A. The Landmark Case of Mobley v. Workday (The Algorithm Rejected Me)

(1) Agent Liability Theory for AI Employment Decisions

Derek Mobley applied to more than 100 companies. He never landed a single interview.

He was Black.

He was in his 40s.

He had been diagnosed with anxiety disorder.

His resume listed a degree from a historically Black college or university (HBCU), with a graduation year of 1995.

He had decades of professional experience.

So why couldn't he get even one interview?

One day he noticed a strange pattern. Rejection emails arrived less than an hour after he submitted his applications. Sometimes in the middle of the night. No human reviewer could work that fast.

He went to a lawyer.

In February 2023, Mobley filed a complaint in the U.S. District Court for the Northern District of California. The defendant was not the 100 companies that had rejected him. It was a single company called Workday. Workday provides human resources management software. Thousands of companies worldwide use its hiring platform. Most of the 100 companies Mobley applied to were running Workday's system. This is where a legally interesting question emerges. Workday was not Mobley's employer. It never tried to hire him. It only provided the hiring tool. So can Workday be held liable for employment discrimination?

Mobley's lawyers invoked a legal doctrine called "Agent Liability." Put simply, it works like this. Suppose you hire a real estate agent to sell your house.

If that agent refuses to show the house to buyers of a certain race, the agent bears liability for discrimination, because the agent acted on your behalf. Mobley's side argued that Workday was no different. Workday did not just provide a tool. It played a central role in deciding who got called for interviews and who got screened out.

Workday pushed back. We are not an employer, they said. Hiring decisions are made by our clients. We are merely software that implements the criteria they set.

On July 12, 2024, Judge Rita Lin denied Workday's motion to dismiss in large part.

The key sentence in her ruling read: "Workday's software does not mechanically implement criteria set by employers; rather, it participates in the decision-making process by determining which applicants to recommend and which to reject."

That single sentence shook the entire AI hiring tool industry.

(2) Recognition of Compound Discrimination Based on Age, Race, and Disability

Mobley's lawsuit was not a simple age discrimination case.

He asserted all three protected characteristics: age (over 40), race (Black), and disability (anxiety disorder). In legal terms, this is called "intersectional discrimination."

The easiest way to explain intersectional discrimination is through math. Suppose being Black costs a 5% disadvantage, being in your 40s costs another 5%, and having a disability costs yet another 5%. Simple addition gives you 15%. But in reality, these disadvantages multiply. A Black person in their 40s with a disability can face a far greater penalty than a white, non-disabled person in their 20s.

Mobley's lawyers argued that AI systems can amplify this kind of compound discrimination.

Workday's algorithm looks at a candidate's graduation year. A 1995 graduation allows a rough estimate of age.

It looks at an HBCU on the resume. Race can be inferred.

It looks at gaps in the employment history. Health issues can be guessed.

The algorithm combines all of this information and assigns a "hiring recommendation score."

The problem is that nobody knows exactly how that score is calculated.

Judge Lin found Mobley's intersectional discrimination claim sufficiently plausible to proceed to trial. She noted that Mobley "applied to over 100 positions across diverse job categories yet was rejected at the screening stage in every single application," and that "rejection emails arrived outside business hours, within one hour of submission." Taken together, these two facts supported a reasonable inference that Workday's algorithm was automatically rejecting applicants based not on qualifications but on protected characteristics.

(3) Analysis of Workday's AI System and Expansion into a Class Action

On May 16, 2025, the case entered a new phase. Judge Lin granted conditional certification of a class action under the Age Discrimination in Employment Act (ADEA).

Here is why class certification matters.

If Mobley litigates alone, Workday only has to compensate him.

But once a class is certified, everyone in a similar situation can join the lawsuit. According to filings Workday's legal team submitted to the court, the total number of applications rejected through Workday's system during the relevant period was 1.1 billion. Filtering for applicants aged 40 and over could yield "hundreds of millions" of people.

Workday seized on this very point as grounds for dismissal. Conducting a class action on behalf of hundreds of millions of people is practically impossible, they argued. Judge Lin's response was blunt: "If the class numbers in the hundreds of millions, that is because Workday stands accused of discriminating against hundreds of millions of people. The breadth of the alleged discrimination cannot serve as a basis for denying notice."

After this ruling, the parties discussed how to notify potential class members of their right to participate.

In December 2025, the court approved a notice plan. In traditional class actions, notice goes out by mail. But mailing hundreds of millions of people is impossible.

Judge Lin permitted notice via social media, and even through Workday's own platform. It was unprecedented in class action history.

The significance of this case goes beyond a single lawsuit. Every vendor that provides AI hiring tools is watching. If Workday loses, companies like HireVue, applicant tracking systems, and Pymetrics could face similar litigation. And the employers using those tools could be next. Judge Lin made this explicit in her ruling: "This lawsuit is one of the first large-scale legal tests of AI hiring tools."

B. EEOC and Federal Agency Enforcement Actions

(1) The iTutorGroup Age Discrimination Settlement: The First Federal Sanction

In August 2023, the U.S. Equal Employment Opportunity Commission (EEOC) made a historic announcement. An online English tutoring company called iTutorGroup agreed to pay $365,000 to settle charges that its AI hiring system engaged in age discrimination.

iTutorGroup's AI system was straightforward.

It checked an applicant's age and automatically rejected women 55 and older and men 60 and older, before any human ever reviewed the resume. This was blatant intentional discrimination. Someone had coded a rule directly into the algorithm: "Reject women over 55."

The EEOC promoted this case widely as the first federal sanction for AI hiring discrimination. Then-EEOC Chair Charlotte Burrows said in a statement: "Employers cannot use AI and algorithms as a shield for discrimination. If the technology makes discriminatory decisions, it is still illegal."

The iTutorGroup case was, in a sense, an easy one. It involved clear intentional discrimination. The harder question lies elsewhere. What happens when an algorithm, without any intent, ends up disadvantaging a particular group? In legal terms, this is called "Disparate Impact." Mobley v. Workday is precisely about this disparate impact theory.

(2) Current State and Standards of AI Hiring Tool Audits

In May 2023, the EEOC published technical guidance on AI hiring tools. The document advised employers on how to comply with Title VII of the Civil Rights Act when using AI tools.

The core message was this:

Regularly test whether your AI tools have a disproportionate impact on protected groups. If disproportionate impact is found, you must demonstrate that the tool is job-related and consistent with business necessity.

Then in January 2025, the situation changed abruptly.

President Donald Trump issued AI-related executive orders immediately after taking office. On January 20, he revoked the Biden administration's AI executive order. On January 23, he signed a new executive order titled "Removing Barriers to American AI Leadership." This order directed federal agencies to review existing AI policies and rescind any that "impede innovation."

On January 27, AI-related guidance documents disappeared from the EEOC website. The May 2023 Title VII guidance, the May 2022 Americans with Disabilities Act (ADA) guidance, and the December 2024 wearable device guidance were all gone. The same happened at the Department of Labor (DOL). Documents like "AI and Inclusive Hiring Framework" and "AI Best Practices" became inaccessible.

But guidance disappearing does not mean the law has disappeared. Title VII of the Civil Rights Act, the Age Discrimination in Employment Act (ADEA), and the Americans with Disabilities Act (ADA) remain in full force.

These laws were enacted by Congress, and the president cannot repeal them by executive order. The continued progress of Mobley v. Workday is proof of that.

In April 2025, President Trump went a step further. Through an executive order titled "Restoring Equal Opportunity and Meritocracy," he directed federal agencies to scale back enforcement based on "disparate impact" theory. This targeted the core legal doctrine at stake in the Mobley lawsuit. But it has no direct effect on private litigation. Mobley's attorneys filed their suit as private individuals, not as a federal agency.

As of 2026, the United States finds itself in a peculiar situation. Federal agencies have effectively halted enforcement against AI hiring discrimination. But private lawsuits continue. And state governments have begun filling the gap.

C. New York City's Experiment

(1) New York City Local Law 144: Mandatory Bias Audits for Automated Employment Decision Tools (AEDT)

On July 5, 2023, New York City became the first jurisdiction in the world to enforce regulations on AI hiring tools. Local Law 144 requires all employers using an "Automated Employment Decision Tool (AEDT)" to do two things.

First, conduct an independent bias audit every year.

Second, publish the audit results and notify applicants that an AI tool is being used.

What is a bias audit? Think of a school exam. A class has 100 male students and 100 female students. They take the test. Fifty of the male students pass. That is a 50% pass rate. Thirty-five of the female students pass. That is a 35% pass rate.

A question arises. Why is the female pass rate lower? Is it because the female students studied less? Or is there something wrong with the test itself?

U.S. employment law created a simple calculation. Divide the female pass rate by the male pass rate. 35 divided by 50. The answer is 0.7. That is 70%. If this number falls below 80%, a red flag goes up. It signals that the test may not be fair.

This is the "Four-Fifths Rule." Four out of five. 80%. If one group's pass rate does not reach 80% of another group's rate, the system comes under suspicion. It means the system may be operating to the disadvantage of a particular group.

A bias audit applies this calculation to AI hiring tools. Male and female. White and Black. Young and middle-aged. You compare each group's selection rate. You check whether it crosses the 80% threshold. If it doesn't, the AI may be discriminating.

In December 2025, the New York City Comptroller issued a report. It noted that two years after the law took effect, meaningful enforcement had barely occurred. Many companies were conducting bias audits only as a formality, or failing to properly disclose the results. Still, the New York City law established an important precedent. Other states and cities began using it as a reference point to create stronger regulations.

(2) Illinois SB 1398: AI Hiring Regulation

Illinois has played a pioneering role in regulating AI in hiring.

The Artificial Intelligence Video Interview Act, in effect since 2020, requires employers to notify applicants in advance and obtain consent when using AI to analyze video interviews.

In 2025, the state went further. HB 3773 takes effect on January 1, 2026. This law prohibits discrimination based on protected characteristics when AI is used in employment-related decisions such as hiring, promotion, and termination. It also requires employers to notify applicants and employees that AI is being used.

What distinguishes the Illinois law is that it regulates AI within the existing framework of anti-discrimination law. Rather than inventing new legal concepts, it made clear that existing anti-discrimination principles apply equally to AI.

(3) Colorado AI Act: Duty to Prevent Disparate Impact in Advance

Colorado attempted the most ambitious AI regulation. SB 24-205, known as the Colorado AI Act, was passed in May 2024. Modeled after the EU AI Act, it introduced a comprehensive regulatory framework for "high-risk AI systems."

The law's core aim is preventing algorithmic discrimination. Developers and deployers of AI systems used for consequential decisions in employment, housing, finance, healthcare, and education must exercise "reasonable care" to prevent discrimination. Specifically, they must conduct impact assessments, establish risk management policies, and inform consumers that AI is being used. But the law was mired in controversy before it even took effect. In May 2025, Colorado Governor Jared Polis sent a letter to the state legislature requesting a delay in implementation. He expressed concern that the law would create a "complex compliance regime" imposing excessive burdens on businesses.

In August 2025, the Colorado legislature convened a special session. Multiple amendment bills were introduced. One called for outright repeal, another drastically narrowed the scope, and yet another expanded small business exemptions. Negotiations proved difficult.

On August 26, Senate Majority Leader Robert Rodriguez submitted a straightforward delay bill instead of a compromise proposal. It pushed the original effective date of February 1, 2026 back by five months to June 30, 2026.

On August 28, Governor Polis signed the bill (SB 25B-004). All substantive requirements of the Colorado AI Act remained intact. Only the timeline shifted.

But a new layer of complexity emerged. On December 11, 2025, President Trump signed an executive order titled "Ensuring a National Framework for American AI Policy." The order criticized "excessive regulation at the state level" as detrimental to AI innovation and directed the Department of Justice to establish an "AI Litigation Task Force" to mount legal challenges against problematic state laws. Observers noted that the Colorado AI Act could be its first target.

This tension between federal and state governments will likely persist for some time. The federal government prioritizes innovation in the AI industry, while state governments prioritize protecting their citizens. Companies caught in between are left confused about which rules to follow.

D. Corporate Dispute Cases

(1) The Intuit/HireVue Case

In 2019, the American Civil Liberties Union (ACLU) and the Electronic Privacy Information Center (EPIC) filed a complaint with the Federal Trade Commission (FTC) requesting an investigation into HireVue.

HireVue is a video interview platform that uses AI to analyze applicants' facial expressions, voice tone, and word choice, then assigns a "hireability score."

The ACLU's argument went like this: HireVue's system relies on "emotion recognition" technology that lacks scientific backing. This technology may work against people of certain races or those with disabilities. For example, a person on the autism spectrum may not display "normal" eye contact or facial expressions. A Black person's facial expressions may be interpreted differently from a white person's.

HireVue responded to this criticism by announcing in 2021 that it had removed the facial analysis feature. But it still uses voice analysis and language analysis. According to a 2023 Boston Globe article, HireVue continues to use machine learning to score applicants' responses. The difference, it says, is that it now analyzes responses converted to text rather than analyzing video and audio.

Major companies including Intuit, Delta Air Lines, T-Mobile, and the Boston Red Sox have used HireVue. The FTC took no formal action on the ACLU's complaint, but the case sparked public debate about AI hiring tools.

(2) Baker v. CVS: The Algorithmic Filter Controversy

Brendan Baker was a resident of Milton, Massachusetts.

In January 2021, he applied for a supply chain management position at CVS Pharmacy. He was not hired. Later, he learned how his interview had been conducted. CVS used HireVue's video interview system. Applicants answer questions in front of a screen, and the video is recorded. The questions went like this: 'What does integrity mean to you?' 'What would you do if you saw someone cheating on a test?' 'Tell us about a time you acted with integrity.'

The recorded video was sent to a third-party platform called Affectiva.

This company, a spinoff from the MIT Media Lab, uses AI to analyze facial expressions, eye contact, voice tone, and intonation. Based on that analysis, HireVue assigns each applicant an 'employability score.' The score includes an assessment of 'integrity and dependability.'

According to HireVue's marketing materials, the system can detect whether an applicant possesses 'natural integrity and a sense of honor.' It claims to assist with 'lie detection' and to 'screen out exaggerators.'

Baker filed a lawsuit in 2023. His legal argument came from an unexpected angle.

Massachusetts has a Lie Detector Statute. The law prohibits employers from administering lie detector tests as a condition of employment. It also requires that every job application include a notice stating that 'it is unlawful in Massachusetts to require or administer a lie detector test as a condition of employment.'

Baker's argument was straightforward.

HireVue's system is, in effect, a lie detector. If it analyzes an applicant's facial expressions and voice to assess 'integrity,' that is an attempt to detect deception. CVS used this system without providing the notice required by law.

On February 16, 2024, Judge Patti B. Saris denied CVS's motion to dismiss. She found that Baker had suffered an 'informational injury' by not receiving the required notice. Had he been notified, he would have viewed the interview 'with a more critical eye.' On July 17, 2024, CVS and Baker reached a settlement. The terms were not disclosed.

The case left behind an important precedent. A lie detector ban written in the 1980s could apply to AI technology of the 2020s.

This is the power of 'technology-neutral' legal interpretation. When a statute prohibits not a specific technology (the polygraph) but a specific purpose (detecting deception), it can reach new technologies that serve the same purpose. As AI continues to advance, this kind of interpretation will only grow more significant.

The Baker case, the Mobley case, and the regulatory tug-of-war between federal and state governments all point to one large question. When AI makes hiring decisions, who bears the responsibility? The developer who built the algorithm? The employer who purchased it? Or does no one have to answer for it at all? The FTC enforcement actions we will examine in the next chapter offer yet another perspective on that question.