Sirius XM Wins AI Hiring Bias Case

TL;DR On September 30, 2026 a federal court granted Sirius XM judgment in Harper v. Sirius XM Radio, LLC, a suit claiming the iCIMS hiring software discriminated against a Black applicant who applied to about 150 roles. HR Dive reported it on October 5. The reported reason is thin evidence, not a finding that AI screening is lawful, and the plaintiff had no lawyer. Keep an evidence file for every automated screening step you use.

An employer won an AI hiring discrimination case this week, and a lot of people will read that as "the lawsuits are not going to land." Read the facts first. The plaintiff represented himself, the court reportedly found he had not shown how the software affected him, and the case that most employers actually fear, the one against Workday, is still going and has survived a state-law motion.

This page covers what we know about Harper v. Sirius XM, what it does not decide, and a template for the one thing that does help you in court: being able to say exactly what your screening tool does.

What happened and when

The ruling came on September 30, 2026. HR Dive published the first report on October 5, 2026, and says Sirius XM was granted summary judgment. Its own article later describes the court granting a motion to dismiss, so we cannot tell you from that source alone which procedural step it was. We have not read the order.

According to the complaint as summarized by Fisher Phillips, iCIMS scored applications using data that can stand in for race, such as school, home zip code and work history. According to the court papers HR Dive quotes, he also said he used several email addresses to avoid "suspected algorithmic penalties for repeat applications." He did get one interview, for an IT desktop support role in late 2023, where he allegedly had to answer questions about the multiple addresses.

What the court reportedly said

Two things, both from HR Dive's summary.

On intentional discrimination: with no direct evidence and no circumstantial evidence, the plaintiff could not make the claim. That is the standard framework for disparate treatment. Without something that points to intent, a rejection is just a rejection.

On disparate impact: the court said he had not shown how the employer's alleged use of AI could have discriminated against him because of his race.

Notice what is missing. Neither point says the software is unbiased. Neither says Sirius XM tested it. They say the plaintiff did not carry his burden. A plaintiff with a lawyer, an expert and discovery into the scoring logic is a different case.

What this does not decide

It is not a ruling that AI screening is lawful. One court, one set of facts, one self-represented plaintiff.

It does not touch the statutes that name AI directly. New York City's Local Law 144, Illinois's AI employment rules and California's automated decision rules each set their own notice and audit duties, and none depends on the outcome of a Title VII case. Our guides cover them: NYC Local Law 144, multi-state bias audit requirements and California ADMT templates.

It does not end the Workday case. HR Dive notes Workday could not defeat the state discrimination claims and the case is still ongoing. Our Workday screening checklist covers what that case asks of employers who use the same kind of tool.

It does not mean federal pressure is coming back. HR Dive places the case against a backdrop of federal retreat on disparate impact: an April 2025 executive order telling agencies to stop enforcing it, and a June 2026 Justice Department view that EEOC's disparate impact guidelines are unconstitutional. We covered the agency side in FTC drops disparate-impact claims and in our EEOC checklist. The practical result is that private plaintiffs and states carry more of the load, which is the reason your own records matter.

Why this still matters if you are small

The most useful detail in the case is not the outcome. It is how the plaintiff got confused about what the software did. He believed repeat applications were penalized, so he created extra email addresses. Whether or not the tool did that, nobody in the story could say.

If a rejected applicant sues you, the first question is what the tool actually does. "We use iCIMS" is not an answer. "It ranks resumes against the job description, it does not reject anyone automatically, a recruiter reviews the top 20, and it treats repeat applications the same as first ones" is an answer, and it is an answer you can only give if you wrote it down before the lawsuit.

Template: AI screening evidence file

Fill one in per tool and per role family. Date it, keep old versions, and re-check it when the vendor ships a major change. This is our suggested structure, not a court or agency form.

AI SCREENING EVIDENCE FILE
Tool and version: [e.g. applicant tracking system name, module, date last updated]
Owner at our company: [name and role]
Roles covered: [job families]
Date written / last reviewed: [date] / [date]

1. WHAT IT DOES
   Inputs it reads: [resume, application form, assessment, video, other]
   Output it produces: [score / rank / yes-no flag / none]
   Can it reject a candidate with no human step?  [yes / no]
   If yes: which stage and which rule: [description]

2. HUMAN REVIEW
   Who reviews candidates the tool ranks low: [role]
   Share of low-ranked candidates a human actually opens: [percent, how measured]

3. REPEAT AND DUPLICATE APPLICATIONS
   How the tool treats a second application or a second email: [description]
   Is that treatment disclosed to candidates?  [yes / no]

4. BIAS TESTING
   Last test date: [date]    Done by: [internal / vendor / third party]
   Groups tested: [race, sex, age, disability, other]
   Result and any change we made: [summary]

5. NOTICE AND ACCOMMODATION
   What candidates are told about automated screening: [text and where shown]
   How a candidate asks for an alternative or an accommodation: [route]

6. RECORDS
   What we keep, where, and for how long: [description]
   Who can pull a candidate's score history on request: [role]

7. VENDOR
   Contract clause on audit access and incident notice: [clause reference]
   Open questions we have asked the vendor: [list and dates]

Section 3 is the one most teams skip. It is also the detail that shaped the plaintiff's own behavior in Harper.

Five questions for your applicant tracking vendor

1. Does the product rank, score or reject, and which of those is on by default?
2. How does it treat a second application from the same person or a second email address?
3. What bias testing exists, who did it, and can we see the summary?
4. Can we export a candidate's score and the inputs behind it?
5. What changes when you update the model, and how will you tell us?

If the vendor cannot answer the first two in writing, treat that as a finding. For a broader process, our small-team AI hiring compliance guide walks through the steps in order.

What we could not verify

• We did not read the court's order. Everything about the ruling comes from HR Dive's October 5 article, which uses both "summary judgment" and "motion to dismiss."
• HR Dive says it asked Sirius XM and iCIMS for comment and did not hear back before publishing. We have no statement from either company.
• The docket number and judge come from a Fisher Phillips summary written when the case was filed in 2025, not from the current docket.
• We could not confirm whether the plaintiff will appeal.
• We did not find a verified post from either party about the ruling, so this page has no embed.