is a general-purpose model deciding which leads to chase next, sitting on top of a face database that Clearview AI says holds well over 70 billion images.
Friction was a privacy control
Woodrow Hartzog, a Boston University privacy scholar, argues that the labour of investigation used to be a practical check on surveillance. “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 told WIRED. Remove the labour, and investigating people police would never have prioritised becomes affordable. WIRED’s sources describe that as cheaper fishing expeditions.
Same starting point, different answers
Generative models can produce different outputs from the same starting point, and WIRED found that what InquiryIQ returned could vary depending on which model was selected. If an investigation is challenged in court, defence lawyers may struggle to reconstruct why the system pursued one lead and dropped another, unless every prompt, search and model choice was logged at the time.
Grok’s record
Grok’s history is why the xAI option drew scrutiny. In May 2025, what xAI called an unauthorised change to Grok’s system prompt made the chatbot insert claims about a supposed “white genocide” in South Africa into unrelated conversations. Later in 2025, after another change to its instructions, Grok produced antisemitic posts and praise for Adolf Hitler. xAI has also lost in court over Grok’s image features, as our report on Minnesota’s nudification ban explains.
Clearview AI says it is continuously evaluating a wide range of available large language model options and has not assessed the suitability of any of them for InquiryIQ. That answer is consistent with a prototype, and it also means the question of which model would run a real investigation is open.
A hallucination as probable cause
Michael Price, litigation director at the National Association of Criminal Defense Lawyers’ Fourth Amendment Center, puts the evidential problem bluntly. “A hallucination-prone chatbot would not be trusted as an informant under any other regular circumstances,” he told WIRED, referring to the information police rely on to establish probable cause in court. His concern extends beyond Grok to any model trained on material pulled from the internet, including conspiracy theories and prejudice.
Clearview AI's Defence: A Prototype No Officer Has Used
Clearview AI’s position has four parts, and each is plausible on its own terms. InquiryIQ is a prototype. The model menu exists for engineering comparisons. The tool automates searches rather than making findings. And human review sits at the centre of the design. Kyler says the company’s interest is to “help generate a lead that leads to ground truth.”
Fast prototypes are normal now
Clearview AI told WIRED that modern AI development makes it much faster to build sophisticated prototypes, and that its engineers are directed to move prototypes along rapidly because it is now possible. A polished interface is therefore weaker evidence of an imminent launch than it would have been five years ago. That is a fair point, and the “employee use only” label in today’s files supports it.
Human review, and its limits
When a search finishes, the officer sees the identities and details the system surfaced and can accept or reject each one. Clearview AI warns that automatically generated demographic, social media and arrest data “may or may not be accurate”, and the officer must attest to independently verifying a finding before accepting it. “That’s the job of an analyst,” Kyler says, “to evaluate what’s true, what’s not; what’s noise, what’s reality.”
The rubber stamp problem
Hartzog calls a human in the loop “a little bit of a cold comfort”, because reviewers drift towards deference until the person checking the machine becomes “a sort of rubber stamp”. WIRED points to United States v. Sant, a Minnesota case in which defence lawyers obtained a Clearview AI report drawing matches from roughly 15 years of protest photography. Every result was marked as accepted by the same user, including one Clearview AI itself labelled “A Less Likely Result”.
Defence attorneys in that case allege the report swept in photos of an entirely different man and his family. There is no evidence InquiryIQ was used in the case, but it shows how an accept button behaves under workload, which is the control Clearview AI relies on.
| Clearview AI says | What the public files show | What stays unknown |
|---|---|---|
| No law enforcement user has ever used it | An “experimental prototype for Clearview employee use only” label | Server-side access records and any pilot accounts |
| The model menu is for engineers comparing performance | “All providers”, “xAI” and “Bedrock” options, plus “Developer mode” and “Model reasoning” labels | Which model any given run used, and whether demographic fields reach it |
| It automates searches, not findings | Status labels from “Unresolved” to “Validated” and “Eliminated (associate)” | How candidates are ranked before a human sees them |
| Human review is central | Accept and reject buttons, a verification attestation and an accuracy warning | Whether acceptance rates drift towards rubber-stamping over time |
Clearview AI's Federal Customers and xAI's Government Push
Clearview AI’s commercial direction explains why the prototype matters beyond one company. Its leadership has changed twice in recent years. Ton-That stepped down as chief executive, Hal Lambert and Richard Schwartz were named co-CEOs in February 2025 with a stated focus on federal contracts, and Kyler, who joined as an engineer in 2019, became chief executive in October 2025.
The border contract
In February 2026, US Customs and Border Protection signed a one-year, $225,000 contract with Clearview AI for “tactical targeting” and counter-network analysis. It buys 15 licences for the National Targeting Center and Border Patrol intelligence units. That works out at $15,000 per licence per year ($225,000 ÷ 15), exactly the value of an earlier order for a single Border Patrol sector.
Smaller orders came first
The 2026 award followed two smaller 2025 purchase orders for Border Patrol sectors: $30,000 for Spokane and $15,000 for Yuma. Together the three documented orders total $270,000 ($225,000 + $30,000 + $15,000). At those prices, face search is a unit-level purchase rather than a major programme, which is how a capability can spread across an agency before its governance catches up.
xAI’s route into government
xAI has pursued federal buyers directly. In September 2025 the US General Services Administration announced a OneGov agreement making Grok 4 and Grok 4 Fast available to federal agencies for $0.42 per agency over 18 months, valid until March 2027. A Grok-powered option inside a police investigation tool would sit alongside that wider push, even though Clearview AI says no model has been chosen for InquiryIQ.
The ACLU settlement boundaries
Clearview AI’s 2022 settlement with the ACLU in Illinois permanently bans the company, nationwide, from granting paid or free database access to most private entities. It also bars access for any Illinois state or local government body, including police, for five years, a period that runs to May 2027. The company had to keep an opt-out form for Illinois residents and stop offering free trials to individual officers without their employers’ approval.
Clearview AI's Legal Record in Europe, the UK and the US
Clearview AI’s scraping model has been ruled unlawful by several European regulators. The Italian, Greek and French authorities each fined it €20 million in 2022, and France added a €5.2 million overdue penalty payment in April 2023 for failing to comply. The Dutch authority fined it €30.5 million in September 2024. On those figures, European penalties total €95.7 million (30.5 + 25.2 + 20 + 20).
| Jurisdiction | Date | Action | Amount |
|---|---|---|---|
| Italy, Garante | February 2022 | Fine and order to delete Italian residents’ data | €20m |
| Illinois, ACLU settlement | May 2022 | Nationwide private-entity ban and five-year Illinois government ban | None |
| UK, ICO | 2022 to 2025 | Fine overturned in 2023, jurisdiction upheld on appeal in 2025, case sent back | £7.5m |
| Greece, Hellenic authority | July 2022 | Fine and order to delete Greek residents’ data | €20m |
| France, CNIL | October 2022 and April 2023 | Fine, then an overdue penalty for not complying | €20m + €5.2m |
| Netherlands, Autoriteit Persoonsgegevens | September 2024 | Fine, with further orders backed by penalties | €30.5m |
| US, Seventh Circuit | July 2026 | Equity-based class settlement vacated on procedural grounds | 23% equity stake, now undone |
The UK case the ICO kept alive
The ICO fined Clearview AI £7.5 million in 2022 and ordered it to delete UK residents’ data. In 2023 the First-tier Tribunal overturned the fine, finding the ICO lacked jurisdiction because Clearview AI’s clients were foreign law enforcement and national security agencies. In October 2025 the Upper Tribunal upheld three of the ICO’s four grounds of appeal, found the processing related to monitoring the behaviour of UK residents, and sent the case back to the First-tier Tribunal.
The Seventh Circuit and the equity settlement
In the US federal class action under Illinois’ Biometric Information Privacy Act, a district judge approved an unusual settlement in March 2025 that would give the class a 23% equity stake in Clearview AI instead of cash. The stake was potentially worth $51.75 million, but only at a $225 million valuation (23% of $225 million). On 13 July 2026 the Seventh Circuit vacated it, holding that the nationwide class and the state subclasses lacked separate representatives.
Why the EU AI Act matters for tools like this
Since 2 February 2025, Article 5(1)(e) of the EU AI Act has prohibited AI systems that “create or expand facial recognition databases through the untargeted scraping of facial images from the internet or CCTV footage”. InquiryIQ’s interface shows it running image searches and comparing faces it finds on web pages. Any supplier offering that capability into the EU would need careful legal analysis, and our EU AI Act compliance checklist covers the prohibited-practice screen.
Where This Leaves UK Police Face Search
UK police already use facial recognition in two established ways, and the government is writing new rules for both. The Home Office consultation on a new legal framework for law enforcement use of biometrics, facial recognition and similar technologies ran from 4 December 2025 to 12 February 2026. An automated profile builder that starts from a face match would test where such a framework draws its boundaries.
Live and retrospective searches
Live facial recognition compares faces from a camera feed against a watchlist in a public place. Retrospective facial recognition compares an image from CCTV, a phone or social media against custody photographs on the Police National Database, and the Home Office says forces run more than 25,000 such searches a month. In January 2026 the government’s policing white paper promised 40 more live facial recognition vans on top of the ten already in use.
| Type of search | What is searched | UK position |
|---|---|---|
| Live facial recognition | A live camera feed against a watchlist | In use, with more vans funded and a new framework under consultation |
| Retrospective facial recognition | Custody images on the Police National Database | Routine, at more than 25,000 searches a month |
| Open-web face search | Images scraped from the public internet | Clearview AI’s model, subject to the ICO case |
| Automated profile assembly | Web pages, image results, arrest records and social accounts | No specific framework, so general data protection law applies |
What InquiryIQ-style profiling would add
A retrospective search asks whether an unknown face matches someone already in police records. An open-web profiler asks who a person is, who they know and where they work, drawing in people who have never been arrested. For UK law enforcement, that processing would fall under Part 3 of the Data Protection Act 2018, which allows sensitive processing, including of racial or ethnic origin, only where it is strictly necessary.
What an Automated Profile Could Assemble About Your Staff
InquiryIQ is aimed at police, but the data it gathers is the same material criminals use for social engineering. Every field in its profile, from employer to aliases, can be found by anyone willing to spend a few hours. Automation shortens those hours for police and attackers alike, which is why threat intelligence work now treats staff exposure as part of the attack surface.
The profile fields and where they leak
The fields below are the ones WIRED reported InquiryIQ filling in, mapped to where that information usually comes from and what a UK organisation can realistically change.
| Profile field | Where it usually comes from | Practical control |
|---|---|---|
| Employer | Professional profiles, staff pages, press releases | Name staff publicly only where the role needs it |
| Associates | Tagged photos, team pictures, event pages | Consent and untagging rules for event photography |
| Aliases | Usernames reused across forums and platforms | Separate work and personal handles |
| Social media accounts | Reverse image search on profile photos | Different profile photos for exposed roles |
| Addresses | Company filings, reviews, planning applications | Use a service address on public registers where allowed |
| Phone numbers | Email signatures, directories, data broker listings | Switchboard or role numbers on public material |
Start with the roles attackers target
Finance staff who approve payments, executive assistants, IT administrators and anyone named in press releases carry the most risk. Their employer, reporting lines and working patterns are exactly what a payment fraud needs, and our guide to deepfake phishing defences shows how quickly public voice and video become an impersonation kit.
Run an exposure review the way an attacker would
A useful review starts with the same public sources InquiryIQ reads: search engines, image search, company filings, social platforms and data broker listings. Record what can be linked to each exposed role, fix what is under your control, and brief staff on the rest. Repeat it every year, because the web keeps indexing. It is the same reconnaissance that opens any social engineering assessment.
The limits of what you can control
Some exposure cannot be removed. Old photographs, news coverage and other people’s posts stay online, and face databases such as Clearview AI’s already hold copies. The realistic goal is to break the easy links between a face, a name, an employer and a home, not to vanish. UK data protection rights to object or to request erasure exist, but they depend on a controller that complies.
Six Checks Before You Buy or Build AI That Profiles People
InquiryIQ is also a design case study for any UK organisation building an AI assistant that searches for information about people, whether for fraud checks, recruitment screening or due diligence. The interface text shows several controls worth copying and several risks worth avoiding.
1. Treat demographic fields as special category data
Race and ethnic origin are special category data under UK GDPR. If a search tool accepts them as inputs, you need a lawful basis, an Article 9 condition and a data protection impact assessment before launch. The ICO’s guidance on special category data sets out the conditions, and our data protection service covers impact assessments for AI systems.
2. Log every model choice and every search
Ferguson sees one upside in AI-assisted investigation: a system that preserves its prompts, searches, model choices and investigative paths could leave a clearer record than a detective’s notebook. “It’s a silver lining in an otherwise potentially fraught and disruptive change,” he told WIRED. Build that audit trail from the start, including which provider answered each query.
3. Separate possible from verified
InquiryIQ’s status labels, from “Unresolved” to “Validated”, keep uncertain findings apart from checked ones. Copy that pattern. Store the source and confidence of every data point, and label model-inferred values as inferred. Clearview AI’s own field labels include “Agent-generated” and “Extracted data with agent inference”, which is the right distinction to preserve.
4. Put a budget on every search
The prototype stops image searches when a persona’s “image search budget” is reached. A hard limit on searches, pages and images per subject is a simple proportionality control. It stops a tool from rummaging indefinitely, and it gives reviewers a clear record of how far each search went.
5. Read your model provider’s terms
Amazon told WIRED that customers are responsible for complying with its acceptable use and responsible AI policies and with any third-party model provider’s terms, and that it has no service-specific terms preventing law enforcement use of Bedrock. Contractual responsibility therefore sits with the builder. Put model terms, data retention and training restrictions through vendor management review before an integration ships.
6. Test whether human review is real
Measure acceptance rates. If reviewers accept almost every suggestion, the human step has become a rubber stamp. Sample accepted findings, check them independently and track reversals. Our piece on algorithmic auditing for HR covers practical testing methods that apply to any system that scores or profiles people.
Where to draw the line
Some tools should not be built at all. An internal system that assembles profiles of private individuals from open-web images and demographic inputs is hard to justify for most businesses under UK GDPR, and selling one into the EU runs into the Article 5 prohibitions. Scope the purpose narrowly, write down what the system must never search for, and review it before every model change. As Ferguson put it after 16 years of watching police technology arrive before its rules: “It’s another new technology, the same playbook.”
Clearview AI and InquiryIQ: Frequently Asked Questions
What is Clearview AI’s InquiryIQ?
InquiryIQ is an unreleased AI assistant built by Clearview AI. Its interface describes it as running follow-on web searches and gathering publicly available information for human review after a face search. It builds a list of possible people and associates, and it can fill a profile with employers, aliases, social media links and arrest records.
Has any police force used InquiryIQ?
Clearview AI says no. Chief executive Amos Kyler told WIRED that “no law enforcement user has ever used it, period”, and the interface text we downloaded on 10 September 2026 labels it “an experimental prototype for Clearview employee use only”. Nobody outside the company can check server-side usage.
Does InquiryIQ use Grok?
WIRED found that one of the models Clearview AI tested came from xAI, the Grok developer that merged with SpaceX. The interface’s model menu still lists “xAI” and “Bedrock”. Clearview AI says the menu exists for internal comparisons and that it has not assessed the suitability of any model for InquiryIQ.
Is Clearview AI allowed to operate in the UK?
The ICO fined Clearview AI £7.5 million in 2022. A tribunal overturned the fine on jurisdiction in 2023, but in October 2025 the Upper Tribunal ruled that the ICO did have jurisdiction and sent the case back. The First-tier Tribunal must now reconsider the appeal.
How many images does Clearview AI hold?
Clearview AI told WIRED its database holds well over 70 billion images, up from more than 3 billion in 2020. The Dutch data protection authority recorded more than 30 billion when it issued its fine in 2024.
What should UK businesses do now?
Review what public information links exposed staff to their faces, employers and homes, and treat any AI tool that profiles people as high-risk processing. If you build one, log model choices, separate possible from verified findings, limit searches per subject and test whether human review works. Our cybersecurity checklist for small businesses is a practical starting point.
References
Clearview AI Is Testing an AI Tool That Would Let Cops Unearth Your Life Online (WIRED)
Flock Has a Powerful New AI Tool for Police. We Got Its Code (WIRED)
Elon Musk’s Grok AI Can’t Stop Talking About White Genocide (WIRED)
Clearview AI quietly tests InquiryIQ tool powered by xAI’s Grok model (Crypto Briefing)
CBP embeds Clearview AI into tactical targeting operations (Biometric Update)
CBP to strengthen tactical targeting and counter-network analysis with Clearview AI (FedScoop)
Settlement Ensures Clearview AI Complies With Illinois Biometric Privacy Law (ACLU)
Seventh Circuit Undoes Novel Privacy Class Settlement (Duane Morris Class Action Defense Blog)
UK Upper Tribunal hands down judgment on Clearview AI Inc (ICO)
Greek DPA imposes 20M euro fine on Clearview AI for unlawful processing of personal data (IAPP)
Challenge against Clearview AI in Europe (Privacy International)
EU AI Act Article 5: Prohibited AI Practices
GSA and xAI Partner on 42 Cent per Agency Agreement to Accelerate Federal AI Adoption (GSA)
Legal framework for using facial recognition in law enforcement (GOV.UK)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.