Clearview AI and the Faceprint That Went to Court

Somewhere in a database you never agreed to be part of, a string of numbers describes the exact geometry of your face. Not a photo. A photo is just pixels sitting on a server. This is something a machine built from the photo: compact, searchable, comparable in milliseconds against millions of others. A detective could upload a grainy still from a gas-station camera and get your name back before finishing a cup of coffee.

That is the object a 2020 lawsuit put in front of a judge in Illinois. And the question the case turned on wasn’t the one most people expected.

What actually happened

Most privacy lawsuits argue about what a company did with your data. This one argued about what the company turned your data into, and why that transformation, not the underlying photo, was the thing the law actually cared about.

In May 2020, the ACLU and several partner organizations sued Clearview AI in Cook County Circuit Court, Illinois, alleging the company violated the state’s Biometric Information Privacy Act (BIPA). Clearview had built a facial-recognition search engine by scraping billions of photos from social media and other public websites, then sold search access to law enforcement agencies and, for a period, to private companies and individuals.

BIPA requires informed, written consent before a company collects a “biometric identifier” from an Illinois resident. Clearview never asked anyone. Its defense reached for the First Amendment: compiling and analyzing publicly available images, it argued, is protected expressive activity, not “collection” in the sense BIPA regulates. (A neat trick, building a for-profit identification engine and calling it free speech.) It was a serious argument, and it could have ended the case before it started.

The court didn’t buy it. In an August 2021 ruling on Clearview’s motion to dismiss, Cook County Circuit Judge Pamela Meyerson rejected the company’s First Amendment defense: sharing a photo publicly doesn’t make it free for any use. That distinction ended up carrying the whole lawsuit. Scraping a photo is one act; turning it into a searchable identifier is another. That single ruling is why everything downstream (the discovery, the technical questions, the eventual settlement) was even possible.

The case ended in May 2022 with a consent order: Clearview agreed to stop selling access to its faceprint database to most private companies and individuals nationwide, and to restrictions on sales to police and government agencies in Illinois specifically. (A “consent order” being, notably, the first form of consent Clearview had ever bothered with.)

The artifact: not the photo, the faceprint

Here’s the part that made the case matter beyond one company’s bad year. A photo of your face, sitting on a public Instagram profile, isn’t a biometric identifier under BIPA. The faceprint is: the string of numbers an algorithm derives from the geometry of your face, built so it can be compared against millions of other such strings in milliseconds. Clearview’s actual product was never “photos.” It was that conversion step: unstructured internet images in, a searchable identity index out. (A company can scrape a billion faces and still insist, with a straight face, that it isn’t in the business of identifying anyone, right up until a police detective uploads a still from a surveillance camera and gets a name back in seconds.)

That distinction is why the case mattered past Clearview. It drew a legal line between “a photo exists online” and “a photo has been converted into a queryable biometric record,” and put the second one squarely inside consent law.

What a computer forensics expert witness actually tests in a facial-recognition case

This is exactly where a computer forensics expert witness does work a lawyer’s summary can’t reach. Underneath every legal argument in a case like this sits a stack of technical questions nobody in the courtroom can answer by reading the pleadings: How did the scraping pipeline collect and retain the source images? What does the matching algorithm actually output (a probabilistic score, a ranked list, a single match), and how is that output stored and indexed? Does the system persist the derived faceprint independently of the source photo, and for how long? Can the database reconstruct, for any individual record, the chain from a specific scraped URL to a specific faceprint to a specific search hit?

The answers decide which of two very different lawsuits you actually have. “The company scraped public photos” is a much weaker claim under most biometric statutes than “the company collected biometric identifiers without consent,” and BIPA is built entirely around the second one. A software expert examines logs, database schemas, and the matching pipeline itself to establish which of those two things happened. That technical finding is often what decides whether the statute even applies.

Why this matters beyond one company

Facial-recognition databases built from scraped photos are not unique to Clearview, and Illinois is not the only jurisdiction with a biometric consent statute. The consent order restricts one company’s sales practices; it doesn’t resolve the broader question of whether mass-scraping public images to build biometric infrastructure is lawful anywhere consent isn’t obtained. As more jurisdictions pass biometric privacy laws, the technical distinction this case turned on (a photo versus the identifier extracted from it) is likely to keep being the fact pattern litigators have to prove.

The above is general information and not legal advice. Case facts are drawn from the sources listed.