Two Lawsuits Every AI Governance Professional Should Be Watching.
Quick summaries of where both cases actually stand.
Meta — Doe v. Meta Platforms (filed July 2026): Twenty-six current and former Meta employees allege that behavioral telemetry scoring systems used in Meta's May 2026 layoffs couldn't account for workers on protected medical, parental, or disability leave — making it potentially the first legal test of AI-assisted layoff selection under the FMLA. The employees allege Meta used internal AI tools, including Metamate, for performance scoring and ranking during a workforce reduction of roughly 8,000 people. The complaint spans 21 causes of action across the FMLA, ADA, Title VII, the Pregnancy Discrimination Act, and state statutes. On July 17, a federal judge denied the employees' emergency request to block the July 22 layoffs. However, the judge flagged "serious questions" about the claims, so the case continues —just without an injunction. Plaintiffs are seeking an independent audit of the algorithmically assisted selection process.
Workday — Mobley v. Workday (filed 2023, ongoing): This is the hiring-screening case, and race is part of it. Derek Mobley is an African American man over 40 with depression and anxiety, suing on behalf of similarly situated applicants — he alleges he was rejected from more than 100 positions at companies using Workday's platform, sometimes within minutes or in the middle of the night. In May 2025, the court conditionally certified a nationwide ADEA collective action, with Workday estimating roughly 1.1 billion applications were rejected through its tools during the relevant period. The case now proceeds on disparate-impact claims based on race under Title VII, disability under the ADA, and age under the ADEA. Recent momentum favors the plaintiffs: in March 2026, Judge Rita Lin rejected Workday's argument that the ADEA doesn't cover job applicants, and on June 22, 2026, she again refused to dismiss most of the discrimination claims — keeping alive the possibility that vendors behind hiring technology can face direct liability for the outcomes their systems produce.
The Algorithm Will See You Now: What Meta and Workday Are Teaching Us About AI Employment Decisions
Two lawsuits are quietly writing the rulebook for AI in the workplace, and if you work in AI governance or just work somewhere that uses AI, you should be watching both.
The cases in brief. Derek Mobley, a Black man over 40, alleges that Workday's AI-powered screening tools rejected him from over a hundred jobs — sometimes within minutes, sometimes in the middle of the night. His case has survived multiple dismissal attempts, been certified as a nationwide age-discrimination collective action, and now proceeds on race, age, and disability claims. Meanwhile, twenty-six Meta employees allege that AI systems used to select people for the company's May 2026 layoffs couldn't account for approved medical, parental, or disability leave, meaning taking protected leave may have effectively lowered your score for keeping your job. A judge declined to pause the layoffs but let the case move forward.
Everything in both cases is still an allegation, not a verdict. But the trajectory matters more than the outcomes.
Where I think this is going.
1. The vendor shield is cracking. Ever since studying for the CISSP and AAISM, one thing the books and manuals have made clear is how the implicit deal has worked for years: employers point at the vendor's algorithm, and vendors say "we don't make hiring decisions." Mobley threatens both halves of that deal — the court has kept alive the theory that a technology vendor can be directly liable for the discriminatory outcomes its tools produce. This goes back to why it's so important to understand how the algorithm actually works. A lot of times we're dealing with vendors that don't want to give that information up, and at what cost? What cost to the deployer's business, their customers, their employees? If this theory holds, "we just provide the software" stops working as a defense, and every AI vendor contract gets renegotiated around bias testing and audit rights.
2. Disparate impact doesn't care about your intent. Nobody at Workday or Meta is accused of wanting to discriminate. That's the point. An algorithm trained on historical data can reproduce historical patterns with perfect statistical honesty and zero malice, and disparate impact law was built for exactly that gap. "The model did it" is not a defense; it's a confession that you didn't test the model. We've seen this before: one of the historic cases you study in AI governance is Amazon's experimental recruiting tool, which learned from years of historical hiring data and taught itself to penalize resumes associated with women. The data doesn't understand that it's discriminating. It's just data — it follows the paths it was given.
3. Opacity is becoming the liability. The more you learn about AI governance, the more one word stands out above the rest: opacity. The Meta plaintiffs' core procedural complaint is that nobody could see the scoring logic or appeal it. Courts and regulators are converging on the same instinct: if a system affects someone's livelihood and no one can explain it, the lack of explanation is the problem. This is really a governance 101 principle — you should be able to explain the decisions you make, including the ones you get from an AI. If you can't, that's an issue, and you have to go back and get the reasons before that system touches another person.
How we navigate this. None of this requires abandoning AI in HR. It requires treating employment AI like the high-risk system it is:
Know where AI touches people. You can't govern what you haven't inventoried. Every tool that scores, ranks, filters, or flags a human being belongs on a register with an owner.
Test for impact before deployment — and keep testing. Bias testing isn't a launch checkbox; models, data, and applicant pools drift. Selection-rate analysis across protected classes should be as routine as vulnerability scanning.
Keep a human decision-maker who can actually decide. A human who rubber-stamps the model's ranking isn't oversight; it's liability theater. The human needs the authority, information, and time to disagree with the machine — and the record should show they sometimes do.
Account for protected statuses by design. The Meta allegations describe a system that couldn't distinguish "unproductive" from "on approved maternity leave." That's not an edge case; it's a foreseeable input your model must handle before it ships.
Push accountability into vendor contracts. Demand bias-testing evidence, audit rights, and clarity on who answers when the tool gets sued. "Trust us" is not a control.
One more thing. I also believe we need governance at the federal level, the way the EU has with the EU AI Act. Right now, AI deployment in the U.S. is closer to the Wild West — a patchwork of state laws and lawsuits filling the gap — and that's dangerous when the systems in question are affecting people's livelihoods. Above all else, this technology should be deployed in a way that protects people: our jobs, our income, our dignity.
The organizations that will come through this era clean aren't the ones that avoided AI. They're the ones that could show their work when someone finally asked. These lawsuits are that ask, arriving on schedule.
These are my personal observations on publicly reported, ongoing litigation — all claims are allegations unless and until proven, and nothing here is legal advice.