HOAN TON-THAT
- Status
- ACTIVE — Co-founder, Clearview AI; chief executive from 2017 through the enforcement record documented in this file
- Hazard
- 76
- ATK / DEF / HP
- 8 / 7 / 6
Behavioral Archetype
THE INDEXER — Subject is the app developer who found the product that made him consequential: a search engine whose corpus is the human face. Before Clearview there were iPhone apps — Trumpet, a video site called ViddyHo — the ordinary churn of a consumer-software career. Then the same instinct was pointed at an unclaimed asset class: every photograph on the open web with a face in it and a name near it. Take what is public, make it searchable, sell the search. Applied to web pages, that instinct built the companies everyone admires. Applied to faces, it ended anonymity in public as a default condition of life, and subject applied it to faces. The throughline is not surveillance doctrine — subject has none on record. It is indexing. The file’s closing line about the technology is the file’s finding about the man: the face is now an index term, and he is the one who indexed it.
Essence Indicators
- Career before Clearview: iOS app developer — Trumpet, and the video site ViddyHo, which drew phishing accusations in contemporaneous 2009 tech-press coverage revisited by the New York Times profile; no charge or adjudication is asserted, and none is documented
- Co-founded Clearview AI in 2017 with Richard Schwartz; early backing included $200,000 from Peter Thiel; the company operated in near secrecy until Kashmir Hill’s New York Times investigation of January 18, 2020, which documented a database of some three billion images and more than 600 law-enforcement agencies using the tool, including the FBI and the Department of Homeland Security
- The odometer statements are his. In March 2023 subject said the database had passed 30 billion images, that police had run roughly one million searches, and that 2,400+ agencies were clients — his own figures, offered by the company’s own founder, unaudited by anyone. The company has stated a goal of 100 billion images. The scale of the dragnet is documented primarily by the subject announcing it
- The enforcement record accrued under his tenure and never stopped the operation: fines from four European regulators (Italy, France, Greece, the Netherlands — the Dutch DPA’s EUR 30.5M in September 2024 “did not stop the US business,” per Reuters), a UK fine currently remitted after three levels of tribunal litigation, the March 2025 Illinois BIPA settlement paid in a ~23% equity stake because the company lacked cash, and a Vermont AG suit dismissed December 2025 for lack of personal jurisdiction — full ledger at the company file (cross-reference
clearview-ai.md)
- The defense travels with the charge. The company’s product is, by US government testing standards, among the most accurate facial-recognition systems available — the file’s finding is the database, not the error rate. Its jurisdictional arguments have repeatedly won: the UK First-tier Tribunal accepted in 2023 that a foreign firm serving foreign law-enforcement clients sat outside UK data-protection law, and Vermont’s suit died on presence. And the wrongful-arrest record that shadows the technology class — NIST’s demographic false-positive findings, the Detroit arrests — is not attributed to Clearview’s product; the Nijeer Parks attribution was retracted by his own attorney and is denied by the company
- The biographical fact the file turns on: a consumer-app developer with no institutional surveillance background built, in three years of obscurity, the face index that states with subpoena power never assembled — and disclosed its scale himself, in billions, as marketing
Social Persona / Impression Management
Immediate impression: Founder-salesman. The register is the product announcement, not the deposition — where the apparatus’s other principals speak in risk-management grammar, subject speaks in inventory milestones.
Energy: Product-first, asset-first. Does not litigate the philosophy of anonymity in public. Announces the number, cites the solved cases, lets the database do the standing.
Impression management strategy: The crime-solver. The work routes to helping police identify suspects, victims, and witnesses after the fact — among the most defensible destinations a surveillance product can choose, and the tool’s accuracy is genuinely high by the government’s own testing. Whether the enterprise is a public-safety instrument with a database attached or a permanent face index with a public-safety front door is the question the litigation record keeps asking, and jurisdiction keeps deferring. That deferral is what makes the position effective rather than simply contested.
Forensic Archetype Comparison
| Pattern | Match Level | Evidence |
|---|---|---|
| The Indexer | MAXIMUM | Built the reverse index of the human face — 3B images at exposure, 30B by his own 2023 statement, 100B the stated goal. |
| The Engineer | HIGH | Came up through consumer app development; personally built the product before the company had a public existence. |
| The Operator | MODERATE | Ran the company through five years of multinational enforcement, but the durable instrument is the asset, not the organization. |
| The Statesman | LOW | No policy apparatus, no commission seats, no revolving door. The apparatus’s institutions sued him; he did not join them. |
| The Financier | LOW | Took Thiel’s early money; did not fund the apparatus. Built the asset the litigation now prices. |
Psychometric Assessment
Big Five (OCEAN):
| Trait | Score | Evidence |
|---|---|---|
| Openness | 76/100 | Consumer apps to a face-search engine is a category jump most developers never make. The instinct — index what is public, sell the search — is fixed; the target moved. |
| Conscientiousness | 68/100 | Built and scaled a single product through five years of fines, suits, and tribunal fights without folding the company. The follow-through is documented; the consent step was skipped by design. |
| Extraversion | 62/100 | Moderate. The database milestones are announced personally, on the record, to the press — the founder as his own press office. |
| Agreeableness | 35/100 | LOW. The product is built on never having asked. Non-consent is not a side effect of the design; it is the input the design runs on. Observation of the documented product, not of private manner. |
| Neuroticism | 32/100 | Low. Kept operating, and kept announcing growth, through four national fines, a remitted fifth, and a settlement paid in the company’s own stock. |
Dark Triad:
| Trait | Score | Notes |
|---|---|---|
| Narcissism | 55/100 | MODERATE. The scale of the database is disclosed by the founder himself, in billions, as marketing. Within normal range for a founder whose product is its own superlative. |
| Machiavellianism | 70/100 | MODERATE-HIGH. The business’s most effective legal product is its jurisdictional posture — foreign clients for foreign courts, insufficient presence for state courts. This is observation of documented litigation outcomes, not an inference about private character. |
| Psychopathy | 35/100 | LOW-MODERATE. No documented personal cruelty. The indifference on the record is procedural — scrape first, litigate after — and it is the company’s method, sustained across every jurisdiction that objected. |
MBTI: ENTP (“The Debater”) — Dominant extraverted intuition. Sees the unclaimed asset class before the law names it, builds the index before the regulator arrives, and treats each enforcement action as a term to be negotiated rather than a wall.
Threat Assessment
| Category | Level | Notes |
|---|---|---|
| Physical threat | NONE | No documented history of personal violence. |
| Institutional threat | HIGH | Built and sold the privatized face-identification layer of US law enforcement — 2,400+ agencies by his own count, with no statute governing most of them. |
| Memetic threat | MODERATE | Subject does not define the field’s vocabulary or write its frameworks. His contribution to the discourse is a number — thirty billion — and the number did more work than most frameworks. |
| Civilizational threat | HIGH | The body layer does not rotate. A face cannot be reissued like a password, and the index of it — assembled without consent, held by a private firm, surviving every remedy — is the layer every other capture in this series ultimately keys to. |
Alignment Analysis
Stated alignment: Help law enforcement identify suspects, victims, and witnesses after the fact, using only publicly available images. Solve crimes. Build an accurate tool.
Observed alignment: Build the largest privately owned index of the human face, disclose its growth personally as marketing, and hold the asset through — and beyond — every enforcement action aimed at it.
Gap assessment: The stated and observed alignments overlap wherever “give police an accurate investigative tool” coincides with “own the face index everyone’s investigation depends on.” The record’s own numbers frame the gap: the tool’s accuracy is high by government testing, the client list is real, and some crimes were surely solved — and the price of that instrument was the enrollment, without consent, of everyone who ever posted a photograph. Europe priced that at roughly ninety-five million euros across four regulators; Illinois priced it at a quarter of the company; Vermont was told the company wasn’t there to be priced. The remedies were partial by design of no one — each system worked as built — and the aggregate is the finding: every wall the record produced turned out to be a fine, and the index survived them all.
The comparison the apparatus file invites is directional. The field’s other builders constructed instruments for watching the information space and pointed them at speech; subject built an instrument for watching bodies and pointed it at faces. Speech, the series’ closer argues, is corrigible — the body is not. The man who indexed the face holds the layer underneath everyone else’s apparatus, and he got there not through doctrine but through the oldest move in software: index what is public, and let the meaning of “public” absorb the loss. The motive is his to know; the database is on the record, in his own numbers.
Convergent Drive Classification
Self-preservation: Survives by locating the company where remedies do not reach — foreign clients for foreign regulators, insufficient presence for state courts. The litigation record is a map of jurisdictions that could not hold it. Goal preservation: The database, once scraped, cannot be un-scraped; deletion orders reach the residents a regulator can prove, and the index keeps the rest. The goal is preserved in the asset’s own physics. Resource acquisition: The resource is faces, and the acquisition curve is announced personally: three billion, thirty billion, one hundred billion stated as the goal. Self-improvement: Each announcement is the same instinct at higher magnitude — index more of what is public, and let scale convert a company’s decision into a fact of nature.
Subject is not an AI system. The drives appear anyway — in the founder whose product is the index of every face that was ever public.
Public footprint: clearview.ai (company).
Sources: The Secretive Company That Might End Privacy as We Know It — New York Times, Jan 18 2020; Clearview CEO claims company’s database of scraped images is now 30 billion strong — Engadget, Mar 2023; $51.75M Settlement in Clearview AI Biometric Privacy Litigation — Regulatory Oversight, Apr 2025; Clearview AI fined $33.6 million by Dutch watchdog — Reuters, Sep 3 2024; Judge throws out Vermont’s lawsuit against facial recognition software firm Clearview AI — VTDigger, Dec 18 2025; First BIPA Litigation: Class Members Receive Equity in Clearview AI — National Law Review, 2025. Enforcement-ledger detail and per-action sourcing at the company file, clearview-ai.md.
Get updates on the Evil Robots series
Newsletter essays on AI escape, deception, and the humans who built them.