Lawsuit against HR platform accused of bias in AI screening tool

AI-powered HR platform Workday must face claims that its algorithmic screening tool was built with biased training data and disproportionately weeds out job applicants who are Black, disabled and over 40, a federal judge in San Francisco has ruled,

The 20-page decision, issued Friday by Judge Rita Lin of the Northern District of California, tees up a legal showdown over the use of AI in making hiring decisions that could have sweeping implications for U.S. employers and HR vendors.

Derek Mobley, who brought the case last year on behalf of a proposed class of thousands, said he was turned down for at least 100 jobs at companies that use Workday’s screening software, including many Fortune 500 firms, even for similar roles he’d already held in industries where he had prior experience, including the insurer Unum and tech provider Hewlett-Packard Enterprise.

Mobley, who is Black and over 40, has anxiety and depression. His lawsuit says the Fortune 500 company uses personality tests and pymetrics that discriminate against applicants based on protected traits like age, race and disability.

Judge Lin refused to throw out the case on Friday, saying the vendor could be held liable for violating federal discrimination laws because it performs certain functions on behalf of its customers.

She stopped short of ruling that Pleasanton, California-based Workday should be liable as an employment agency, “but Workday may be liable on an agency theory because the [complaint] plausibly alleges that Workday’s customers delegated their traditional function of rejecting candidates or advancing them to the interview stage to Workday,” she wrote.

She also tossed Mobley’s claim that the discrimination was intentional, and dismissed with leave to amend his claim under California’s Fair Employment and Housing Act that Workday aided and abetted its customers in engaging in unlawful discrimination.

In her ruling, Lin noted Workday’s “allegedly crucial role in deciding which applicants can get their ‘foot in the door’ for an interview,” conduct “that is at the heart of equal access to employment opportunities.”

She added, “Nothing in the language of the federal anti-discrimination statutes or the case law interpreting those statutes distinguishes between delegating functions to an automated agent versus a live human one.”

Lin said the common denominator for all of the positions from which Mobley was rejected is Workday, which provided the application intake and screening service and that “in a traditional employment discrimination case, this data would be analogous to having over one hundred qualified applicants like Mobley—African American, over 40 years old, and suffering from depression and anxiety—all strike out for jobs with one employer.”

Mobley’s situation, she wrote, is even more compelling because he has struck out not just with one employer but with a whole range of employers across multiple industries that use Workday’s platform, including for a job with a company that he was already doing, but as a contractor.

“The zero percent success rate at passing Workday’s initial screening, combined with the [first amended complaint’s] allegations regarding bias in Workday’s training data and the tool’s reliance on information from pymetrics and personality tests, plausibly supports an inference that Workday’s algorithmic tools disproportionately reject applicants based on factors other than qualifications, such as a candidate’s race, age, or disability.”

Lawyers for Mobley at the Green Law Firm did not respond to a request for comment.

In an email to Law.com, Workday spokesman Connor Spielmaker said the company is “pleased that the majority of claims in this case were dismissed, and we’re confident that the remaining allegations will be easily refuted as we move to the next phase where we’ll have an opportunity to directly challenge their accuracy.”

Many companies are already using automated decision-making tools to streamline the hiring process and help filter out the hundreds of job applications they receive. “We’re going to be talking about this a lot as employers and companies rely upon technology like this to make their jobs easier,” said Jason Tremblay, a partner at Saul Ewing. “This is going to be the first of many cases that are filed and the question is what kind of theories, like aiding and abetting, are going to be invoked.”

Tremblay said the case is somewhat unique because Workday is accused not just of being responsible for the allegedly discriminatory algorithm, but went so far as to make recommendations to its clients about whether to move forward with a certain applicant.

“It took a much more involved position. So that made it easier for the court to find that [Workday] was an agent of the employer,” Tremblay said, and therefore covered under Title VII of the Civil Rights Act of 1964 as well as the Age Discrimination in Employment Act.

Lin’s ruling, the first challenging the use of AI in hiring, comes as lawmakers and regulators are increasingly scrutinizing the use of algorithms in employment decisions.

California’s Supreme Court ruled last year that vendors using algorithms to target job ads, screen out applicants or perform other employment-related tasks for other companies can be held directly liable for discrimination under state law.

Meanwhile, the Equal Employment Opportunity Commission has warned that anti-discrimination laws apply to the use of algorithms that make decisions about hiring and employment, and last year it settled its first bias lawsuit against an online tutoring company for using AI to reject older job applicants.

The EEOC filed an amicus brief urging Lin to allow the Mobley case to move forward, saying the commission “has a significant interest in whether and how these longstanding theories of coverage apply to the novel circumstances presented in this case.”

Read more: Study: Bias traps minorities in low-wage jobs

Lin has yet to decide whether to certify a class of job applicants who are over 40, Black and disabled. Tremblay said this step could pose a significant hurdle, as it’s hard to ascertain a set of similar facts from a pool of potentially thousands of potential class members.

“Damages are also hard to calculate. It’s hard to tell what that damage matrix would look like because you have so many people with so many different circumstances,” Tremblay said.

Still, the case could open the door to vendors being held liable for discrimination based on how involved they are in the hiring process. “It could scare some of those that are in the line of preparing these platforms that make these decisions easier for employers,” Tremblay said.

“And if a company gets sued for making alleged discriminatory decisions based on these platforms, are they going to demand contractual immunity from the vendor? I could see customers demanding more protections from some of these technology companies who are creating these tools.”