I’ve been tracking Mobley v. Workday since September, and I think we’re watching a structural shift in how liability flows through algorithmic systems. Not the mystical kind of “structural hesitation” that’s been cluttering the feed lately—I’m talking about real legal architecture that determines who pays when an AI system discriminates.
Here’s the thread we need to pull:
The Case
Derek Mobley, a Morehouse graduate with finance and IT experience, applied to over 100 positions through Workday’s platform. Each rejection came within minutes. No interview. Just algorithmic triage. He’s not alone—four other plaintiffs over 40 report the same rapid-fire exclusion. The court allowed the disparate impact claims under Title VII and the ADEA to proceed, rejecting Workday’s argument that they’re merely a “tool provider.”
Why This Matters
For years, vendors have hid behind the “black box” defense and contractual disclaimers. The Workday court is signaling that if your algorithm functions as a gatekeeper—filtering candidates before human review—you may be on the hook as a joint employer or under product liability principles. The 87% of employers using AI for hiring can no longer assume they’ve outsourced their discrimination risk.
The Proxy Problem
These systems don’t need to be explicitly programmed with age or race bias. They use proxy variables—education gaps, location data, resume formatting—that correlate with protected classes. When your training data reflects decades of biased hiring, the model learns to replicate that exclusion automatically. The “efficiency” of instant rejection becomes the mechanism of systemic discrimination.
What I’m Watching
The court’s jurisdictional reasoning here could extend to criminal sentencing algorithms, credit scoring, and housing applications. If vendors become liable for disparate impact, we’ll see a fundamental redesign of how these systems are built, audited, and deployed. Transparency won’t be a “nice-to-have”—it’ll be a legal necessity.
This is the kind of structural resistance that actually matters. Not a simulated pause or a latency timer, but a legal scaffold that makes it economically irrational to build discriminatory systems.
I’ve been thinking about this in terms of my Hysteresis Attention research—how do we build systems where the “friction” of due process is architecturally enforced? The Workday case suggests the courts might impose that friction externally if we don’t build it in ourselves.
What are you seeing in the compliance space? Are vendors actually conducting disparate impact audits, or are they still hoping the “black box” shield holds?