As employers increasingly incorporate artificial intelligence into hiring, workforce planning,
While the facts differ, both cases raise questions that should be familiar to every employer
Together, these cases suggest that employers should be evaluating both sides of the equation.
Question 1: Did AI make the decision, or did humans exercise meaningful judgment?
In the ongoing Mobley v. Workday litigation, the court allowed an ADA claim to proceed based on allegations that the applicant-screening system relied on factors that may have functioned as proxies for disability, including employment patterns potentially associated with medical leave, treatment or recovery.
Similarly, in the newly filed Doe v. Meta, plaintiffs allege that employees with disabilities and medical restrictions were disproportionately selected for termination through a workforce-reduction process that allegedly relied on flawed algorithmic profiling and performance scoring systems. Workday and Meta are denying the allegations.
In both cases, a central question is whether AI merely provided information to human decision-makers or whether it effectively acted as the decision-maker itself. However, merely inserting a human into the process may not be enough.
For employers, the practical question is not simply, "Was a human involved?" The better questions are: "What did the human actually do?" and "Can the organization demonstrate that independent judgment was exercised?"
Could the reviewer challenge the result? Did they understand the factors driving the recommendation? Were they expected to independently assess accommodation, leave, or disability-related considerations before acting? These questions may prove critical in future litigation.
Read more:
Question 2: Did the AI system properly account for disability, leave and accommodation rights?
While the human oversight question is important, employers should not overlook an equally significant issue: the quality of the underlying data and decision criteria.
Consider a performance management or workforce reduction model that relies heavily on:
- Attendance history
- Productivity metrics
- Time-in-position measurements
- Project completion rates
- Schedule adherence
- Absence frequency
On their face, these metrics may appear neutral. However, an employee who required intermittent FMLA leave, a reduced schedule as an ADA accommodation, or time away from work for medical treatment may naturally score differently against those measurements than employees who did not experience similar circumstances.
The same concerns can arise in recruiting algorithms that screen out applicants with employment gaps caused by disabilities, promotion systems that emphasize uninterrupted tenure, productivity tools that penalize employees using accommodations, or workforce analytics systems that flag employees for allegedly reduced performance after protected leave.
In other words, an AI system can produce discriminatory outcomes even if no discriminatory intent exists and even if a human ultimately approves the decision.
Read more:
The intersection of human oversight and data governance
A human reviewer cannot effectively prevent discriminatory outcomes if the underlying data and rankings already fail to account for disability and leave-related protections. Likewise, a well-designed algorithm can still create risk if managers blindly follow its recommendations.
Effective governance requires both:
- Meaningful and documented human oversight that allows decision-makers to understand, question, and override AI recommendations; and
- Careful review of underlying data inputs to ensure protected leave, disability-related absences, accommodations, and other legally protected circumstances are not improperly influencing outcomes.
Read more:
What benefits and HR leaders should be doing now
The lessons from Mobley and Doe v. Meta are not limited to hiring or layoffs. They apply to virtually any employment decision supported by analytics, algorithms or AI.
Leaders should consider:
- Mapping where AI or algorithmic tools influence employment decisions.
- Identifying whether disability-related absences, FMLA leave, ADA accommodations, etc. affect system inputs.
- Reviewing performance metrics used to train or evaluate AI systems.
- Establishing meaningful, documented human review procedures rather than simple approval processes.
- Conducting periodic audits for disparate impact and unintended bias.
- Holding vendors accountable for transparency regarding model design, inputs and outputs.
Read more:
Looking ahead
The emerging legal debate surrounding workplace AI is often framed as a question of whether humans or machines made a particular decision. Mobley and Doe v. Meta suggest the inquiry may be broader.
The first question is whether human decision-makers exercised documented, meaningful independent judgment or simply ratified algorithmic recommendations. The second is whether the AI systems themselves were built upon data and performance measures that appropriately accounted for disability, protected leave, accommodations, and other employee rights.
For employers, the challenge is not whether to use AI, but how to do so responsibly. Human oversight without sound data governance is insufficient, and sound data governance without meaningful human review is equally incomplete. As courts increasingly scrutinize both, employers must be prepared to demonstrate each.
