Workday v. Mobley: What legal impacts are ahead for HR leaders

As previously covered, the litigation and regulatory risks facing HR departments adopting artificial intelligence are continuing to emerge. The initial wave of AI-related risks facing HR departments tested legal theories focused on discrimination, and the second wave has focused on the use of AI to create “consumer reports” about job applicants in violation of the Fair Credit Reporting Act (FCRA). Now, courts have started reaching the merits of the first wave of AI-related discrimination cases, which will likely open the door for similar suits in the future.

This article discusses the recent court decision addressing AI-related discrimination claims and what employers can do to mitigate these risks.

Key takeaways from recent court cases

• Courts have held that AI-powered hiring tools, like Workday’s AI recommendation system, can support federal and state discrimination claims.
• As part of the lawsuit against Workday, Workday was required to turn over a list of all companies that use its AI-powered hiring tools. Employers using Workday’s AI-powered HR tools since 2020 should take steps now to minimize risks.
• Both the owner of AI-powered HR tools, as well as their business customers, can be held liable under California’s Fair Employment and Housing Act.

Latest updates from Mobley v. Workday

In February 2023, a class action was filed against Workday (Mobley v. Workday, Inc.), alleging that Workday’s AI-driven HR tools, such as HiredScore AI and Candidate Skills Match, discriminate on the basis of race, age and disability in violation of several federal and California laws. The case alleges that Workday’s AI recommendation system, Candidate Skills Match, which scored and ranked individuals, produced discriminatory outcomes at scale across hundreds of employer-customers.

The court previously conditionally certified an Age Discrimination in Employment Act (ADEA) class of persons aged 40 and older whose applications were processed by Workday’s AI recommendation system from September 2020 to the present. Earlier this year, the court also allowed a claim against Workday’s AI recommendation system to proceed under California’s Fair Employment and Housing Act (FEHA). In addition to letting the FEHA claim proceed, the court also held that the FEHA claim can extend to Workday’s AI-powered hiring tools used by non-California employers to evaluate non-California applicants.

FEHA’s application to California-based AI vendors

The court’s most significant holding is that FEHA reaches Workday’s conduct as a California-based agent engaging in allegedly discriminatory screening on behalf of employer-customers nationwide. The court rejected three principal arguments for dismissal:

• Direct vs. derivative liability. Workday argued its liability as an “agent” turns on its employer-customers’ liability. The court held this is “at odds with the FEHA’s concept of agency liability”. Workday is directly liable for its “own engagement in FEHA-regulated activities on the employer’s behalf” independent of employer-customer liability.
• Extraterritoriality. Workday argued FEHA cannot be applied based solely on California operations. The court disagreed and held that because Workday designs, develops and operates the tools from California, plaintiffs adequately alleged tortious discriminatory conduct occurring in California.
• Products-liability analogy. The court rejected Workday’s analogy to choice-of-law analysis in products-liability cases, reasoning that such cases do not illuminate FEHA’s text, structure or remedial intent to “prevent and deter unlawful employment practices.” The court declined to reach Workday’s constitutional arguments, leaving them for a fuller factual record.

Key legal implications

• Nationwide reach: A California-domiciled AI hiring-tool vendor may face FEHA liability for discriminatory outcomes nationwide, independent of whether any employer-customer would itself be subject to FEHA.
• FEHA “agent” liability: The court held that Workday, operating as an “agent” of its customers, could still be held liable and “share liability” with its customers.
• Product-suite sweep: Liability theories and collective definitions may encompass a vendor’s entire AI feature set, including acquired products running on third-party systems, under a “unified policy” theory.

How employers can mitigate risks

To mitigate risks from potential AI-related lawsuits, employers and HR departments should take certain mitigation measures, including:

• Develop clear company policies and guidelines regarding use of AI tools in the workplace, especially with HR-related functions such as recruitment, promotions, benefits, disciplinary actions, etc.
• Before contracting with an AI vendor, request more information from the vendor to better understand the legal implications of the vendor’s AI tools. Companies may want to create a list of standard questions posed to an AI vendor.
• When contracting with an AI vendor, HR and legal departments should consider the allocation of responsibilities and risks in key employment compliance areas, including:

1. 1. compliance with federal, state and local anti‑discrimination and employment laws;
   2. ADA and accessibility compliance in the screening process;
   3. data sources and accuracy of data;
   4. bias testing, impact assessments and audit rights (including access to information needed to evaluate adverse impact);
   5. data security and retention obligations;
   6. compliance with applicable laws regarding required notice and applicant consent; and
   7. allocation of liability and indemnification for claims arising from the AI tool or application.

• Before deploying an AI tool, ensure that all necessary disclosures are made to those affected by the AI tool (e.g., job applicants), and obtain any legally required consent. In general, the more transparent a company is, the more insulated the company is from legal risks.
• Once an AI tool has been deployed, maintain adequate human oversight of any AI tools and document such oversight. In the HR context, human oversight should look for indicia of bias, disparate impact, inaccuracy of information or lack of transparency.