Discarded, the Complaint Alleges,
“Before a Human Being Ever Looks.”

Two job applicants have asked a federal court to rule on what an AI hiring screen legally is. Their complaint alleges Eightfold AI’s Match Score grades candidates from 0 to 5, that employers typically review only the highly ranked — and that the person being scored never sees the report, because under Eightfold’s reading of the law, there isn’t one.

By William Douglas Baker

Founding Editor

The instrument at the center of Kistler v. Eightfold AI is called the Match Score. According to the complaint, it is an AI model built on what Eightfold’s own marketing, quoted in the filing, calls the “world’s largest, self-refreshing source of talent data,” and it grades job applicants from 0 to 5, in half-point increments, on their “likelihood of success.” Employers sort the scores into a ranked list and, the complaint alleges, typically review only the candidates at the top. Then comes the sentence this edition takes its headline from: “Lower-ranked candidates are often discarded before a human being ever looks at their application.”

Erin Kistler holds a computer-science degree from Ohio State and has roughly nineteen years in product management. In December 2025, the complaint recounts, she applied for two senior product-manager roles at PayPal — applications routed, per the filing, through a link carrying “eightfold.ai/careers” in its address. She and co-plaintiff Sruti Bhaumik now represent a proposed class of applicants scored by a system the complaint says they were never told existed. What it pleads Ms. Kistler never received is specific: “a standalone disclosure providing clear and conspicuous notice that a consumer report based on her personal data would be obtained.”

The complaint’s description of what feeds the score reaches well past the résumé. Eightfold’s technology, it alleges, “lurks in the background of job applications,” assembling “social media profiles, location data, internet and device activity, cookies and other tracking” into a profile of “behavior, attitudes, intelligence, aptitudes and other characteristics that applicants never included in their job application.” Whether every clause of that description survives discovery is the litigation’s work. The structure it describes is this record’s: data the applicant never submitted, weighed by a model the applicant cannot see, delivered to a decision-maker the applicant never meets.

The legal theory is older than the technology by half a century, and the complaint states it without apology: “There is no AI-exemption to these laws.” The Fair Credit Reporting Act has governed third parties that assemble and evaluate information about people for employment decisions since 1970 — Congress’s answer, the complaint notes, to a market in secret “dossiers.” If a Match Score is a consumer report and Eightfold is a consumer reporting agency, familiar machinery follows, distributed across the parties the statute assigns it to: the employer procuring the report must give the applicant a standalone disclosure and obtain authorization; the reporting agency must obtain the employer’s certification of compliance before furnishing it; and the applicant acquires rights to see, correct, and dispute what was assembled, including protections before an adverse action lands. California’s ICRAA imposes overlapping, and in some respects additional, requirements. The complaint alleges the system operated with none of those protections in place.

The theory did not arrive from nowhere. In November 2024, the Consumer Financial Protection Bureau issued Circular 2024-06, “Background Dossiers and Algorithmic Scores for Hiring, Promotion, and Other Employment Decisions,” taking the position that a company using outside data to generate employment scores could qualify as a consumer reporting agency. The Bureau withdrew that circular on May 12, 2025, months before this action was filed. The guidance is historical context now, not governing policy — which is what makes the posture clean. If the plaintiffs win, they will have won on the statute itself.

Eightfold’s answer, filed April 20, is categorical: the plaintiffs “do not plausibly plead that Eightfold is a consumer reporting agency,” and the federal and state claims fall with that premise. The plaintiffs opposed on June 18; Eightfold replied July 9. A hearing was set for August 4 — then, by order entered July 24, Judge Yvonne Gonzalez Rogers vacated it, reserving the choice between resetting argument and deciding the motion on the papers. No new hearing appears on the docket. As of this edition’s publication, no ruling has issued; the most recent entry is a routine attorney admission, dated August 12. The question of what this instrument legally is sits fully briefed and undecided.

What the machine assembled, and who never saw it

In this record’s clinical register, the shape has a name: Decision Exclusion, one of the five clinical characteristics of Digital Derangement Syndrome™ — a decision about a person, made by a system the person is structurally outside of. Edition No. 061 filed its cousin three days ago: a file present in the room and, per that complaint, never opened. This is the harder variant. Here there is no file to point at, because per this complaint the person was never told a file existed. Exclusion by architecture, the complaint alleges, not by oversight — no notice, no copy, no meaningful opportunity to dispute, and, for those ranked out of consideration, potentially no human review at all.

Why this ruling reaches past hiring

Read narrowly, Kistler is about résumés. Read structurally, it is an unusually direct test of a question automated evaluation keeps raising: when a machine assembles a picture of you and hands a decision-maker a number, does the person behind the number have any right to the picture? The complaint names the customers whose applicant flows allegedly run through the platform — Microsoft, Morgan Stanley, Starbucks, BNY, PayPal, Chevron, Bayer — and cites industry reporting (¶ 7) that nearly two-thirds of large companies now use AI screening of some kind. Whatever the court decides, the screening layer is not an experiment. It is the front door.

That is why this record files it. The professionals this record is written for rarely meet a Match Score by that name. They meet the shape constantly: a lender’s pre-read, a panel’s automated shortlist, an answer engine’s ranking of who counts as the authority worth naming. In every one of those rooms, the deciding layer assembles its own picture from whatever it can reach — and the difference between being weighed and being discarded unread is whether your authority was encoded where the system actually looked. The complaint calls its version a consumer report. The Inference Era has not yet been asked what to call the rest.

A ruling either way sharpens the record. If the motion is denied on the statutory question, the court will have held that these allegations plausibly place the Match Score inside the federal consumer-reporting framework, with the disclosure machinery that follows. If it is granted, the meaning depends on the why: the court could reject the theory outright, find the pleaded facts insufficient, or grant leave to amend. What no ruling on this motion can say is that every AI screen is, or is not, a report about the person it scores. The motion is with the court. This record will file its answer when it lands.