1.1 billion
Update, November 2026: After this piece was published, the EU adopted the Digital Omnibus on AI. It took effect on July 27, 2026 and moved the high-risk obligations for employment AI from August 2, 2026 to December 2, 2027. The Article 50 transparency rules still applied from August 2, 2026 as scheduled. The requirements themselves did not change, so the five steps below still apply. The deadline is now December 2027.
The number is 1.1 billion.
That's how many job applications Workday's court filings admit its software has rejected since September 2020. Stripped from consideration before any human looked. By any measure, this is the largest exhibit of automated employment decision-making ever entered into a court record.
Most HR teams are still calling it efficiency.
Here's what I want you to sit with for a moment. Every one of those 1.1 billion rejections was made on behalf of an organization. Not by Workday. By the organizations that enabled the tool, trusted the output, and never asked hard questions about what was actually happening inside the model.
That is where the legal and ethical weight lands. And it's landing now.
The case
In March, a federal court allowed the age discrimination claims in Mobley v. Workday to proceed as a collective action. The lead plaintiff is Derek Mobley, an African American man over forty with anxiety and depression. He holds a finance degree from Morehouse College and has spent years working in IT and financial services roles at Hewlett Packard Enterprise, the IRS, and AT&T.
He applied to more than a hundred jobs at companies using Workday's AI hiring tools. Every single one rejected him. In many cases the rejection came within minutes, sometimes in the middle of the night, before any human could have read his application.
The class he now represents could include hundreds of millions of people.
The court has also ordered Workday to produce a comprehensive list of every employer that enabled HiredScore AI, the hiring tool Workday acquired in early 2024. Those employers become the next class of defendants in the cases that follow. Some are companies you've heard of. Some are companies you advise. Some are companies on your board.
If your organization used these tools and your name shows up on that list, the legal exposure doesn't sit with Workday. It sits with you.
The research context behind Mobley matters here. A Brookings Institution study published in April 2025 ran controlled tests on AI resume screening tools currently in wide commercial use. The results were stark. White-sounding names were preferred 85.1% of the time. Black-sounding names led in only 8.6% of cases. When researchers looked specifically at how the systems treated Black men compared to white men, equal selection rates occurred exactly zero percent of the time. Not rarely. Zero.
That study tested tools that are in production today, at companies that haven't asked these questions. Ninety-eight point four percent of Fortune 500 companies run applications through an applicant tracking system, according to Jobscan's 2024 usage report. The Mobley case is not an outlier story. It is the pattern, finally documented.
The calendar
On August 2, eighty-two days from now, the EU AI Act's high-risk obligations come into force. The European Commission's Digital Omnibus proposal would have deferred those obligations to December 2027, but it failed at the April 28 trilogue. Without an agreement, the original timeline holds.
Every AI system used by a European employer for recruitment, candidate selection, performance evaluation, work allocation, monitoring, promotion, or termination is now classified as high-risk. That triggers risk assessments, technical documentation, bias testing, human oversight, transparency disclosures, and continuous monitoring. The obligations bind every organization with European employees, regardless of where it's headquartered.
Colorado's story is a different kind of cautionary tale. The state's AI Act was supposed to take effect February 1. It got delayed to June 30. A court then enjoined the state attorney general from enforcing it until rulemaking is complete, and the governor's workgroup has proposed a replacement framework, though that replacement has not yet passed the legislature. The result is regulatory whiplash. Most companies are responding by doing exactly nothing, because they're waiting for clarity that may not arrive before the federal exposure does.
Ontario, where ScaleX is based, took a quieter path. Since January 1, every employer with twenty-five or more employees has had to disclose, on every public job posting, whether AI is used to screen, assess, or select applicants. The province issued no detailed guidance. The obligation simply began. Ontario decided that the absence of clarity was less expensive than the absence of disclosure.
That's the right instinct. More organizations should adopt it.
The pattern we keep seeing
We've spent the past year sitting with people leaders inside organizations that bought hiring AI before they understood it. The conversation is almost always the same.
The vendor demoed the tool. Procurement signed the contract. Someone in HR Operations enabled the screening logic. The CHRO understood it as "we use AI for resume screening." The CEO understood it as "we have AI in our HR stack."
And then I ask the questions that matter. What does the model actually weight? What populations was it tested against? What is its false rejection rate? Who reviews the decisions it makes? Who has the authority to override it?
Nobody can answer.
When an AI system makes a hiring decision, the organization is speaking through that decision. The vendor isn't the one speaking. The people who enabled the system without understanding it are the speaker. And the speaker owns what gets said.
What the research tells us is that this isn't just a governance gap. It's also an oversight problem that compounds over time. A University of Washington study published in November 2025 put 528 participants in front of an AI hiring tool and watched what happened when the AI's recommendations were biased. When the bias was moderate, participants mirrored it completely. When the bias was severe, about ninety percent still followed the AI's lead. Eighty percent of organizations in the study had no structural mechanism to ensure humans actually overrode AI recommendations.
The authors' conclusion was direct: "Unless bias is obvious, people were perfectly willing to accept the AI's biases."
This is the part that keeps me up. Organizations bought these tools in good faith. They wanted to hire better and faster. The problem is not that they used AI. The problem is that they used AI without governance, which means they made consequential decisions about people's careers without understanding the structure of those decisions. And then they put a human in the loop who, according to the research, was almost certain to follow the AI anyway.
That is a governance failure. And it is not rare. It is the norm.
What to do before August 2
This is the part of the conversation where leaders sometimes feel overwhelmed, and I understand that. The legal landscape is complex, the tools are opaque, and the regulatory environment is moving faster than most organizations can track. But the work itself is not complicated. It is specific, and it is doable, and the organizations that do it now will be in a fundamentally different position than those that don't.
Here are the five things we tell every leader we work with.
1. Map every AI touchpoint in your hiring funnel this week.
Start at sourcing and end at offer. List every tool that touches a candidate decision: sourcing platforms, resume screeners, scheduling tools, assessment platforms, interview analysis software, offer prediction models. If you don't know what's in your stack, you cannot govern it. This step doesn't require budget. It requires someone to sit down and make the list.
2. Demand a model card from every vendor.
A model card is a vendor's written documentation of what the system does, what it was trained on, what populations it was tested against, and what its known failure modes are. If your vendor cannot produce one, that is your answer. Vendors who cannot document what their product does cannot be trusted with your hiring decisions. The Brookings study tested tools that are in wide commercial use right now. If you don't know whether your vendor was tested on the same populations as your applicant pool, you are taking a legal and ethical risk you have not priced.
3. Name a human owner for every automated decision point.
"Human in the loop" is not a policy. The UW research showed that eighty percent of organizations had no structural mechanism to ensure humans actually overrode AI recommendations. A named human with authority to override, time in their calendar to review, and a documented record of decisions is a policy. Every AI touchpoint in your hiring funnel needs one. Write down the name. Put it in writing. That document is the beginning of your governance structure.
4. Write your AI-in-HR policy before August 2.
Not a values statement. An operating policy. What you will and will not deploy AI to decide. What the controls are at each decision point. What happens when the system is wrong and who is responsible. If this document does not exist in your organization today, August 2 is your deadline. The EU AI Act's high-risk requirements describe, in detail, what this document needs to cover. Use it as your template.
5. Audit your job postings this week.
If you employ twenty-five or more people in Ontario, open your last ten public job postings right now. Check whether they disclose AI use in screening, assessment, or selection. Many organizations that are legally required to make this disclosure are still not doing it. This is the lowest-lift compliance item on this list and the most visible gap we see. Fix it today.
None of these steps require a new budget line or a new team. They require ownership, a few difficult vendor conversations, and the willingness to look honestly at what your organization has actually built.
The audit has already started
Here's what I want you to take away from all of this.
The legal deadlines are not when you begin examining what you've built. They are when the rest of the world begins examining it for you.
The Workday discovery process will produce the largest public dataset on automated hiring outcomes ever assembled. The 1.1 billion is just the beginning. The employers on that HiredScore list will have their practices examined in federal court, in the press, and in boardrooms. Some of them will be case studies. Some of them will be cases.
You still have time to choose which one you are.
Eighty-two days isn't nothing. With the right structure, it's enough time to complete all five steps above and brief your board before the August deadline lands. The leaders who do this work in the next quarter will be the ones who, in twelve months, won't be writing apology emails.
Ethics by Design isn't a slower way to operate. It is the operating system that lets fast organizations move without breaking. It is the difference between scaling and stretching. For every organization currently using AI in hiring, it is the difference between being a case study and being a case.
The 1.1 billion is already in the record. The next number is yours.
Sources
- Mobley v. Workday, Inc., Case No. 3:23-cv-00770 (N.D. Cal.). Court filings, conditional certification order, and discovery rulings through March 2026. courtlistener.com
- EU Commission Digital Omnibus proposal on AI (November 19, 2025). EU AI Act, Regulation (EU) 2024/1689, Annex III. Trilogue status: April 28, 2026 meeting ended without agreement. europarl.europa.eu
- Colorado SB 25B-004 (signed August 28, 2025). Delays AI Act implementation to June 30, 2026. April 27, 2026 court injunction enjoining AG enforcement pending rulemaking. leg.colorado.gov
- Ontario Working for Workers Four Act, 2024 (Bill 149). Amendments to the Employment Standards Act, 2000. AI disclosure requirement in force January 1, 2026. ontario.ca
- Workday announcement of intent to acquire HiredScore, February 26, 2024. Acquisition completed April 2024. investor.workday.com
- Jobscan's Fortune 500 ATS Usage Report (2024), Jobscan.

Maysa Hawwash
Founder & CEO, ScaleX
HR strategist and author of the Ethics by Design newsletter, helping organizations grow with integrity.
More in AI Governance

June 9, 2026
When Good Intentions Aren't Enough: The Governance Gap in Humanitarian AI
The humanitarian sector is deploying AI faster than it is building the governance to match. When the population you serve is in crisis, displaced, and undocumented, that lag is not a footnote. It is a failure.

May 28, 2026
When the Pope and an 18-Year-Old Student Say the Same Thing at the Same Time
A university student in Vancouver and the Pope of the Catholic Church arrived at the same diagnosis of artificial intelligence in the same week. That is not a coincidence. It is a signal.

February 19, 2026
Ethics Is Not a Principle. It's an Architecture.
Every organization deploying AI today is quietly redesigning its moral system, not through mission statements, but through workflows.
